- 04 Feb, 2026 1 commit
-
-
wanglong3 authored
feat: support moe sum when topk==9 bugfix: Fix mtp model load error when eable shared experts fusion.
-
- 26 Jan, 2026 1 commit
-
-
zhuwenwen authored
-
- 23 Jan, 2026 1 commit
-
-
zhuwenwen authored
-
- 22 Jan, 2026 2 commits
- 21 Jan, 2026 4 commits
-
-
zhuwenwen authored
for qwen3, VLLM_USE_FUSED_RMS_ROPE=1 (default)
-
laibao authored
- 移除 VLLM_USE_MARLIN_W16A16_MOE 环境变量 - 初始化阶段基于 lightop 探测并缓存 _marlin_w16a16_moe_enabled,满足条件强制 use_nn_moe=False - 权重加载后按缓存结果一次性 Marlin pack;运行时按 packed 标记走 Marlin fast path
-
laibao authored
新增 VLLM_USE_FUSED_RMS_ROPE 分支,走 fused 路径 注册 torch.ops.vllm.rms_rotary_embedding_fuse(direct_register_custom_op) cos_sin_cache 自动转 device/dtype 并缓存,避免每次重复拷贝
-
SAC_fanth authored
-
- 20 Jan, 2026 2 commits
- 19 Jan, 2026 2 commits
- 17 Jan, 2026 5 commits
- 16 Jan, 2026 7 commits
-
-
zhuwenwen authored
-
zhuwenwen authored
add VLLM_USE_FUSED_CACHE_QUANT_BMM_MLA to use fused rmsnorm + contiguous + rope(for dpsk-v3) + concat_and_cache_mla + q quant, control bmm(todo) + cat +mla (fp8)
-
laibao authored
新增环境变量 VLLM_MOE_ROUTER_CAPTURE / DIR / RANK / MAX_LAYERS / NUM_TOKENS_* 用于开关与过滤控制 新增 router_capture.py,支持按 num_tokens 分桶抓取 router logits 并落盘 在 qwen3_moe 中接入抓取逻辑,默认关闭,仅在开启时记录 固定 skip_profile / skip_stack_funcs 为默认启用,避免抓到 warmup/profile 形状 统一配置入口到 vllm.envs,作为运行时基准
-
zhuwenwen authored
-
zhuwenwen authored
-
zhuwenwen authored
-
zhuwenwen authored
-
- 15 Jan, 2026 3 commits
- 14 Jan, 2026 2 commits
- 13 Jan, 2026 3 commits
- 12 Jan, 2026 6 commits
- 09 Jan, 2026 1 commit
-
-
laibao authored
- 在 post-load hook 中对 w13/w2 做 per-expert Marlin pack,并替换为 packed 参数 - Marlin fast path 仅接受 packed 权重;未预打包则 fail fast,避免运行时 packing 峰值/不确定性 - 更新 Marlin wrapper 的入参与 shape 推导(从 packed layout 计算 K/N)
-