Search results

AWQ GQA support

awqFor any issue / PR related to AWQ supportFor any issue / PR related to AWQ support

enhancementNew feature or requestNew feature or request

Status: Open.

Feature

#2808 In vllm-project/llm-compressor;

· HDCharles opened on Jun 8, 2026

[Bug]: AWQ w/ DDP and CPU offloading hits CUDA OOM

awqFor any issue / PR related to AWQ supportFor any issue / PR related to AWQ support

bugSomething isn't workingSomething isn't working

Status: Open.

#2790 In vllm-project/llm-compressor;

· brian-dellabetta opened on Jun 2, 2026

W8A8 INT8 support for Qwen3.6-35B-A3B (Qwen3.5 MoE architecture)

enhancementNew feature or requestNew feature or request

Status: Open.

#2787 In vllm-project/llm-compressor;

· Priyjain-amd opened on Jun 2, 2026

[Bug]: Python 3.14 incompatibility via pydantic eval_type_backport on Recipe.dict()

Status: Open.

#2778 In vllm-project/llm-compressor;

· u7k4rs6 opened on Jun 1, 2026

[Bug] sequential_epoch_end qparam writeback crashes with Inplace update to inference tensor outside InferenceMode on speculative-decode heads (DeepSeek-V4 MTP)

Status: Open.

#2745 In vllm-project/llm-compressor;

· pasta-paul opened on May 21, 2026

[Bug] Multi-rank calibration deadlocks in offload-cache code paths; propagate_error=False (only on deepseekv4-experimental) is the existing workaround

Status: Open.

#2743 In vllm-project/llm-compressor;

· pasta-paul opened on May 20, 2026

[Feature] Native intra-calibration resume in SequentialPipeline

Status: Open.

#2741 In vllm-project/llm-compressor;

· pasta-paul opened on May 20, 2026

GPTQModifier compress_module_list line 304: int(num_samples) on CUDA scalar causes synchronous device->host sync (stream stall under multi-rank decoupled MoE / sharded module sets)

Status: Open.

#2736 In vllm-project/llm-compressor;

· pasta-paul opened on May 20, 2026

DSv4 canonical example drops MTP layer; load_quantizable_moe regex anchored at ^layers excludes mtp.* block

Status: Open.

#2735 In vllm-project/llm-compressor;

· pasta-paul opened on May 20, 2026

GPTQModifier hangs on multi-rank with sharded MoE experts (dist.reduce/broadcast on disjoint module sets)

Status: Open.

#2734 In vllm-project/llm-compressor;

· pasta-paul opened on May 20, 2026

[Bug]: Mix-precision vLLM loading failure (MXFP4+MXFP8)

bugSomething isn't workingSomething isn't working

Status: Open.

#2730 In vllm-project/llm-compressor;

· xin3he opened on May 20, 2026

[Bug]: # Internvl quantization for language part: KeyError: 'vision_tower.embeddings.cls_token'

bugSomething isn't workingSomething isn't working

Status: Open.

#2715 In vllm-project/llm-compressor;

· Jacob682 opened on May 17, 2026