All issues

Search results

[Bug][XPU] 0.26.0-b1: Gemma 4 26B-A4B AWQ-INT4 graph capture dies — compiled AOT code calls arg.item() (host sync) inside command graph: "wait method cannot be used for an event associated with a command graph"

Status: Open.

#725 In intel/llm-scaler;

· jinsoojinsoo opened on Sep 22, 2026

[Bug][XPU] MTP spec decode (gemma4_mtp) with AWQ-INT4 target fails: '_C' object has no attribute 'gptq_marlin_repack' — Marlin repack not built for XPU

Status: Open.

#724 In intel/llm-scaler;

· jinsoojinsoo opened on Sep 22, 2026

[Bug][XPU] 0.26.0-b2: TP=2 fails on dual Arc Pro B70 — DEVICE_LOST / collective hang in eager, piecewise, and full-graph modes (b1 works)

Status: Open.

#723 In intel/llm-scaler;

· jinsoojinsoo opened on Sep 22, 2026

runaway degeneration on B70 sym_int4 TP8 with Qwen3‑Coder‑Next (scaler‑vllm:0.26.0‑b2), V1 engine: normal chat echo system prompt, tool call generates massive empty tool_calls

Status: Open.

#722 In intel/llm-scaler;

· zhouwenjing001 opened on Sep 22, 2026

[Bug][XPU] --kv-cache-dtype fp8 emits every second token as a duplicate on Arc Pro B65 (Battlemage)

Status: Open.

#721 In intel/llm-scaler;

· TheNorthWestPassage opened on Sep 21, 2026

[Issue] Native CPU KV offload crashes with AssertionError in _build_store_jobs under high concurrency (XPU, 0.26.0-b2)

Status: Open.

#720 In intel/llm-scaler;

· russalchang opened on Sep 18, 2026

scaler-vllm:0.26.0-b2 GPT-OSS -120B sym_int4/fp8, Gemma4-26B fp8 runaway-nIt\n degeneration on B70

Status: Open.

#716 In intel/llm-scaler;

· zhouwenjing001 opened on Sep 17, 2026

Omni: ComfyUI 运行SenseNova模型会OOM

Status: Open.

#710 In intel/llm-scaler;

· zhm-algo opened on Sep 15, 2026

intel/llm-scaler-vllm:0.26.0-b2:: Gemma 12B model loading error

Status: Open.

#705 In intel/llm-scaler;

· polarities opened on Sep 13, 2026

Out-of-memory when serving Qwen3.8-27B with image intel/llm-scaler-vllm:0.26.0-b2

Status: Open.

#703 In intel/llm-scaler;

· noobHappylife opened on Sep 11, 2026

0.26.0-b2: XPU Graph / torch.compile cannot start with sym_int4 — logger.info_once() inside the traced apply() path hard-fails Dynamo

Status: Open.

#699 In intel/llm-scaler;

· jsetsuda opened on Sep 10, 2026

gpt-oss-120b (MXFP4, TP=4, 4x Arc Pro B70): XPU Graph mode produces garbage ("!!!!" / empty tokens) on both 0.21.0-b3.1 and 0.26.0-b2; eager mode is correct

Status: Open.

#698 In intel/llm-scaler;

· dps041983-svg opened on Sep 10, 2026