- 03 Feb, 2026 1 commit
-
-
laibao authored
- 新增 pr/kvpress_v2.md 文档，整理 KVPress v2 相关的 15 个 PR - 包含各 PR 的目的、范围、内容、风险和待处理问题 - 涵盖 KV 压缩配置、预算计算、Triton 内核、调度层逻辑、runner 侧实现等模块
-
- 27 Jan, 2026 4 commits
- 23 Jan, 2026 2 commits
- 22 Jan, 2026 5 commits
- 21 Jan, 2026 5 commits
- 20 Jan, 2026 2 commits
- 06 Jan, 2026 5 commits
- 05 Jan, 2026 5 commits
- 04 Jan, 2026 9 commits
-
-
zhuwenwen authored
V0.9.2 dev pdncclinit See merge request dcutoolkit/deeplearing/vllm!333
-
xiabo authored
-
xiabo authored
-
zhuwenwen authored
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
-
zhuwenwen authored
perf: 加速 v1 InputBatch.add_request 的 token_ids 拷贝 See merge request dcutoolkit/deeplearing/vllm!330
-
jujl1 authored
-
laibao authored
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）
-
zhuwenwen authored
perf: DS-量化模型融合qa和kva的gemm See merge request dcutoolkit/deeplearing/vllm!329
-
wujl5 authored
-
- 31 Dec, 2025 2 commits
