All issues

Search results

ENG-HYBRID-PLACEMENT: map llama.cpp's four placement surfaces onto --offload-config's vllm_cpp key, and re-anchor the spec at b10451

Status: Open.

#2015 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

index row for #1954 wrongly claims #907 documents the same test and assertion

Status: Open.

#2013 In mudler/vllm.cpp;

· joral opened on Aug 26, 2026

docs/FEATURES.md carries no row for hybrid CPU/GPU expert placement, so the -cmoe gap is invisible

Status: Open.

#2012 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

GLM-5.3-Flash has no GGUF and no tool can make one: author the glm5_next converter and gate its k-quant encoders byte-for-byte against llama.cpp b10451 (W7a of #1998)

Status: Open.

#2011 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

DFlash2's draft-context position invariant is ungated: deleting it leaves the whole suite green

Status: Open.

#2009 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

DFlash2 speculative decoding serves only ONE sequence: at c=2 the draft context accumulation desynchronises and EngineCore stops

Status: Open.

#2008 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

Attention and recurrent state are allocated from two separate pools, so c=32 at k=8 is unservable where vLLM and SGLang serve it

Status: Open.

#2007 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

No Tenstorrent equivalent of gpu_clock_state.py: every TT figure on record is clock-unattributed, including #2003

Status: Open.

#2005 In mudler/vllm.cpp;

· lu-zero opened on Aug 26, 2026

TT P150 at post-W2c main: the host-hybrid opt-out outperforms the host-free eager DEFAULT by 1.24x — #1604 flip premise inverted

Status: Open.

#2003 In mudler/vllm.cpp;

· lu-zero opened on Aug 26, 2026

Speculative decoding routes every step to the greedy-only RejectionSampler with no temperature guard, so a temperature-1.0 request decodes greedily and vt::RandomSample is never called

Status: Open.

#2002 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

GLM-5.3-Flash (Glm5NextForConditionalGeneration): a hybrid KDA + DSA + mHC 321B MM model no oracle implements and no device here can hold

enhancementNew feature or requestNew feature or request

Status: Open.

#1998 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026

MODEL-MM-QWEN4-EXP W8: e2e through examples/server, the llama.cpp concurrency ladder, and the four document surfaces

Status: Open.

#1996 In mudler/vllm.cpp;

· localai-bot opened on Aug 26, 2026