Commits · 1a56d6cbbd045b3923f7ba62d3dbc256c5971b17 · OpenDAS / vllm_cscc

16 Sep, 2025 1 commit
- 添加mori ep · 1a56d6cb
  王敏 authored Sep 16, 2025
  
  1a56d6cb
10 Sep, 2025 1 commit
- 临时添加mori代码 · 3cb11400
  王敏 authored Sep 10, 2025
  
  3cb11400
07 Sep, 2025 1 commit
- 临时添加cudagraph代码，目前还有问题 · fe70dcb2
  王敏 authored Sep 07, 2025
  
  fe70dcb2
06 Sep, 2025 4 commits
- feat: slimquant_w4a8量化加入ep_moe适配，修改ep_moe中调用接口为：self.quant_method.apply_ep() · 9c86df96
  jujl1 authored Sep 05, 2025
  
  9c86df96
- 解决大EP mtp报错问题 · 121db653
  王敏 authored Sep 06, 2025
  
  121db653
- 修复all2all报维度不匹配问题 · 8f66f64b
  王敏 authored Sep 06, 2025
  
  8f66f64b
- 修复all2all报维度不匹配问题 · 578d3e97
  王敏 authored Sep 06, 2025
  
  578d3e97
05 Sep, 2025 1 commit
- dummy_run时使用随机输入 · cde83ab0
  王敏 authored Sep 05, 2025
  
  cde83ab0
04 Sep, 2025 4 commits
- 去掉无效代码 · 4e22d272
  王敏 authored Sep 04, 2025
  
  4e22d272
- 1.优化大EP，合入grouped gemm · d997afc4
  王敏 authored Sep 04, 2025
```
2.解决mtp >1 大EP推理all gather卡住问题
```
  d997afc4
- 暂时去掉profilling标志位，避免影响其他模型 · 6f5d76dc
  王敏 authored Sep 04, 2025
  
  6f5d76dc
- add glm-4.5-air-tp8 nn moe configs · 7a97637e
  zhuwenwen authored Sep 04, 2025
  
  7a97637e
03 Sep, 2025 2 commits
- [fix]fix tests of v1 and worker · a7668e46
  zhuwenwen authored Sep 03, 2025
  
  a7668e46
- fix deepseek pp + mtp issue · bfffd59a
  lizhigong authored Sep 02, 2025
  
  bfffd59a
02 Sep, 2025 1 commit
- remove topk_ids · 4a62a3eb
  zhuwenwen authored Sep 02, 2025
  
  4a62a3eb
01 Sep, 2025 3 commits
- [fix]fix tests of fused_moe · 58bbb720
  zhuwenwen authored Sep 01, 2025
  
  58bbb720
- [BugFix]支持v1 engine pp deepseek · 4a946680
  zhuwenwen authored Sep 01, 2025
  
  4a946680
- add dca and sparse attention support on rocm · 12112a4d
  zhuwenwen authored Sep 01, 2025
  
  12112a4d
29 Aug, 2025 3 commits
- fix bug in zero-overhead core · 7d1b0c5e
  lizhigong authored Aug 29, 2025
  
  7d1b0c5e
- add benchmark_moe.py to perf · fb94600a
  zhuwenwen authored Aug 29, 2025
  
  fb94600a
- 修复residual FP16 overflow，解决mtp采样率和数据集精度的冲突 · 080ed180
  yangql authored Aug 29, 2025
  
  080ed180
28 Aug, 2025 2 commits
- [feat]上传初版基于all2all通信的大EP代码 · d04683a4
  王敏 authored Aug 28, 2025
  
  d04683a4
- 优化 block_tables 的生成逻辑，增加对混合情况的检测，确保在存在空和非空块时正确计算最大块长度。 · 4932783d
  laibao authored Aug 28, 2025
  
  4932783d
27 Aug, 2025 2 commits
- feat: w4a8 marlin合入主分支, 通过-q slimquant_w4a8_marlin 可以启用 · 533d4c72
  jujl1 authored Aug 27, 2025
  
  533d4c72
- when use tbo, deepseek prefill use eager · 8702264b
  lizhigong authored Aug 27, 2025
  
  8702264b
26 Aug, 2025 4 commits
- feat: w4a8 marlin 初版合入 · d0181e5a
  jujl1 authored Aug 25, 2025
  
  d0181e5a
- [fix]fix tests of kernels · 31584b45
  zhuwenwen authored Aug 26, 2025
  
  31584b45
- debug and fix error · a01df4a6
  lizhigong authored Aug 26, 2025
  
  a01df4a6
- v1 engine eager tbo support mla attention · 91ccb1cc
  lizhigong authored Aug 26, 2025
  
  91ccb1cc
25 Aug, 2025 3 commits
- fix self.drafter error · ba29eebb
  lizhigong authored Aug 25, 2025
  
  ba29eebb
- 临时上传大ep代码 · dbd0bda6
  王敏 authored Aug 25, 2025
  
  dbd0bda6
- Remove the synchronization introduced by the verification logic when mtp is enabled. · 2fe2d237
  lizhigong authored Aug 25, 2025
  
  2fe2d237
22 Aug, 2025 2 commits
- update cuda_graph_sizes · 15347448
  zhuwenwen authored Aug 22, 2025
  
  15347448
- support adaptive max_num_seqs and cuda_graph_sizes · 65225897
  zhuwenwen authored Aug 22, 2025
  
  65225897
21 Aug, 2025 2 commits
- revert 修改flashmla的接口 · 9b67a1e0
  zhuwenwen authored Aug 21, 2025
  
  9b67a1e0
- update gen_model_json · 9c266e36
  zhuwenwen authored Aug 21, 2025
  
  9c266e36
20 Aug, 2025 4 commits
- set v1 attention use fa · f7beb354
  zhuwenwen authored Aug 20, 2025
  
  f7beb354
- update value_stride_order · 716852ab
  zhuwenwen authored Aug 20, 2025
  
  716852ab
- [fix]v0 SamplerOutput在非tree decoding时不传入logits · 01692be6
  王敏 authored Aug 20, 2025
  
  01692be6
- add ggml_dequantize · 9a4a94ee
  zhuwenwen authored Aug 20, 2025
  
  9a4a94ee