Conversation
…decoding Add a new Spark recipe for serving Qwen3.6-27B-FP8 across two DGX Spark nodes using Ray distributed executor and RoCE interconnect. Key features: - Dual-node tensor parallelism (TP=2) via Ray distributed executor backend - Qwen MTP (Multi-Token Prediction) speculative decoding with 3 speculative tokens for improved throughput - FlashAttention attention backend for optimal performance - Chunked prefill and prefix caching for efficient request handling - RoCE network configuration (NCCL) with explicit HCA and socket interface settings for low-latency GPU-to-GPU communication - Custom chat template via fix-qwen3.6-enhanced-chat-template mod - Tool call parsing (qwen3_coder) and reasoning parser (qwen3) support - Conservative generation defaults (temp=0.2, top_p=0.90, top_k=20) - 256K max context length with 16K max batched tokens and 8 max sequences Target hardware: two DGX Spark nodes connected via RoCE (200GbE). Model: Qwen/Qwen3.6-27B-FP8 from HuggingFace.
|
hi @mmike87 is this PR missing fix-qwen3.6-enhanced-chat-template? |
Author
Yes, sorry - I'll fix it. |
Author
This should be fixed in the PR. Thanks! |
|
@mmike87 nice! thanks a lot |
Author
|
What's the approval process? |
|
i'm not so sure too. i think you may have to tag @eugr for a reviewer? |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
…decoding
Add a new Spark recipe for serving Qwen3.6-27B-FP8 across two DGX Spark nodes using Ray distributed executor and RoCE interconnect.
Key features:
Target hardware: two DGX Spark nodes connected via RoCE (200GbE).
Model: Qwen/Qwen3.6-27B-FP8 from HuggingFace.