diff --git a/models.json b/models.json index 319c901c..c81bec8e 100644 --- a/models.json +++ b/models.json @@ -1,2988 +1,23 @@ [ - { - "name": "pyt_huggingface_gpt2", - "url": "https://github.com/huggingface/transformers", - "dockerfile": "docker/pyt_huggingface", - "scripts": "scripts/huggingface_gpt2/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp16", - "tags": [ - "pyt", - "fp16", - "gpt2" - ], - "args": "" - }, - { - "name": "pyt_huggingface_bert", - "url": "https://github.com/huggingface/transformers", - "dockerfile": "docker/pyt_huggingface", - "scripts": "scripts/huggingface_bert/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "bert" - ], - "args": "" - }, - { - "name": "pyt_vllm_deepseek-r1", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-R1-0528.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo deepseek-ai/DeepSeek-R1-0528 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_deepseek-r1_fp4", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-R1-0528-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/DeepSeek-R1-0528-MXFP4 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_deepseek-v3.2", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-V3.2.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo deepseek-ai/DeepSeek-V3.2 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_deepseek-v3.2_fp4", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-V3.2-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/DeepSeek-V3.2-MXFP4 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_deepseek-v4-flash", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-V4-Flash.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo deepseek-ai/DeepSeek-V4-Flash --config configs/default.yaml" - }, - { - "name": "pyt_vllm_deepseek-v4-pro", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-V4-Pro.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo deepseek-ai/DeepSeek-V4-Pro --config configs/default.yaml" - }, - { - "name": "pyt_vllm_gpt-oss-20b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_gpt-oss-20b.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo openai/gpt-oss-20b --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_gpt-oss-120b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_gpt-oss-120b.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo openai/gpt-oss-120b --config configs/default.yaml" - }, - { - "name": "pyt_vllm_gpt-oss-120b_w4a8", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_gpt-oss120b-w-mxfp4-a-fp8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo amd/gpt-oss120b-w-mxfp4-a-fp8 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_glm-5.1_fp8", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_GLM-5.1-FP8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo zai-org/GLM-5.1-FP8 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_glm-5.1_fp4", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_GLM-5.1-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/GLM-5.1-MXFP4 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_glm-5.2_fp8", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_GLM-5.2-FP8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo zai-org/GLM-5.2-FP8 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_glm-5.2_fp4", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_GLM-5.2-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/GLM-5.2-MXFP4 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_kimi-k2.6", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Kimi-K2.6.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo moonshotai/Kimi-K2.6 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_kimi-k2.6_fp4", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Kimi-K2.6-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/Kimi-K2.6-MXFP4 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_kimi-k3", - "url": "", - "dockerfile": "docker/pyt_vllm_kimi_k3", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Kimi-K3.csv", - "tags": [ - "pyt", - "vllm", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo moonshotai/Kimi-K3 --config configs/default.yaml" - }, - { - "name": "pyt_atom_kimi-k3", - "url": "", - "dockerfile": "docker/pyt_atom_kimi_k3", - "scripts": "scripts/atom/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Kimi-K3.csv", - "tags": [ - "pyt", - "atom", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo moonshotai/Kimi-K3 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_llama-3.1-8b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-8B-Instruct.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo meta-llama/Llama-3.1-8B-Instruct --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_llama-3.1-8b_fp8", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-8B-Instruct-FP8-KV.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo amd/Llama-3.1-8B-Instruct-FP8-KV --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_llama-3.1-405b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-405B-Instruct.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo meta-llama/Llama-3.1-405B-Instruct --config configs/default.yaml" - }, - { - "name": "pyt_vllm_llama-3.1-405b_fp8", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-405B-Instruct-FP8-KV.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo amd/Llama-3.1-405B-Instruct-FP8-KV --config configs/default.yaml" - }, - { - "name": "pyt_vllm_llama-3.1-405b_fp4", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-405B-Instruct-MXFP4-Preview.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/Llama-3.1-405B-Instruct-MXFP4-Preview --config configs/default.yaml" - }, - { - "name": "pyt_vllm_llama-3.3-70b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.3-70B-Instruct.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo meta-llama/Llama-3.3-70B-Instruct --config configs/default.yaml" - }, - { - "name": "pyt_vllm_llama-3.3-70b_fp8", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.3-70B-Instruct-FP8-KV.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo amd/Llama-3.3-70B-Instruct-FP8-KV --config configs/default.yaml" - }, - { - "name": "pyt_vllm_llama-3.3-70b_fp4", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.3-70B-Instruct-MXFP4-Preview.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/Llama-3.3-70B-Instruct-MXFP4-Preview --config configs/default.yaml" - }, - { - "name": "pyt_vllm_minimax-m2.7", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_MiniMax-M2.7.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo MiniMaxAI/MiniMax-M2.7 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_minimax-m2.7_fp4", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_MiniMax-M2.7-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/MiniMax-M2.7-MXFP4 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_minimax-m3_mxfp8", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_MiniMax-M3-MXFP8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo MiniMaxAI/MiniMax-M3-MXFP8 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_minimax-m3_fp4", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_MiniMax-M3-MXFP4.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo amd/MiniMax-M3-MXFP4 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_mixtral-8x7b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Mixtral-8x7B-Instruct-v0.1.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo mistralai/Mixtral-8x7B-Instruct-v0.1 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_mixtral-8x7b_fp8", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Mixtral-8x7B-Instruct-v0.1-FP8-KV.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo amd/Mixtral-8x7B-Instruct-v0.1-FP8-KV --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_mixtral-8x22b", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Mixtral-8x22B-Instruct-v0.1.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo mistralai/Mixtral-8x22B-Instruct-v0.1 --config configs/default.yaml" - }, - { - "name": "pyt_vllm_mixtral-8x22b_fp8", - "url": "", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Mixtral-8x22B-Instruct-v0.1-FP8-KV.csv", - "tags": [ - "pyt", - "vllm", - "vllm_default", - "inference" - ], - "timeout": -1, - "args": - "--model_repo amd/Mixtral-8x22B-Instruct-v0.1-FP8-KV --config configs/default.yaml" - }, - { - "name": "pyt_vllm_qwen3-8b", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-8B.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3-8B --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3-32b", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-32B.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3-32B --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3-30b-a3b", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-30B-A3B-Thinking-2507.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3-30B-A3B-Thinking-2507 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3-30b-a3b_fp8", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-30B-A3B-Thinking-2507-FP8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3-235b-a22b", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-235B-A22B-Thinking-2507.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3-235B-A22B-Thinking-2507 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3-235b-a22b_fp8", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-235B-A22B-Thinking-2507-FP8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3-235B-A22B-Thinking-2507-FP8 --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3.5-397b-a17b", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3.5-397B-A17B.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3.5-397B-A17B --config configs/extended.yaml" - }, - { - "name": "pyt_vllm_qwen3.5-397b-a17b_fp8", - "data": "huggingface", - "dockerfile": "docker/pyt_vllm", - "scripts": "scripts/vllm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3.5-397B-A17B-FP8.csv", - "tags": [ - "pyt", - "vllm", - "vllm_extended", - "inference" - ], - "timeout": -1, - "args": - "--model_repo Qwen/Qwen3.5-397B-A17B-FP8 --config configs/extended.yaml" - }, - { - "name": "pyt_train_flux", - "url": "", - "dockerfile": "docker/pytorch_train", - "scripts": "scripts/pytorch_train/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Flux.csv", - "tags": [ - "pyt", - "training" - ], - "timeout": -1, - "args": - "--model_repo pyt_train_flux" - }, - { - "name": "pyt_train_stable-diffusion-xl", - "url": "", - "dockerfile": "docker/pytorch_train", - "scripts": "scripts/pytorch_train/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Stable-Diffusion-XL.csv", - "tags": [ - "pyt" - ], - "timeout": -1, - "args": - "--model_repo pyt_train_stable-diffusion-xl" - }, - { - "name": "pyt_train_dlrm", - "url": "", - "dockerfile": "docker/pytorch_train", - "scripts": "scripts/pytorch_train/run.sh", - "data": "", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DLRM.csv", - "tags": [ - "pyt" - ], - "timeout": -1, - "args": - "--model_repo pyt_train_dlrm" - }, - { - "name": "pyt_train_mochi-1", - "url": "", - "dockerfile": "docker/pytorch_train", - "scripts": "scripts/pytorch_train/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Mochi-1.csv", - "tags": [ - "pyt" - ], - "timeout": -1, - "args": "--model_repo pyt_train_mochi-1" - }, - { - "name": "pyt_train_hunyuan-video", - "url": "", - "dockerfile": "docker/pytorch_train", - "scripts": "scripts/pytorch_train/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Hunyuan-video.csv", - "tags": [ - "pyt" - ], - "timeout": -1, - "args": "--model_repo pyt_train_hunyuan-video" - }, - { - "name": "pyt_train_wan2_1-i2v", - "url": "", - "dockerfile": "docker/pytorch_train", - "scripts": "scripts/pytorch_train/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Wan2_1-i2v.csv", - "tags": [ - "pyt" - ], - "timeout": -1, - "args": "--model_repo pyt_train_wan2_1-i2v" - }, - { - "name": "pyt_mochi_video_inference", - "dockerfile": "docker/pyt_mochi_inference", - "scripts": "scripts/mochi/run_mochi.sh", - "n_gpus": "1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "inference" - ], - "args": "--model_repo genmo/mochi-1-preview --tunableop off" - }, - { - "name": "jax_maxtext_train_llama-2-7b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-2-7B.csv", - "tags": [ - "jax", - "llama2", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_llama-2-7b" - }, - { - "name": "jax_maxtext_train_llama-2-7b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_Llama-2-7B_nanoo_fp8.csv", - "tags": [ - "jax", - "llama2", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_llama-2-7b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_llama-2-7b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Llama-2-7B_fp8.csv", - "tags": [ - "jax", - "llama2", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_llama-2-7b --quantization fp8" - }, - { - "name": "jax_maxtext_train_llama-2-70b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-2-70B.csv", - "tags": [ - "jax", - "llama2", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_llama-2-70b" - }, - { - "name": "jax_maxtext_train_llama-2-70b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_Llama-2-70B_nanoo_fp8.csv", - "tags": [ - "jax", - "llama2", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_llama-2-70b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_llama-2-70b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Llama-2-70B_fp8.csv", - "tags": [ - "jax", - "llama2", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_llama-2-70b --quantization fp8" - }, - { - "name": "jax_maxtext_train_llama-3.1-8b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-8B.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_llama-3.1-8b" - }, - { - "name": "jax_maxtext_train_llama-3.1-8b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_Llama-3.1-8B_nanoo_fp8.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_llama-3.1-8b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_llama-3.1-8b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Llama-3.1-8B_fp8.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_llama-3.1-8b --quantization fp8" - }, - { - "name": "jax_maxtext_train_llama-3.1-70b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.1-70B.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_llama-3.1-70b" - }, - { - "name": "jax_maxtext_train_llama-3.1-70b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Llama-3.1-70B_fp8.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_llama-3.1-70b --quantization fp8" - }, - { - "name": "jax_maxtext_train_llama-3.3-70b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Llama-3.3-70B.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_llama-3.3-70b" - }, - { - "name": "jax_maxtext_train_llama-3.3-70b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Llama-3.3-70B_fp8.csv", - "tags": [ - "jax", - "llama3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_llama-3.3-70b --quantization fp8" - }, - { - "name": "jax_maxtext_train_deepseek-v2-lite-16b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-V2-lite.csv", - "tags": [ - "jax", - "deepseekv2", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_deepseek-v2-lite-16b" - }, - { - "name": "jax_maxtext_train_deepseek-v2-lite-16b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_DeepSeek-V2-lite_nanoo_fp8.csv", - "tags": [ - "jax", - "deepseekv2", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_deepseek-v2-lite-16b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_deepseek-v2-lite-16b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_DeepSeek-V2-lite_fp8.csv", - "tags": [ - "jax", - "deepseekv2", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_deepseek-v2-lite-16b --quantization fp8" - }, - { - "name": "jax_maxtext_train_mixtral-8x7b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Mixtral-8x7B.csv", - "tags": [ - "jax", - "mixtral", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_mixtral-8x7b" - }, - { - "name": "jax_maxtext_train_mixtral-8x7b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_Mixtral-8x7B_nanoo_fp8.csv", - "tags": [ - "jax", - "mixtral", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_mixtral-8x7b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_mixtral-8x7b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Mixtral-8x7B_fp8.csv", - "tags": [ - "jax", - "mixtral", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_mixtral-8x7b --quantization fp8" - }, - { - "name": "jax_maxtext_train_qwen3-14b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-14B.csv", - "tags": [ - "jax", - "qwen3", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_qwen3-14b" - }, - { - "name": "jax_maxtext_train_qwen3-14b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_Qwen3-14B_nanoo_fp8.csv", - "tags": [ - "jax", - "qwen3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_qwen3-14b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_qwen3-14b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Qwen3-14B_fp8.csv", - "tags": [ - "jax", - "qwen3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_qwen3-14b --quantization fp8" - }, - { - "name": "jax_maxtext_train_qwen3-30b-a3b", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Qwen3-30B-A3B.csv", - "tags": [ - "jax", - "qwen3", - "training" - ], - "timeout": -1, - "args": - "--model_repo jax_maxtext_train_qwen3-30b-a3b" - }, - { - "name": "jax_maxtext_train_qwen3-30b-a3b_nanoo_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "nanoo_fp8", - "multiple_results": "perf_Qwen3-30B-A3B_nanoo_fp8.csv", - "tags": [ - "jax", - "qwen3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx950", - "args": - "--model_repo jax_maxtext_train_qwen3-30b-a3b --quantization nanoo_fp8" - }, - { - "name": "jax_maxtext_train_qwen3-30b-a3b_fp8", - "url": "", - "dockerfile": "docker/jax_maxtext", - "scripts": "scripts/jax-maxtext/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "fp8", - "multiple_results": "perf_Qwen3-30B-A3B_fp8.csv", - "tags": [ - "jax", - "qwen3", - "training" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo jax_maxtext_train_qwen3-30b-a3b --quantization fp8" - }, - { - "name": "pyt_chai1_inference", - "dockerfile": "docker/pyt_chai1_inference", - "scripts": "scripts/pyt_chai1_inference/run.sh", - "n_gpus": "1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "inference", - "pyt" - ], - "timeout": 28800, - "args": "", - "multiple_results": "perf_pyt_chai1_inference.csv" - }, - { - "name": "pyt_clip_inference", - "dockerfile": "docker/pyt_clip_inference", - "scripts": "scripts/pyt_clip_inference/run.sh", - "n_gpus": "1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "inference" - ], - "args": "--tunableop off" - }, - { - "name": "pyt_janus_pro_inference", - "url": "", - "dockerfile": "docker/pyt_janus_pro_inference", - "scripts": "scripts/pyt_janus_pro_inference/run.sh", - "n_gpus": "1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_pyt_janus_pro_inference.csv", - "tags": [ - "pyt", - "inference", - "janus-pro", - "multimodal" - ], - "timeout": -1, - "args": "" - }, - { - "name": "pyt_wan2.1_inference", - "dockerfile": "docker/pyt_wan2.1_inference", - "scripts": "scripts/pyt_wan2.1_inference/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "inference" - ], - "args": "--num_gpu 1" - }, - { - "name": "dummy_multi", - "dockerfile": "docker/dummy", - "scripts": "scripts/dummy/run_multi.sh", - "n_gpus": "1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "dummies" - ], - "args": "", - "multiple_results": "perf_dummy.csv" - }, - { - "name": "pyt_sglang_deepseek-r1-distill-qwen-32b", - "url": "", - "dockerfile": "docker/pyt_sglang", - "scripts": "scripts/sglang/run.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_DeepSeek-R1-Distill-Qwen-32B.csv", - "tags": [ - "pyt", - "sglang", - "inference" - ], - "timeout": 28800, - "args": - "--model_repo deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --test_option latency --num_gpu 8 --datatype bfloat16 --dataset random --batch_size 1,8,32 --lat_input_output_len '128:128;128:1024;1024:128;1024:1024'" - }, - { - "name": "pyt_sglang_kimi-k3", - "url": "", - "dockerfile": "docker/pyt_sglang_kimi_k3", - "scripts": "scripts/sglang/run_kimi_k3.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Kimi-K3.csv", - "tags": [ - "pyt", - "sglang", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo moonshotai/Kimi-K3 --config configs/kimi_k3.yaml --variant nospec" - }, - { - "name": "pyt_sglang_kimi-k3_dspark", - "url": "", - "dockerfile": "docker/pyt_sglang_kimi_k3", - "scripts": "scripts/sglang/run_kimi_k3.sh", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_Kimi-K3.csv", - "tags": [ - "pyt", - "sglang", - "inference" - ], - "timeout": -1, - "skip_gpu_arch": "gfx942", - "args": - "--model_repo moonshotai/Kimi-K3 --config configs/kimi_k3.yaml --variant dspark" - }, - { - "name": "sglang-disagg-deepseek-r1-overlay", - "url": "", - "dockerfile": "docker/sglang_disagg_inference_full_overlay", - "scripts": "scripts/sglang_disagg/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_sglang-disagg-DeepSeek-R1.csv", - "tags": [ - "pyt", - "sglang", - "inference", - "disaggregated", - "deepseek-r1", - "moe", - "ep-internode", - "rccl-tp", - "mori-a2a" - ], - "timeout": 14400, - "args": "" - }, - { - "name": "pyt_hy_video", - "url": "", - "dockerfile": "docker/pyt_hy_video", - "scripts": "scripts/pyt_hy_video/run_720p.sh", - "n_gpus": "8", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_pyt_hy_video.csv", - "tags": [ - "pyt", - "inference", - "video" - ], - "timeout": -1, - "args": - "--model_repo microsoft/phi-4 --test_option latency --num_gpu 1 --datatype float16" - }, - { - "name": "pyt_huggingface_stable_diffusion_xl_2k_lora_finetuning", - "url": "https://github.com/huggingface/diffusers", - "dockerfile": "docker/pyt_training_huggingface", - "scripts": "scripts/pyt_huggingface_diffusers/run.sh", - "data": "huggingface", - "n_gpus": "8", - "owner": "mad.support@amd.com", - "training_precision": "fp16", - "multiple_results": "results_pyt_huggingface_stable_diffusion_xl_lora._finetuning.csv", - "timeout": -1, - "tags": [ - "training", - "perf", - "pyt", - "lora", - "diffusers", - "stable-diffusion-xl" - ], - "args": - "" - }, - { - "name": "pyt_ncf_training", - "url": "https://github.com/ROCm/DeepLearningExamples", - "data": "", - "dockerfile": "docker/pyt_ncf_training", - "scripts": "scripts/ncf/run.sh", - "n_gpus": "1", - "owner": "mad.support@amd.com", - "training_precision": "fp32", - "multiple_results": "results_ncf.csv", - "tags": [ - "pyt", - "training", - "recommendation", - "ncf" - ], - "timeout": -1, - "args": - "" - }, - { - "name": "pyt_xdit_flux", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2i", - "pyt", - "xdit" - ], - "args": "--workload flux", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_flux_kontext", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "ti2i", - "pyt", - "xdit" - ], - "args": "--workload flux_kontext", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_flux_2", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2i", - "ti2i", - "tmi2i", - "pyt", - "xdit" - ], - "args": "--workload flux2", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_flux_2_klein", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2i", - "pyt", - "xdit" - ], - "args": "--workload flux2_klein", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_hunyuanvideo", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2v", - "pyt", - "xdit" - ], - "args": "--workload hunyuanvideo", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_hunyuanvideo_1_5", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2v", - "pyt", - "xdit" - ], - "args": "--workload hunyuanvideo_1_5", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_ltx_2_3", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2av", - "pyt", - "xdit" - ], - "args": "--workload ltx2_3", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_qwen_image", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2i", - "pyt", - "xdit" - ], - "args": "--workload qwen_image", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_qwen_image_edit", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "ti2i", - "pyt", - "xdit" - ], - "args": "--workload qwen_image_edit", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_sd_3_5", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2i", - "pyt", - "xdit" - ], - "args": "--workload stablediffusion_3_5", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_wan_2_1", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "i2v", - "pyt", - "xdit" - ], - "args": "--workload wan2_1", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_wan_2_2", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "i2v", - "pyt", - "xdit" - ], - "args": "--workload wan2_2", - "multiple_results": "results.csv" - }, - { - "name": "pyt_xdit_z_image", - "dockerfile": "docker/pyt_xdit", - "scripts": "scripts/pyt_xdit/run.sh", - "n_gpus": "8", - "owner": "nsakkine@amd.com", - "training_precision": "", - "tags": [ - "inference", - "diffusion", - "t2i", - "pyt", - "xdit" - ], - "args": "--workload z_image", - "multiple_results": "results.csv" - }, - { - "name": "pyt_sglang_disagg_mori_io_llama-3.1-8b", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": 0, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Llama-3.1-8B-Instruct", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_io_qwen3-32b", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Qwen3-32B", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_io_llama-3.3-70b-fp8", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "amd-Llama-3.3-70B-Instruct-FP8-KV", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_io_llama-3.1-405b-fp8", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Llama-3.1-405B-Instruct-FP8-KV", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_io_mixtral-8x7b", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Mixtral-8x7B-Instruct-v0.1", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_io_deepseek-v3", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_io_deepseek-r1", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_io", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-R1", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_dp_deepseek-v3", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_dp", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3", - "xP": "1", - "yD": "1", - "DP_MODE": "1", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mori_dp_deepseek-r1", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mori_dp", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-R1", - "xP": "1", - "yD": "1", - "DP_MODE": "1", - "RUN_MORI": "1", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_llama-3.1-8b", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Llama-3.1-8B-Instruct", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_qwen3-32b", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Qwen3-32B", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_llama-3.3-70b-fp8", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "amd-Llama-3.3-70B-Instruct-FP8-KV", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_llama-3.1-405b-fp8", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Llama-3.1-405B-Instruct-FP8-KV", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_mixtral-8x7b", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Mixtral-8x7B-Instruct-v0.1", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_deepseek-v3", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_sglang_disagg_mooncake_deepseek-r1", - "url": "", - "dockerfile": "docker/sglang_disagg_inference", - "scripts": "scripts/sglang_disagg/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "sglang", - "sglang_disagg", - "mooncake", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-R1", - "xP": "1", - "yD": "1", - "DP_MODE": "0", - "RUN_MORI": "0", - "USE_CX7_NICS": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_nixl_deepseek-v3", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "nixl", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_nixl_deepseek-r1", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "nixl", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-R1", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_nixl_deepseek-v3-5layer", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "nixl", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3-5layer", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_nixl_llama-3.1-405b-fp8", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "nixl", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "Llama-3.1-405B-Instruct-FP8-KV", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_nixl_llama-3.3-70b-fp8", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "nixl", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "amd-Llama-3.3-70B-Instruct-FP8-KV", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_nixl_gpt-oss-120b", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "nixl", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "gpt-oss-120b", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_mori_deepseek-v3", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "mori_ep", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3", - "xP": "1", - "yD": "1", - "RUN_MORI": "1", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_mori_deepseek-r1", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "mori_ep", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-R1", - "xP": "1", - "yD": "1", - "RUN_MORI": "1", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_mori_deepseek-v3-5layer", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "mori_ep", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3-5layer", - "xP": "1", - "yD": "1", - "RUN_MORI": "1", - "RUN_DEEPEP": "0", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_deepep_deepseek-v3", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "deepep", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "1", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_deepep_deepseek-r1", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "deepep", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-R1", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "1", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_vllm_disagg_deepep_deepseek-v3-5layer", - "url": "", - "dockerfile": "docker/vllm_disagg_inference", - "scripts": "scripts/vllm_dissag/run_xPyD_models.slurm", - "data": "huggingface", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "vllm", - "vllm_disagg", - "deepep", - "inference" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE_NAME": "", - "MODEL_NAME": "DeepSeek-V3-5layer", - "xP": "1", - "yD": "1", - "RUN_MORI": "0", - "RUN_DEEPEP": "1", - "BENCHMARK_COMBINATIONS": "1024/1024" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_large_ep_bench_1n", - "url": "", - "dockerfile": "docker/large_ep_benchmark", - "scripts": "scripts/large-ep-benchmark/run_benchmark.sbatch", - "data": "", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "large_ep", - "deepep", - "mori_ep", - "benchmark" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", - "DOCKER_IMAGE_NAME": "", - "IBDEVICES": "mlx5_0", - "SKIP_DEEPEP": "0" - }, - "args": "-N 1 -n 1" - }, - { - "name": "pyt_large_ep_bench_1n_mori_only", - "url": "", - "dockerfile": "docker/large_ep_benchmark", - "scripts": "scripts/large-ep-benchmark/run_benchmark.sbatch", - "data": "", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "large_ep", - "large_ep_scaling", - "mori_ep", - "benchmark" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", - "DOCKER_IMAGE_NAME": "", - "IBDEVICES": "mlx5_0", - "SKIP_DEEPEP": "1" - }, - "args": "-N 1 -n 1" - }, - { - "name": "pyt_large_ep_bench_2n", - "url": "", - "dockerfile": "docker/large_ep_benchmark", - "scripts": "scripts/large-ep-benchmark/run_benchmark.sbatch", - "data": "", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "large_ep", - "deepep", - "mori_ep", - "benchmark" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", - "DOCKER_IMAGE_NAME": "", - "IBDEVICES": "mlx5_0", - "SKIP_DEEPEP": "0" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_large_ep_bench_2n_mori_only", - "url": "", - "dockerfile": "docker/large_ep_benchmark", - "scripts": "scripts/large-ep-benchmark/run_benchmark.sbatch", - "data": "", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "large_ep", - "large_ep_scaling", - "mori_ep", - "benchmark" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", - "DOCKER_IMAGE_NAME": "", - "IBDEVICES": "mlx5_0", - "SKIP_DEEPEP": "1" - }, - "args": "-N 2 -n 2" - }, - { - "name": "pyt_large_ep_bench_4n_mori_only", - "url": "", - "dockerfile": "docker/large_ep_benchmark", - "scripts": "scripts/large-ep-benchmark/run_benchmark.sbatch", - "data": "", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "pyt", - "large_ep", - "large_ep_scaling", - "mori_ep", - "benchmark" - ], - "timeout": -1, - "distributed": { - "launcher": "slurm_multi" - }, - "env_vars": { - "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", - "DOCKER_IMAGE_NAME": "", - "IBDEVICES": "mlx5_0", - "SKIP_DEEPEP": "1" - }, - "args": "-N 4 -n 4" - }, - { - "name": "primus_train", - "dockerfile": "docker/primus", - "dockercontext": ".", - "scripts": "scripts/primus_train", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "tags": [ - "training", - "primus", - "megatron", - "pretrain" - ], - "args": "", - "env": { - "HSA_NO_SCRATCH_RECLAIM": "1", - "PRIMUS_TURBO_ATTN_V3_ATOMIC_FP32": "1", - "NVTE_CK_IS_V3_ATOMIC_FP32": "1" + { + "name": "primus_train", + "dockerfile": "docker/primus", + "dockercontext": ".", + "scripts": "scripts/primus_train", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "training", + "primus", + "megatron", + "pretrain" + ], + "args": "", + "env": { + "HSA_NO_SCRATCH_RECLAIM": "1", + "PRIMUS_TURBO_ATTN_V3_ATOMIC_FP32": "1", + "NVTE_CK_IS_V3_ATOMIC_FP32": "1" + } } - }, - { - "name": "primus_pyt_megatron_lm_train_llama-3.1-8b_overlay", - "url": "", - "dockerfile": "docker/primus_megatron_train_rccl_overlay", - "scripts": "scripts/primus_megatron-lm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_primus-megatron-Llama-3.1-8B.csv", - "tags": [ - "pyt", - "pretrain", - "llama3", - "training", - "primus", - "scaleout" - ], - "timeout": -1, - "args": "--model_repo primus_pyt_megatron_lm_train_llama-3.1-8b" - }, - { - "name": "primus_pyt_megatron_lm_train_llama-3.1-70b_overlay", - "url": "", - "dockerfile": "docker/primus_megatron_train_rccl_overlay", - "scripts": "scripts/primus_megatron-lm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_primus-megatron-Llama-3.1-70B.csv", - "tags": [ - "pyt", - "pretrain", - "llama3", - "training", - "primus", - "scaleout" - ], - "timeout": -1, - "args": "--model_repo primus_pyt_megatron_lm_train_llama-3.1-70b" - }, - { - "name": "primus_pyt_megatron_lm_train_llama-3.1-405b_overlay", - "url": "", - "dockerfile": "docker/primus_megatron_train_rccl_overlay", - "scripts": "scripts/primus_megatron-lm/run.sh", - "n_gpus": "-1", - "owner": "mad.support@amd.com", - "training_precision": "", - "multiple_results": "perf_primus-megatron-Llama-3.1-405B.csv", - "tags": [ - "pyt", - "pretrain", - "llama3", - "training", - "primus", - "scaleout" - ], - "timeout": -1, - "args": "--model_repo primus_pyt_megatron_lm_train_llama-3.1-405b" - } ] diff --git a/scripts/atom/models.json b/scripts/atom/models.json new file mode 100644 index 00000000..0520c0c3 --- /dev/null +++ b/scripts/atom/models.json @@ -0,0 +1,21 @@ +[ + { + "name": "pyt_atom_kimi-k3", + "dockerfile": "../../docker/pyt_atom_kimi_k3", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Kimi-K3.csv", + "tags": [ + "pyt", + "atom", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo moonshotai/Kimi-K3 --config configs/default.yaml" + } +] diff --git a/scripts/dummy/models.json b/scripts/dummy/models.json new file mode 100644 index 00000000..780d3aa8 --- /dev/null +++ b/scripts/dummy/models.json @@ -0,0 +1,15 @@ +[ + { + "name": "dummy_multi", + "dockerfile": "../../docker/dummy", + "scripts": "run_multi.sh", + "n_gpus": "1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "dummies" + ], + "args": "", + "multiple_results": "perf_dummy.csv" + } +] diff --git a/scripts/huggingface_bert/models.json b/scripts/huggingface_bert/models.json new file mode 100644 index 00000000..65aa5f03 --- /dev/null +++ b/scripts/huggingface_bert/models.json @@ -0,0 +1,16 @@ +[ + { + "name": "pyt_huggingface_bert", + "dockerfile": "../../docker/pyt_huggingface", + "scripts": "run.sh", + "url": "https://github.com/huggingface/transformers", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "bert" + ], + "args": "" + } +] diff --git a/scripts/huggingface_gpt2/models.json b/scripts/huggingface_gpt2/models.json new file mode 100644 index 00000000..6f1da511 --- /dev/null +++ b/scripts/huggingface_gpt2/models.json @@ -0,0 +1,17 @@ +[ + { + "name": "pyt_huggingface_gpt2", + "dockerfile": "../../docker/pyt_huggingface", + "scripts": "run.sh", + "url": "https://github.com/huggingface/transformers", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp16", + "tags": [ + "pyt", + "fp16", + "gpt2" + ], + "args": "" + } +] diff --git a/scripts/jax-maxtext/models.json b/scripts/jax-maxtext/models.json new file mode 100644 index 00000000..15008566 --- /dev/null +++ b/scripts/jax-maxtext/models.json @@ -0,0 +1,468 @@ +[ + { + "name": "jax_maxtext_train_llama-2-7b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-2-7B.csv", + "tags": [ + "jax", + "llama2", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_llama-2-7b" + }, + { + "name": "jax_maxtext_train_llama-2-7b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_Llama-2-7B_nanoo_fp8.csv", + "tags": [ + "jax", + "llama2", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_llama-2-7b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_llama-2-7b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Llama-2-7B_fp8.csv", + "tags": [ + "jax", + "llama2", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_llama-2-7b --quantization fp8" + }, + { + "name": "jax_maxtext_train_llama-2-70b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-2-70B.csv", + "tags": [ + "jax", + "llama2", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_llama-2-70b" + }, + { + "name": "jax_maxtext_train_llama-2-70b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_Llama-2-70B_nanoo_fp8.csv", + "tags": [ + "jax", + "llama2", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_llama-2-70b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_llama-2-70b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Llama-2-70B_fp8.csv", + "tags": [ + "jax", + "llama2", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_llama-2-70b --quantization fp8" + }, + { + "name": "jax_maxtext_train_llama-3.1-8b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-8B.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_llama-3.1-8b" + }, + { + "name": "jax_maxtext_train_llama-3.1-8b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_Llama-3.1-8B_nanoo_fp8.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_llama-3.1-8b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_llama-3.1-8b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Llama-3.1-8B_fp8.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_llama-3.1-8b --quantization fp8" + }, + { + "name": "jax_maxtext_train_llama-3.1-70b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-70B.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_llama-3.1-70b" + }, + { + "name": "jax_maxtext_train_llama-3.1-70b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Llama-3.1-70B_fp8.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_llama-3.1-70b --quantization fp8" + }, + { + "name": "jax_maxtext_train_llama-3.3-70b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.3-70B.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_llama-3.3-70b" + }, + { + "name": "jax_maxtext_train_llama-3.3-70b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Llama-3.3-70B_fp8.csv", + "tags": [ + "jax", + "llama3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_llama-3.3-70b --quantization fp8" + }, + { + "name": "jax_maxtext_train_deepseek-v2-lite-16b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-V2-lite.csv", + "tags": [ + "jax", + "deepseekv2", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_deepseek-v2-lite-16b" + }, + { + "name": "jax_maxtext_train_deepseek-v2-lite-16b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_DeepSeek-V2-lite_nanoo_fp8.csv", + "tags": [ + "jax", + "deepseekv2", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_deepseek-v2-lite-16b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_deepseek-v2-lite-16b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_DeepSeek-V2-lite_fp8.csv", + "tags": [ + "jax", + "deepseekv2", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_deepseek-v2-lite-16b --quantization fp8" + }, + { + "name": "jax_maxtext_train_mixtral-8x7b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Mixtral-8x7B.csv", + "tags": [ + "jax", + "mixtral", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_mixtral-8x7b" + }, + { + "name": "jax_maxtext_train_mixtral-8x7b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_Mixtral-8x7B_nanoo_fp8.csv", + "tags": [ + "jax", + "mixtral", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_mixtral-8x7b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_mixtral-8x7b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Mixtral-8x7B_fp8.csv", + "tags": [ + "jax", + "mixtral", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_mixtral-8x7b --quantization fp8" + }, + { + "name": "jax_maxtext_train_qwen3-14b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-14B.csv", + "tags": [ + "jax", + "qwen3", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_qwen3-14b" + }, + { + "name": "jax_maxtext_train_qwen3-14b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_Qwen3-14B_nanoo_fp8.csv", + "tags": [ + "jax", + "qwen3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_qwen3-14b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_qwen3-14b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Qwen3-14B_fp8.csv", + "tags": [ + "jax", + "qwen3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_qwen3-14b --quantization fp8" + }, + { + "name": "jax_maxtext_train_qwen3-30b-a3b", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-30B-A3B.csv", + "tags": [ + "jax", + "qwen3", + "training" + ], + "timeout": -1, + "args": "--model_repo jax_maxtext_train_qwen3-30b-a3b" + }, + { + "name": "jax_maxtext_train_qwen3-30b-a3b_nanoo_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "nanoo_fp8", + "multiple_results": "perf_Qwen3-30B-A3B_nanoo_fp8.csv", + "tags": [ + "jax", + "qwen3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx950", + "args": "--model_repo jax_maxtext_train_qwen3-30b-a3b --quantization nanoo_fp8" + }, + { + "name": "jax_maxtext_train_qwen3-30b-a3b_fp8", + "dockerfile": "../../docker/jax_maxtext", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "fp8", + "multiple_results": "perf_Qwen3-30B-A3B_fp8.csv", + "tags": [ + "jax", + "qwen3", + "training" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo jax_maxtext_train_qwen3-30b-a3b --quantization fp8" + } +] diff --git a/scripts/large-ep-benchmark/models.json b/scripts/large-ep-benchmark/models.json new file mode 100644 index 00000000..c6b2fd80 --- /dev/null +++ b/scripts/large-ep-benchmark/models.json @@ -0,0 +1,142 @@ +[ + { + "name": "pyt_large_ep_bench_1n", + "dockerfile": "../../docker/large_ep_benchmark", + "scripts": "run_benchmark.sbatch", + "url": "", + "data": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "large_ep", + "deepep", + "mori_ep", + "benchmark" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", + "DOCKER_IMAGE_NAME": "", + "IBDEVICES": "mlx5_0", + "SKIP_DEEPEP": "0" + }, + "args": "-N 1 -n 1" + }, + { + "name": "pyt_large_ep_bench_1n_mori_only", + "dockerfile": "../../docker/large_ep_benchmark", + "scripts": "run_benchmark.sbatch", + "url": "", + "data": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "large_ep", + "large_ep_scaling", + "mori_ep", + "benchmark" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", + "DOCKER_IMAGE_NAME": "", + "IBDEVICES": "mlx5_0", + "SKIP_DEEPEP": "1" + }, + "args": "-N 1 -n 1" + }, + { + "name": "pyt_large_ep_bench_2n", + "dockerfile": "../../docker/large_ep_benchmark", + "scripts": "run_benchmark.sbatch", + "url": "", + "data": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "large_ep", + "deepep", + "mori_ep", + "benchmark" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", + "DOCKER_IMAGE_NAME": "", + "IBDEVICES": "mlx5_0", + "SKIP_DEEPEP": "0" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_large_ep_bench_2n_mori_only", + "dockerfile": "../../docker/large_ep_benchmark", + "scripts": "run_benchmark.sbatch", + "url": "", + "data": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "large_ep", + "large_ep_scaling", + "mori_ep", + "benchmark" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", + "DOCKER_IMAGE_NAME": "", + "IBDEVICES": "mlx5_0", + "SKIP_DEEPEP": "1" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_large_ep_bench_4n_mori_only", + "dockerfile": "../../docker/large_ep_benchmark", + "scripts": "run_benchmark.sbatch", + "url": "", + "data": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "large_ep", + "large_ep_scaling", + "mori_ep", + "benchmark" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE": "rocm/pytorch-private:large-ep-benchmark-rocm720-mori-42e89547-20260605", + "DOCKER_IMAGE_NAME": "", + "IBDEVICES": "mlx5_0", + "SKIP_DEEPEP": "1" + }, + "args": "-N 4 -n 4" + } +] diff --git a/scripts/mochi/models.json b/scripts/mochi/models.json new file mode 100644 index 00000000..2120c96a --- /dev/null +++ b/scripts/mochi/models.json @@ -0,0 +1,15 @@ +[ + { + "name": "pyt_mochi_video_inference", + "dockerfile": "../../docker/pyt_mochi_inference", + "scripts": "run_mochi.sh", + "n_gpus": "1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "inference" + ], + "args": "--model_repo genmo/mochi-1-preview --tunableop off" + } +] diff --git a/scripts/ncf/models.json b/scripts/ncf/models.json new file mode 100644 index 00000000..23d2abc4 --- /dev/null +++ b/scripts/ncf/models.json @@ -0,0 +1,21 @@ +[ + { + "name": "pyt_ncf_training", + "dockerfile": "../../docker/pyt_ncf_training", + "scripts": "run.sh", + "url": "https://github.com/ROCm/DeepLearningExamples", + "data": "", + "n_gpus": "1", + "owner": "mad.support@amd.com", + "training_precision": "fp32", + "multiple_results": "results_ncf.csv", + "tags": [ + "pyt", + "training", + "recommendation", + "ncf" + ], + "timeout": -1, + "args": "" + } +] diff --git a/scripts/primus_megatron-lm/models.json b/scripts/primus_megatron-lm/models.json new file mode 100644 index 00000000..6ff03bee --- /dev/null +++ b/scripts/primus_megatron-lm/models.json @@ -0,0 +1,62 @@ +[ + { + "name": "primus_pyt_megatron_lm_train_llama-3.1-8b_overlay", + "dockerfile": "../../docker/primus_megatron_train_rccl_overlay", + "scripts": "run.sh", + "url": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_primus-megatron-Llama-3.1-8B.csv", + "tags": [ + "pyt", + "pretrain", + "llama3", + "training", + "primus", + "scaleout" + ], + "timeout": -1, + "args": "--model_repo primus_pyt_megatron_lm_train_llama-3.1-8b" + }, + { + "name": "primus_pyt_megatron_lm_train_llama-3.1-70b_overlay", + "dockerfile": "../../docker/primus_megatron_train_rccl_overlay", + "scripts": "run.sh", + "url": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_primus-megatron-Llama-3.1-70B.csv", + "tags": [ + "pyt", + "pretrain", + "llama3", + "training", + "primus", + "scaleout" + ], + "timeout": -1, + "args": "--model_repo primus_pyt_megatron_lm_train_llama-3.1-70b" + }, + { + "name": "primus_pyt_megatron_lm_train_llama-3.1-405b_overlay", + "dockerfile": "../../docker/primus_megatron_train_rccl_overlay", + "scripts": "run.sh", + "url": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_primus-megatron-Llama-3.1-405B.csv", + "tags": [ + "pyt", + "pretrain", + "llama3", + "training", + "primus", + "scaleout" + ], + "timeout": -1, + "args": "--model_repo primus_pyt_megatron_lm_train_llama-3.1-405b" + } +] diff --git a/scripts/pyt_chai1_inference/models.json b/scripts/pyt_chai1_inference/models.json new file mode 100644 index 00000000..1a94fb38 --- /dev/null +++ b/scripts/pyt_chai1_inference/models.json @@ -0,0 +1,17 @@ +[ + { + "name": "pyt_chai1_inference", + "dockerfile": "../../docker/pyt_chai1_inference", + "scripts": "run.sh", + "n_gpus": "1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "inference", + "pyt" + ], + "timeout": 28800, + "args": "", + "multiple_results": "perf_pyt_chai1_inference.csv" + } +] diff --git a/scripts/pyt_clip_inference/models.json b/scripts/pyt_clip_inference/models.json new file mode 100644 index 00000000..fb5410cf --- /dev/null +++ b/scripts/pyt_clip_inference/models.json @@ -0,0 +1,15 @@ +[ + { + "name": "pyt_clip_inference", + "dockerfile": "../../docker/pyt_clip_inference", + "scripts": "run.sh", + "n_gpus": "1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "inference" + ], + "args": "--tunableop off" + } +] diff --git a/scripts/pyt_huggingface_diffusers/models.json b/scripts/pyt_huggingface_diffusers/models.json new file mode 100644 index 00000000..0df17109 --- /dev/null +++ b/scripts/pyt_huggingface_diffusers/models.json @@ -0,0 +1,23 @@ +[ + { + "name": "pyt_huggingface_stable_diffusion_xl_2k_lora_finetuning", + "dockerfile": "../../docker/pyt_training_huggingface", + "scripts": "run.sh", + "url": "https://github.com/huggingface/diffusers", + "data": "huggingface", + "n_gpus": "8", + "owner": "mad.support@amd.com", + "training_precision": "fp16", + "multiple_results": "results_pyt_huggingface_stable_diffusion_xl_lora._finetuning.csv", + "timeout": -1, + "tags": [ + "training", + "perf", + "pyt", + "lora", + "diffusers", + "stable-diffusion-xl" + ], + "args": "" + } +] diff --git a/scripts/pyt_hy_video/models.json b/scripts/pyt_hy_video/models.json new file mode 100644 index 00000000..4fd5f93b --- /dev/null +++ b/scripts/pyt_hy_video/models.json @@ -0,0 +1,19 @@ +[ + { + "name": "pyt_hy_video", + "dockerfile": "../../docker/pyt_hy_video", + "scripts": "run_720p.sh", + "url": "", + "n_gpus": "8", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_pyt_hy_video.csv", + "tags": [ + "pyt", + "inference", + "video" + ], + "timeout": -1, + "args": "--model_repo microsoft/phi-4 --test_option latency --num_gpu 1 --datatype float16" + } +] diff --git a/scripts/pyt_janus_pro_inference/models.json b/scripts/pyt_janus_pro_inference/models.json new file mode 100644 index 00000000..0b9a55d0 --- /dev/null +++ b/scripts/pyt_janus_pro_inference/models.json @@ -0,0 +1,20 @@ +[ + { + "name": "pyt_janus_pro_inference", + "dockerfile": "../../docker/pyt_janus_pro_inference", + "scripts": "run.sh", + "url": "", + "n_gpus": "1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_pyt_janus_pro_inference.csv", + "tags": [ + "pyt", + "inference", + "janus-pro", + "multimodal" + ], + "timeout": -1, + "args": "" + } +] diff --git a/scripts/pyt_wan2.1_inference/models.json b/scripts/pyt_wan2.1_inference/models.json new file mode 100644 index 00000000..1ac8357f --- /dev/null +++ b/scripts/pyt_wan2.1_inference/models.json @@ -0,0 +1,15 @@ +[ + { + "name": "pyt_wan2.1_inference", + "dockerfile": "../../docker/pyt_wan2.1_inference", + "scripts": "run.sh", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "inference" + ], + "args": "--num_gpu 1" + } +] diff --git a/scripts/pyt_xdit/models.json b/scripts/pyt_xdit/models.json new file mode 100644 index 00000000..7cfd3628 --- /dev/null +++ b/scripts/pyt_xdit/models.json @@ -0,0 +1,225 @@ +[ + { + "name": "pyt_xdit_flux", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2i", + "pyt", + "xdit" + ], + "args": "--workload flux", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_flux_kontext", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "ti2i", + "pyt", + "xdit" + ], + "args": "--workload flux_kontext", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_flux_2", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2i", + "ti2i", + "tmi2i", + "pyt", + "xdit" + ], + "args": "--workload flux2", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_flux_2_klein", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2i", + "pyt", + "xdit" + ], + "args": "--workload flux2_klein", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_hunyuanvideo", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2v", + "pyt", + "xdit" + ], + "args": "--workload hunyuanvideo", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_hunyuanvideo_1_5", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2v", + "pyt", + "xdit" + ], + "args": "--workload hunyuanvideo_1_5", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_ltx_2_3", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2av", + "pyt", + "xdit" + ], + "args": "--workload ltx2_3", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_qwen_image", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2i", + "pyt", + "xdit" + ], + "args": "--workload qwen_image", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_qwen_image_edit", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "ti2i", + "pyt", + "xdit" + ], + "args": "--workload qwen_image_edit", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_sd_3_5", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2i", + "pyt", + "xdit" + ], + "args": "--workload stablediffusion_3_5", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_wan_2_1", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "i2v", + "pyt", + "xdit" + ], + "args": "--workload wan2_1", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_wan_2_2", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "i2v", + "pyt", + "xdit" + ], + "args": "--workload wan2_2", + "multiple_results": "results.csv" + }, + { + "name": "pyt_xdit_z_image", + "dockerfile": "../../docker/pyt_xdit", + "scripts": "run.sh", + "n_gpus": "8", + "owner": "nsakkine@amd.com", + "training_precision": "", + "tags": [ + "inference", + "diffusion", + "t2i", + "pyt", + "xdit" + ], + "args": "--workload z_image", + "multiple_results": "results.csv" + } +] diff --git a/scripts/pytorch_train/models.json b/scripts/pytorch_train/models.json new file mode 100644 index 00000000..57280eb9 --- /dev/null +++ b/scripts/pytorch_train/models.json @@ -0,0 +1,99 @@ +[ + { + "name": "pyt_train_flux", + "dockerfile": "../../docker/pytorch_train", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Flux.csv", + "tags": [ + "pyt", + "training" + ], + "timeout": -1, + "args": "--model_repo pyt_train_flux" + }, + { + "name": "pyt_train_stable-diffusion-xl", + "dockerfile": "../../docker/pytorch_train", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Stable-Diffusion-XL.csv", + "tags": [ + "pyt" + ], + "timeout": -1, + "args": "--model_repo pyt_train_stable-diffusion-xl" + }, + { + "name": "pyt_train_dlrm", + "dockerfile": "../../docker/pytorch_train", + "scripts": "run.sh", + "url": "", + "data": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DLRM.csv", + "tags": [ + "pyt" + ], + "timeout": -1, + "args": "--model_repo pyt_train_dlrm" + }, + { + "name": "pyt_train_mochi-1", + "dockerfile": "../../docker/pytorch_train", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Mochi-1.csv", + "tags": [ + "pyt" + ], + "timeout": -1, + "args": "--model_repo pyt_train_mochi-1" + }, + { + "name": "pyt_train_hunyuan-video", + "dockerfile": "../../docker/pytorch_train", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Hunyuan-video.csv", + "tags": [ + "pyt" + ], + "timeout": -1, + "args": "--model_repo pyt_train_hunyuan-video" + }, + { + "name": "pyt_train_wan2_1-i2v", + "dockerfile": "../../docker/pytorch_train", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Wan2_1-i2v.csv", + "tags": [ + "pyt" + ], + "timeout": -1, + "args": "--model_repo pyt_train_wan2_1-i2v" + } +] diff --git a/scripts/sglang/models.json b/scripts/sglang/models.json new file mode 100644 index 00000000..bf67267c --- /dev/null +++ b/scripts/sglang/models.json @@ -0,0 +1,58 @@ +[ + { + "name": "pyt_sglang_deepseek-r1-distill-qwen-32b", + "dockerfile": "../../docker/pyt_sglang", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-R1-Distill-Qwen-32B.csv", + "tags": [ + "pyt", + "sglang", + "inference" + ], + "timeout": 28800, + "args": "--model_repo deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --test_option latency --num_gpu 8 --datatype bfloat16 --dataset random --batch_size 1,8,32 --lat_input_output_len '128:128;128:1024;1024:128;1024:1024'" + }, + { + "name": "pyt_sglang_kimi-k3", + "dockerfile": "../../docker/pyt_sglang_kimi_k3", + "scripts": "run_kimi_k3.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Kimi-K3.csv", + "tags": [ + "pyt", + "sglang", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo moonshotai/Kimi-K3 --config configs/kimi_k3.yaml --variant nospec" + }, + { + "name": "pyt_sglang_kimi-k3_dspark", + "dockerfile": "../../docker/pyt_sglang_kimi_k3", + "scripts": "run_kimi_k3.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Kimi-K3.csv", + "tags": [ + "pyt", + "sglang", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo moonshotai/Kimi-K3 --config configs/kimi_k3.yaml --variant dspark" + } +] diff --git a/scripts/sglang_disagg/models.json b/scripts/sglang_disagg/models.json new file mode 100644 index 00000000..a9205492 --- /dev/null +++ b/scripts/sglang_disagg/models.json @@ -0,0 +1,537 @@ +[ + { + "name": "pyt_sglang_disagg_mori_io_llama-3.1-8b", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": 0, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Llama-3.1-8B-Instruct", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_io_qwen3-32b", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Qwen3-32B", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_io_llama-3.3-70b-fp8", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "amd-Llama-3.3-70B-Instruct-FP8-KV", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_io_llama-3.1-405b-fp8", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Llama-3.1-405B-Instruct-FP8-KV", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_io_mixtral-8x7b", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Mixtral-8x7B-Instruct-v0.1", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_io_deepseek-v3", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_io_deepseek-r1", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_io", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-R1", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_dp_deepseek-v3", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_dp", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3", + "xP": "1", + "yD": "1", + "DP_MODE": "1", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mori_dp_deepseek-r1", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mori_dp", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-R1", + "xP": "1", + "yD": "1", + "DP_MODE": "1", + "RUN_MORI": "1", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_llama-3.1-8b", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Llama-3.1-8B-Instruct", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_qwen3-32b", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Qwen3-32B", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_llama-3.3-70b-fp8", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "amd-Llama-3.3-70B-Instruct-FP8-KV", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_llama-3.1-405b-fp8", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Llama-3.1-405B-Instruct-FP8-KV", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_mixtral-8x7b", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Mixtral-8x7B-Instruct-v0.1", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_deepseek-v3", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_sglang_disagg_mooncake_deepseek-r1", + "dockerfile": "../../docker/sglang_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "sglang", + "sglang_disagg", + "mooncake", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-R1", + "xP": "1", + "yD": "1", + "DP_MODE": "0", + "RUN_MORI": "0", + "USE_CX7_NICS": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "sglang-disagg-deepseek-r1-overlay", + "dockerfile": "../../docker/sglang_disagg_inference_full_overlay", + "scripts": "run.sh", + "url": "", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_sglang-disagg-DeepSeek-R1.csv", + "tags": [ + "pyt", + "sglang", + "inference", + "disaggregated", + "deepseek-r1", + "moe", + "ep-internode", + "rccl-tp", + "mori-a2a" + ], + "timeout": 14400, + "args": "" + } +] diff --git a/scripts/vllm/models.json b/scripts/vllm/models.json new file mode 100644 index 00000000..87b7c9c9 --- /dev/null +++ b/scripts/vllm/models.json @@ -0,0 +1,753 @@ +[ + { + "name": "pyt_vllm_deepseek-r1", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-R1-0528.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo deepseek-ai/DeepSeek-R1-0528 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_deepseek-r1_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-R1-0528-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/DeepSeek-R1-0528-MXFP4 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_deepseek-v3.2", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-V3.2.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo deepseek-ai/DeepSeek-V3.2 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_deepseek-v3.2_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-V3.2-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/DeepSeek-V3.2-MXFP4 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_deepseek-v4-flash", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-V4-Flash.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo deepseek-ai/DeepSeek-V4-Flash --config configs/default.yaml" + }, + { + "name": "pyt_vllm_deepseek-v4-pro", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_DeepSeek-V4-Pro.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo deepseek-ai/DeepSeek-V4-Pro --config configs/default.yaml" + }, + { + "name": "pyt_vllm_gpt-oss-20b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_gpt-oss-20b.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo openai/gpt-oss-20b --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_gpt-oss-120b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_gpt-oss-120b.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo openai/gpt-oss-120b --config configs/default.yaml" + }, + { + "name": "pyt_vllm_gpt-oss-120b_w4a8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_gpt-oss120b-w-mxfp4-a-fp8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo amd/gpt-oss120b-w-mxfp4-a-fp8 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_glm-5.1_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_GLM-5.1-FP8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo zai-org/GLM-5.1-FP8 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_glm-5.1_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_GLM-5.1-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/GLM-5.1-MXFP4 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_glm-5.2_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_GLM-5.2-FP8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo zai-org/GLM-5.2-FP8 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_glm-5.2_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_GLM-5.2-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/GLM-5.2-MXFP4 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_kimi-k2.6", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Kimi-K2.6.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo moonshotai/Kimi-K2.6 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_kimi-k2.6_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Kimi-K2.6-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/Kimi-K2.6-MXFP4 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_kimi-k3", + "dockerfile": "../../docker/pyt_vllm_kimi_k3", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Kimi-K3.csv", + "tags": [ + "pyt", + "vllm", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo moonshotai/Kimi-K3 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_llama-3.1-8b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-8B-Instruct.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo meta-llama/Llama-3.1-8B-Instruct --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_llama-3.1-8b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-8B-Instruct-FP8-KV.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo amd/Llama-3.1-8B-Instruct-FP8-KV --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_llama-3.1-405b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-405B-Instruct.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo meta-llama/Llama-3.1-405B-Instruct --config configs/default.yaml" + }, + { + "name": "pyt_vllm_llama-3.1-405b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-405B-Instruct-FP8-KV.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo amd/Llama-3.1-405B-Instruct-FP8-KV --config configs/default.yaml" + }, + { + "name": "pyt_vllm_llama-3.1-405b_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.1-405B-Instruct-MXFP4-Preview.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/Llama-3.1-405B-Instruct-MXFP4-Preview --config configs/default.yaml" + }, + { + "name": "pyt_vllm_llama-3.3-70b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.3-70B-Instruct.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo meta-llama/Llama-3.3-70B-Instruct --config configs/default.yaml" + }, + { + "name": "pyt_vllm_llama-3.3-70b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.3-70B-Instruct-FP8-KV.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo amd/Llama-3.3-70B-Instruct-FP8-KV --config configs/default.yaml" + }, + { + "name": "pyt_vllm_llama-3.3-70b_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Llama-3.3-70B-Instruct-MXFP4-Preview.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/Llama-3.3-70B-Instruct-MXFP4-Preview --config configs/default.yaml" + }, + { + "name": "pyt_vllm_minimax-m2.7", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_MiniMax-M2.7.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo MiniMaxAI/MiniMax-M2.7 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_minimax-m2.7_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_MiniMax-M2.7-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/MiniMax-M2.7-MXFP4 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_minimax-m3_mxfp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_MiniMax-M3-MXFP8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo MiniMaxAI/MiniMax-M3-MXFP8 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_minimax-m3_fp4", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_MiniMax-M3-MXFP4.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "skip_gpu_arch": "gfx942", + "args": "--model_repo amd/MiniMax-M3-MXFP4 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_mixtral-8x7b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Mixtral-8x7B-Instruct-v0.1.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo mistralai/Mixtral-8x7B-Instruct-v0.1 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_mixtral-8x7b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Mixtral-8x7B-Instruct-v0.1-FP8-KV.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo amd/Mixtral-8x7B-Instruct-v0.1-FP8-KV --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_mixtral-8x22b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Mixtral-8x22B-Instruct-v0.1.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo mistralai/Mixtral-8x22B-Instruct-v0.1 --config configs/default.yaml" + }, + { + "name": "pyt_vllm_mixtral-8x22b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Mixtral-8x22B-Instruct-v0.1-FP8-KV.csv", + "tags": [ + "pyt", + "vllm", + "vllm_default", + "inference" + ], + "timeout": -1, + "args": "--model_repo amd/Mixtral-8x22B-Instruct-v0.1-FP8-KV --config configs/default.yaml" + }, + { + "name": "pyt_vllm_qwen3-8b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-8B.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3-8B --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3-32b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-32B.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3-32B --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3-30b-a3b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-30B-A3B-Thinking-2507.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3-30B-A3B-Thinking-2507 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3-30b-a3b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-30B-A3B-Thinking-2507-FP8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3-235b-a22b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-235B-A22B-Thinking-2507.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3-235B-A22B-Thinking-2507 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3-235b-a22b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3-235B-A22B-Thinking-2507-FP8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3-235B-A22B-Thinking-2507-FP8 --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3.5-397b-a17b", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3.5-397B-A17B.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3.5-397B-A17B --config configs/extended.yaml" + }, + { + "name": "pyt_vllm_qwen3.5-397b-a17b_fp8", + "dockerfile": "../../docker/pyt_vllm", + "scripts": "run.sh", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "multiple_results": "perf_Qwen3.5-397B-A17B-FP8.csv", + "tags": [ + "pyt", + "vllm", + "vllm_extended", + "inference" + ], + "timeout": -1, + "args": "--model_repo Qwen/Qwen3.5-397B-A17B-FP8 --config configs/extended.yaml" + } +] diff --git a/scripts/vllm_dissag/models.json b/scripts/vllm_dissag/models.json new file mode 100644 index 00000000..6fc3849a --- /dev/null +++ b/scripts/vllm_dissag/models.json @@ -0,0 +1,374 @@ +[ + { + "name": "pyt_vllm_disagg_nixl_deepseek-v3", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "nixl", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_nixl_deepseek-r1", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "nixl", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-R1", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_nixl_deepseek-v3-5layer", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "nixl", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3-5layer", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_nixl_llama-3.1-405b-fp8", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "nixl", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "Llama-3.1-405B-Instruct-FP8-KV", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_nixl_llama-3.3-70b-fp8", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "nixl", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "amd-Llama-3.3-70B-Instruct-FP8-KV", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_nixl_gpt-oss-120b", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "nixl", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "gpt-oss-120b", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_mori_deepseek-v3", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "mori_ep", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3", + "xP": "1", + "yD": "1", + "RUN_MORI": "1", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_mori_deepseek-r1", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "mori_ep", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-R1", + "xP": "1", + "yD": "1", + "RUN_MORI": "1", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_mori_deepseek-v3-5layer", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "mori_ep", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3-5layer", + "xP": "1", + "yD": "1", + "RUN_MORI": "1", + "RUN_DEEPEP": "0", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_deepep_deepseek-v3", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "deepep", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "1", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_deepep_deepseek-r1", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "deepep", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-R1", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "1", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + }, + { + "name": "pyt_vllm_disagg_deepep_deepseek-v3-5layer", + "dockerfile": "../../docker/vllm_disagg_inference", + "scripts": "run_xPyD_models.slurm", + "url": "", + "data": "huggingface", + "n_gpus": "-1", + "owner": "mad.support@amd.com", + "training_precision": "", + "tags": [ + "pyt", + "vllm", + "vllm_disagg", + "deepep", + "inference" + ], + "timeout": -1, + "distributed": { + "launcher": "slurm_multi" + }, + "env_vars": { + "DOCKER_IMAGE_NAME": "", + "MODEL_NAME": "DeepSeek-V3-5layer", + "xP": "1", + "yD": "1", + "RUN_MORI": "0", + "RUN_DEEPEP": "1", + "BENCHMARK_COMBINATIONS": "1024/1024" + }, + "args": "-N 2 -n 2" + } +] diff --git a/tools/migrate_to_dir_models.py b/tools/migrate_to_dir_models.py new file mode 100755 index 00000000..26a59e97 --- /dev/null +++ b/tools/migrate_to_dir_models.py @@ -0,0 +1,194 @@ +#!/usr/bin/env python3 +""" +Migrate MAD root models.json to per-directory models.json files. + +For each model in the root models.json, this script: + 1. Derives the script directory from the model's "scripts" field. + 2. Translates absolute-from-root paths to paths relative to that script directory. + 3. Writes a models.json inside scripts// containing all models for that dir. + 4. Rewrites the root models.json to remove migrated entries (retaining only those + whose script dir already has a get_models_json.py, which cannot coexist with + a models.json per the madengine discovery rules). + +Path translation rules (all root paths start with their top-level prefix): + dockerfile: "docker/X" -> "../../docker/X" + scripts: "scripts/" -> "." + "scripts//" -> "." + "scripts//f.sh" -> "f.sh" + "scripts//…" -> "../../scripts//…" + dockercontext: kept as-is (project-root-relative) — the engine does NOT normalize it + +Usage (run from MAD repository root): + python3 tools/migrate_to_dir_models.py [--dry-run] + +Options: + --dry-run Print what would be written without writing any files. +""" + +import argparse +import json +import os +import sys + + +def extract_script_dir(scripts_field: str) -> str: + """Return the first-level subdirectory under scripts/ for the given scripts path.""" + # Normalise trailing slashes and split + parts = scripts_field.rstrip("/").split("/") + # Expected: ["scripts", "", ...] + if len(parts) < 2 or parts[0] != "scripts": + raise ValueError(f"Unexpected scripts path format: {scripts_field!r}") + return parts[1] + + +def translate_path(root_path: str, script_dir: str) -> str: + """Translate a project-root-relative path to be relative to scripts//. + + The engine resolves dir-specific paths via: + os.path.normpath(os.path.join("scripts", script_dir, translated_path)) + So we need: + translated_path such that normpath("scripts//" + translated) == root_path + + For a root path "scripts//..." -> the relative part after stripping that prefix. + For "scripts//..." -> "../../scripts//..." + For "docker/..." -> "../../docker/..." + """ + root_path_stripped = root_path.rstrip("/") + + # Path sits inside this script dir + prefix_bare = f"scripts/{script_dir}" + prefix_slash = f"scripts/{script_dir}/" + + if root_path_stripped == prefix_bare or root_path == prefix_slash: + # Points to the directory itself — use "." which normalises to "scripts/" + return "." + + if root_path.startswith(prefix_slash): + # Points to a file/subpath inside this dir — strip the prefix + return root_path[len(prefix_slash):] + + # Path is in a different top-level directory (docker/ or scripts//) + # Go up two levels from scripts// to reach the project root + return "../../" + root_path_stripped + + +def build_dir_entry(model: dict, script_dir: str) -> dict: + """Return a new model dict with paths translated for a dir-specific models.json.""" + entry = {} + + # name: keep as-is; the engine prepends "/" automatically + entry["name"] = model["name"] + + # Required fields with path translation + entry["dockerfile"] = translate_path(model["dockerfile"], script_dir) + entry["scripts"] = translate_path(model["scripts"], script_dir) + + # dockercontext is used directly by DockerBuilder (not normalized by the engine), + # so keep the original project-root-relative path unchanged. + if "dockercontext" in model: + entry["dockercontext"] = model["dockercontext"] + + # Copy all non-path fields in their original order + skip_keys = {"name", "dockerfile", "scripts", "dockercontext"} + for key, value in model.items(): + if key not in skip_keys: + entry[key] = value + + return entry + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) + parser.add_argument("--dry-run", action="store_true", help="Print actions without writing files") + args = parser.parse_args() + + # Always run from MAD repository root + root_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) + models_json_path = os.path.join(root_dir, "models.json") + scripts_dir = os.path.join(root_dir, "scripts") + + print(f"Reading {models_json_path} ...") + with open(models_json_path) as f: + all_models = json.load(f) + + print(f"Total models: {len(all_models)}") + + # Find dirs that already have get_models_json.py (cannot add models.json alongside) + dirs_with_dynamic = set() + for dirname in os.listdir(scripts_dir): + dir_path = os.path.join(scripts_dir, dirname) + if os.path.isdir(dir_path) and os.path.exists(os.path.join(dir_path, "get_models_json.py")): + dirs_with_dynamic.add(dirname) + + if dirs_with_dynamic: + print(f"Dirs with get_models_json.py (will be skipped): {sorted(dirs_with_dynamic)}") + + # Group models by their script directory + by_dir: dict[str, list[dict]] = {} + skipped_models: list[dict] = [] + + for model in all_models: + try: + script_dir = extract_script_dir(model["scripts"]) + except ValueError as e: + print(f"WARNING: {e} — leaving model {model['name']!r} in root models.json", file=sys.stderr) + skipped_models.append(model) + continue + + if script_dir in dirs_with_dynamic: + print(f" Skipping {model['name']!r}: dir {script_dir!r} has get_models_json.py") + skipped_models.append(model) + continue + + by_dir.setdefault(script_dir, []).append(model) + + print(f"\nDirectories to create models.json in: {len(by_dir)}") + print(f"Models remaining in root models.json: {len(skipped_models)}") + + # Write per-directory models.json files + total_written = 0 + for script_dir, models in sorted(by_dir.items()): + dir_path = os.path.join(scripts_dir, script_dir) + out_path = os.path.join(dir_path, "models.json") + + entries = [build_dir_entry(m, script_dir) for m in models] + + if args.dry_run: + print(f"\n[DRY RUN] Would write {out_path} ({len(entries)} models):") + print(json.dumps(entries[:2], indent=4)) + if len(entries) > 2: + print(f" ... ({len(entries) - 2} more entries)") + else: + if not os.path.isdir(dir_path): + print(f"WARNING: scripts dir does not exist: {dir_path}", file=sys.stderr) + skipped_models.extend(models) + continue + + if os.path.exists(out_path): + print(f"WARNING: {out_path} already exists — overwriting", file=sys.stderr) + + with open(out_path, "w") as f: + json.dump(entries, f, indent=4) + f.write("\n") + + print(f" Wrote {out_path} ({len(entries)} models)") + + total_written += len(entries) + + # Rewrite root models.json with only skipped models + new_root_path = models_json_path + if args.dry_run: + print(f"\n[DRY RUN] Would rewrite {new_root_path} with {len(skipped_models)} remaining models") + else: + with open(new_root_path, "w") as f: + json.dump(skipped_models, f, indent=4) + f.write("\n") + print(f"\nRewritten {new_root_path} with {len(skipped_models)} remaining models") + + print(f"\nDone. Models written to per-dir files: {total_written}") + if args.dry_run: + print("(dry run — no files were modified)") + + +if __name__ == "__main__": + main()