GitLab

feat: enable shared expert overlap.

See merge request dcutoolkit/deeplearing/vllm!339

feat：为 GLM4 和 Llama 模型新增 MultiModalConfigProxy，以支持扁平配置与嵌套的多模态配置（text_config）

See merge request dcutoolkit/deeplearing/vllm!336

V0.9.2 dev pdncclinit

See merge request dcutoolkit/deeplearing/vllm!333

fix: 只有当kv block中不含有MTP的假数据时才会被cached，以修复cache_full_blocks同一个kv block保存两次的bug

See merge request dcutoolkit/deeplearing/vllm!331

perf: 加速 v1 InputBatch.add_request 的 token_ids 拷贝

See merge request dcutoolkit/deeplearing/vllm!330

新增环境变量开关 VLLM_V1_FAST_TOKEN_ID_COPY（默认关闭）。开启后在 CachedRequestState 中缓存 prompt_token_ids 的 np.int32，并在 add_request 里用 np.copyto 写入 token_ids_cpu，避免长 prompt 场景反复 list->NumPy 转换开销（尤其是抢占/反复进出 batch 时）

perf: DS-量化模型融合qa和kva的gemm

See merge request dcutoolkit/deeplearing/vllm!329

fix: chunk-prefill生成的token被错误返回

See merge request dcutoolkit/deeplearing/vllm!327

V0.9.2 dev main+mtp zero

See merge request dcutoolkit/deeplearing/vllm!325

V1 采样器：新增 reduced top-k/top-p 采样路径

See merge request dcutoolkit/deeplearing/vllm!321

修复w8a8 triton config 择优位运算可能引发torch compile 编译错误，修复smquant w8a8 权重后处理位置

See merge request dcutoolkit/deeplearing/vllm!320

# Conflicts:
#	vllm/envs.py

新增环境变量 VLLM_V1_USE_REDUCED_TOPK_TOPP_SAMPLER 用于开关控制
扩展 SamplingMetadata，增加 max_top_k 与 has_any_no_top_k
在 InputBatch 侧计算 top-k 的主机端汇总信息，避免 device 同步
更新 Sampler/TopKTopPSampler 传递并使用新参数以启用优化采样

mla模型P、D单实例单机的任意切分方式（满足D的tp>=P的tp)使用

See merge request dcutoolkit/deeplearing/vllm!315

update VLLM_USE_LIGHTOP_RMS_ROPE_CONCAT impl