Commits · 389ebcf734ec1217d152a0adcc49ab7b5d26a504 · OpenDAS / vllm_cscc · GitLab

24 Mar, 2025 1 commit
- [fix]修复fused_moe.py中fused_moe接口未初始化moe_ep_size导致的deekseek等模型报错 · 389ebcf7
  zhuwenwen authored Mar 24, 2025
  
  389ebcf7
17 Mar, 2025 1 commit
- update fused_moe.py · cbeaf0b5
  zhuwenwen authored Mar 17, 2025
  
  cbeaf0b5
15 Mar, 2025 1 commit
- 支持fusemoe对int4的scale合zero合并读取操作 · 2b47d4fa
  zhuwenwen authored Mar 15, 2025
  
  2b47d4fa
14 Mar, 2025 1 commit
- 增加deepseek v3 block-int8量化支持 · ba1ff372
  zhuwenwen authored Mar 14, 2025
```
增加线性int8 gemm配置
```
  ba1ff372
12 Mar, 2025 1 commit
- [feat]添加BW deepseek_v3 fused_moe configs · abac3adc
  zhuwenwen authored Mar 12, 2025
  
  abac3adc
11 Mar, 2025 1 commit
- [feat]优化fused_moe tuning，上传K100_AI的configs · e3a0d6bb
  zhuwenwen authored Mar 11, 2025
  
  e3a0d6bb
28 Feb, 2025 1 commit
- 1. 修复w8a8找不到对应config的bug · baf7d06a
  zhuwenwen authored Feb 28, 2025
```
2. 删除vllm的w8a8 config
```
  baf7d06a
20 Feb, 2025 3 commits
- [Bugfix] Fix deepseekv3 grouped topk error (#13474) · ed6e9075
  chenxiaobing authored Feb 20, 2025
```
Signed-off-by: Chen-XiaoBing <chenxb002@whu.edu.cn>
```
  ed6e9075
- 适配deepseekv3\v2 moe awq的推理支持 · 47bd229c
  yangql authored Feb 20, 2025
  
  47bd229c
- [ROCm][MoE] mi300 mixtral8x7B perf for specific BS (#13577) · 0d243f2a
  Divakar Verma authored Feb 19, 2025
```
Signed-off-by: Divakar Verma <divakar.verma@amd.com>
```
  0d243f2a
19 Feb, 2025 2 commits
- add deepseek-v2 tp2 fused_moe_config · 177520a9
  zhuwenwen authored Feb 19, 2025
  
  177520a9
- [ROCm][MoE configs] mi325 mixtral & mi300 qwen_moe (#13503) · 8aada19d
  Divakar Verma authored Feb 19, 2025
  
  8aada19d
18 Feb, 2025 5 commits
- [V1] Optimize handling of sampling metadata and req_ids list (#13244) · 30172b49
  Nick Hill authored Feb 18, 2025
```
Signed-off-by: Nick Hill <nhill@redhat.com>
```
  30172b49
- add deepseek-v2 tp1 fused_moe_config · f0ac5f23
  zhuwenwen authored Feb 18, 2025
  
  f0ac5f23
- 添加w8a8 关于0.7.2的修改 · 8c61b2ed
  gaoqiong authored Feb 18, 2025
  
  8c61b2ed
- [Misc] Remove dangling references to `SamplingType.BEAM` (#13402) · efbe8544
  Harry Mellor authored Feb 18, 2025
  
  efbe8544
- [Bugfix][CI][V1] Work around V1 + CUDA Graph + torch._scaled_mm fallback issue (#13425) · b3942e15
  Tyler Michael Smith authored Feb 17, 2025
```
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
```
  b3942e15
17 Feb, 2025 2 commits
- [fix]1.解决deepseekv2/v3 nn layout加载报错问题;2.上传ep1和ep2的fused_moe config · 249fca2a
  王敏 authored Feb 17, 2025
  
  249fca2a
- [Model] Support Mamba2 (Codestral Mamba) (#9292) · 1f69c4a8
  Tyler Michael Smith authored Feb 17, 2025
```
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
Co-authored-by: Yu Chin Fabian Lim <flim@sg.ibm.com>
```
  1f69c4a8
16 Feb, 2025 1 commit
- [Quant] Add `SupportsQuant` to phi3 and clip (#13104) · 12913d17
  Kyle Sayers authored Feb 15, 2025
  
  12913d17
15 Feb, 2025 1 commit
- [AMD] [Model] DeepSeek tunings (#13199) · ed0de3e4
  rasmith authored Feb 15, 2025
  
  ed0de3e4
14 Feb, 2025 3 commits
- [Core] Reduce TTFT with concurrent partial prefills (#10235) · 3bcb8c75
  Joe Runde authored Feb 14, 2025
```
Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>
Signed-off-by: Prashant Gupta <prashantgupta@us.ibm.com>
Co-authored-by: Prashant Gupta <prashantgupta@us.ibm.com>
Co-authored-by: Cody Yu <hao.yu.cody@gmail.com>
```
  3bcb8c75
- [Quant][Perf] Use moe_wna16 kernel by default for MoEs with many experts (#13236) · 5e5c8e09
  Michael Goin authored Feb 14, 2025
```
Signed-off-by: mgoin <mgoin64@gmail.com>
```
  5e5c8e09
- [Kernel][Bugfix] Refactor and Fix CUTLASS 2:4 Sparse Kernels (#13198) · c1e37bf7
  Tyler Michael Smith authored Feb 13, 2025
```
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
```
  c1e37bf7
13 Feb, 2025 3 commits
- Optimize moe_align_block_size for deepseek_v3 (#12850) · 2344192a
  Michael Goin authored Feb 13, 2025
```
Signed-off-by: mgoin <mgoin64@gmail.com>
```
  2344192a
- Allow Unsloth Dynamic 4bit BnB quants to work (#12974) · cb944d58
  Daniel Han authored Feb 12, 2025
  
  cb944d58
- 添加moe模型部分fused_moe config · a10c70e7
  王敏 authored Feb 13, 2025
  
  a10c70e7
12 Feb, 2025 3 commits
- [Bugfix] Allow fallback to AWQ from AWQMarlin at per-layer granularity (#13119) · 09972e71
  Michael Goin authored Feb 12, 2025
  
  09972e71
- [CORE] [QUANT] Support for GPTQModel's `dynamic` quantization per module override/control (#7086) · 36a08630
  Qubitium-ModelCloud authored Feb 13, 2025
  
  36a08630
- 优化ep moe · 44d451ff
  王敏 authored Feb 12, 2025
  
  44d451ff
11 Feb, 2025 3 commits
- Fix initializing GGUF weights for ColumnParallelLinear when using tensor parallel > 1 (#13023) · 2b25b7d2
  Szymon Ożóg authored Feb 11, 2025
  
  2b25b7d2
- [feat]添加ep moe功能 · d0dafaf4
  王敏 authored Feb 11, 2025
  
  d0dafaf4
- update fused_moe.py · a27fdb55
  zhuwenwen authored Feb 11, 2025
  
  a27fdb55
08 Feb, 2025 4 commits
- update layer.py · 54944679
  zhuwenwen authored Feb 08, 2025
  
  54944679
- [Hardware][Intel-Gaudi] Enable long-contexts + LoRA support for Intel Gaudi (#12812) · 2880e21e
  Sanju C Sudhakaran authored Feb 08, 2025
```
Signed-off-by: Sanju C Sudhakaran <scsudhakaran@habana.ai>
```
  2880e21e
- [feat] add nn_moe · 37b63c24
  zhuwenwen authored Feb 08, 2025
  
  37b63c24
- add bw fused_moe_config · 2dc7ec2f
  zhuwenwen authored Feb 08, 2025
  
  2dc7ec2f
07 Feb, 2025 2 commits
- [ROCm] [Feature] [Doc] [Dockerfile] [BugFix] Support Per-Token-Activation... · eaa92d44
  TJian authored Feb 08, 2025
```
[ROCm] [Feature] [Doc] [Dockerfile] [BugFix] Support Per-Token-Activation Per-Channel-Weight FP8 Quantization Inferencing (#12501)
```
  eaa92d44
- PR #12718 (#12718) · 538fab93
  Amit Garg authored Feb 07, 2025
  
  538fab93
06 Feb, 2025 1 commit
- [MISC] Check space in the file names in the pre commit checks (#12804) · 741429a4
  Lu Fang authored Feb 06, 2025
```
Signed-off-by: Lu Fang <lufang@fb.com>
```
  741429a4