GitLab

Fix：Extend MAX_VPT from 32 to 256 to accommodate large-scale MoE models (e.g., GLM-5-quantized model).

优化015 fp8 kvscale

See merge request dcutoolkit/deeplearing/vllm!473

perf:Deepseek v2模型增加rmsQuant和siluMulQuant融合

See merge request dcutoolkit/deeplearing/vllm!468

support qwen3-asr

See merge request dcutoolkit/deeplearing/vllm!466

[perf]glm4_moe模型适配rmsquant和silu_quant融合算子

See merge request dcutoolkit/deeplearing/vllm!467

[perf]添加Module支持split qkv+rmsnorm+rope+kvcache融合算子，GLM4_MOE完成适配

See merge request dcutoolkit/deeplearing/vllm!465

修复dsa的workspace的bug，以及添加环境变量关闭DSAVLLM_DISABLE_DSA=1

See merge request dcutoolkit/deeplearing/vllm!463

修复channel-int8 的block_shape读取bug

See merge request dcutoolkit/deeplearing/vllm!462

feat:fix dsa

See merge request dcutoolkit/deeplearing/vllm!457

解决custom allreduce在K100AI上新模型报错问题

See merge request dcutoolkit/deeplearing/vllm!459

[perf]合入lightop topp_topk 融合算子

See merge request dcutoolkit/deeplearing/vllm!456

修复awq-marlin的bug

See merge request dcutoolkit/deeplearing/vllm!454

fix(qwen3): 修复VL场景fused RoPE分支条件，并更正qwen3_moe中rms_mrope参数顺序

See merge request dcutoolkit/deeplearing/vllm!451

[perf]优化异步调度+并行解码 step之间的空泡，实现kernel提前下发

See merge request dcutoolkit/deeplearing/vllm!449

修复qwen3.5的使用dtype为fp16的picecwise的推理模式

See merge request dcutoolkit/deeplearing/vllm!448

feat(sampler): 增加 reduced topk+topp 采样快速路径以降低全词表 softmax 开销

See merge request dcutoolkit/deeplearing/vllm!447

新增 VLLM_V1_USE_REDUCED_TOPK_TOPP_SAMPLER 开关并补充适用场景说明

在 V1 GPU 输入批预计算 max_top_k/has_any_no_top_k，native sampler 满足条件时走 reduced fast path，异常自动回退

修复block-wise 用triton报错的bug

See merge request dcutoolkit/deeplearing/vllm!444