Search results

[bug] W4A16 fused MoE: runtime launch resolve has no CUDA-graph capture guard, so a first-use CuTe compile can run inside a captured graph

Status: Open.

#98 In local-inference-lab/sparkinfer;

· malaiwah opened on Jul 30, 2026

[correctness] PCIe DCP A2A: signed int slot_ overflows so staging_[-1] aliases signals_/self_signal_ — eager channel writes staged attention over a peer's barrier region and null-derefs

Status: Open.

#97 In local-inference-lab/sparkinfer;

· malaiwah opened on Jul 30, 2026

[perf] W4A16 launch path + indexer env parsing: full kernel rebuild before every compile-cache probe, prewarmed launches unreachable at runtime, per-call getenv on the paged-indexer prefill route

Status: Open.

#96 In local-inference-lab/sparkinfer;

· malaiwah opened on Jul 30, 2026

[correctness] DCP A2A capture channel adoption: graphs sharing one channel bake host-side staging-slot parity; stale capture stream-key mappings can hand a graph's channel to later captures

Status: Open.

#95 In local-inference-lab/sparkinfer;

· malaiwah opened on Jul 30, 2026

[memory] PCIe comm slabs + indexer scratch: per-capture channel slabs, per-call fold allocs, compressed-wire ring oversizing, fp32 tile_logits HWM

Status: Open.

#94 In local-inference-lab/sparkinfer;

· malaiwah opened on Jul 30, 2026

[memory] Trellis/W4A16 scratch sizing: unpinned block-size sweep reserves +65 MiB/rank worst case; plan scales unbounded with max_tokens

Status: Open.

#93 In local-inference-lab/sparkinfer;

· malaiwah opened on Jul 30, 2026

Size issue trying to run unsloth/Qwen3.6-27B-NVFP4 with b12x_attn (fathomless-firmament-ds4-v10 image)

Status: Open.

#29 In local-inference-lab/sparkinfer;

· jeremylea opened on Jul 11, 2026

Add a LICENSE file

Status: Open.

#19 In local-inference-lab/sparkinfer;

· ashish-iqm opened on Jun 26, 2026

benchmark_mla.py: default --tp-size gives num_local_heads=8, rejected by unified_sm120 prefill (HPB=16)

Status: Open.

#13 In local-inference-lab/sparkinfer;

· SeedSource opened on Jun 17, 2026

Request for SGLang code you may have for using b12x for DeekSeek v4 Flash inference

Status: Open.

#10 In local-inference-lab/sparkinfer;

· arcticjoe opened on Jun 8, 2026

nvidia/MiniMax-M2.7-NVFP4 problem "Allocated: 118784 bytes. Max: 101376 bytes."

Status: Open.

#5 In local-inference-lab/sparkinfer;

· paxren opened on May 17, 2026