Kane2011
b4fef2cf29
[MetaxGPU] Support FastDeploy on metax gpu ( #3241 )
...
* [MetaxGPU] Support FastDeploy on metax gpu
* Update metax_worker.py
1. change worker log;
2. remove custom allreduce, adapt it later;
3. remove cuda graph;
* Update __init__.py
1. remove metax's key work comment
* Update __init__.py
1. remove metax's key word comment;
2. add fused_moe_kernel_paddle import
---------
Co-authored-by: yongqiangma <xing.wo@163.com >
2025-08-13 11:11:54 +08:00
RichardWooSJTU
283da92bfa
fix ep lm head ( #3244 )
...
Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com >
2025-08-12 15:38:28 +08:00
Jiang-Jia-Jun
c56c99837a
Revert "[BugFix] num_seqs ( #3291 )" ( #3316 )
...
This reverts commit e0aeac58e1
2025-08-11 16:16:51 +08:00
Yuanle Liu
9571c458f0
enhance eos_tokens ( #3274 )
...
* enhance eos_tokens
* update
* update
2025-08-11 14:47:52 +08:00
Zero Rains
42af0b4b64
[V1 Loader] Support DeepSeekV3(bf16) ( #3294 )
...
* Support new loader for DeepSeekV3(bf16)
* update paddle version
* remove useless attr
2025-08-11 13:39:28 +08:00
lizexu123
e0aeac58e1
[BugFix] num_seqs ( #3291 )
...
* fix num_seqs
* merge develop
2025-08-11 13:38:55 +08:00
gaoziyuan
a799d14df1
[Bugfix] Fix model accuracy in some ops ( #3231 )
...
* fix noaux_tc op
* fix
* update
* fix qk norm
* fix linear for prequant loader
* test
* fix
* fix
* rm some print
* fix noaux_tc op
* test
* Fix the confused enable_early_stop when only set early_stop_config (#3214 )
* fix the confused early_stop_config when only set early_stop_config
* pre-commit
* write a general method
* Add ci case for min token and max token (#3229 )
Co-authored-by: xujing43 <xujing43@baidu.com >
* add some evil cases (#3240 )
* add repitation early stop cases
* add repitation early stop cases
* add bad cases
* add bad cases
* add evil cases
* qwen3_moe (#3084 )
* [Feature] support seed parameter (#3161 )
* support seed
* fix
* add SamplingMetadata seed test
* The next_tokens values are inconsistent!
* add air and rejection seed test
* fix
* add SamplingParams seed test
* fix seed=0
* Default to defualt
* fix
* fix args_utils
* fix review
* fix review
* fix
* fix
* add xpu,gcu,iluvatar support seed
* fix
* 【Fix Bug】 修复 fa3 支持集中式bug (#3235 )
* fix fa3 集中式bug
* 增加qknorm参数
* fix qk norm
* fix
* update
* fix linear for prequant loader
* fix
* fix
* rm some print
* fix
* fix moe init weight&scale
* fix moe init weight&scale
---------
Co-authored-by: bukejiyu <395822456@qq.com >
Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com >
Co-authored-by: Zero Rains <linjunlu@zerorains.top >
Co-authored-by: xjkmfa <108254620+xjkmfa@users.noreply.github.com >
Co-authored-by: xujing43 <xujing43@baidu.com >
Co-authored-by: Divano <dddivano@outlook.com >
Co-authored-by: bukejiyu <52310069+bukejiyu@users.noreply.github.com >
Co-authored-by: lizexu123 <39205361+lizexu123@users.noreply.github.com >
Co-authored-by: yangjianfengo1 <125249383+yangjianfengo1@users.noreply.github.com >
Co-authored-by: qingqing01 <dangqingqing@baidu.com >
2025-08-08 17:30:37 +08:00
Zero Rains
ce1f353c70
Move create_parameters to __init__ in FuseMOE for CultassBackend and TritonBackend ( #3148 )
...
* w4a8 bug
* fix w4a8 bug
* remove code
* modify the triton backend
* fix ep
* fix the bug with tensor_wise_fp8 in triton backend
* fix the RL
* fix bug by merge
* fix the bug in w4a8
* fix the tensor_wise_fp8 bug
* fix RL
2025-08-08 15:55:47 +08:00
freeliuzc
71267840f7
【Fix】fix mtp bug ( #3139 )
2025-08-08 13:30:12 +08:00
bukejiyu
b76b17fc1b
qwen3 0.3B fix ( #3255 )
Deploy GitHub Pages / deploy (push) Has been cancelled
2025-08-08 11:35:40 +08:00
yzwu
fbdd6b0663
[Iluvatar GPU] Optimze attention and moe performance ( #3234 )
2025-08-08 10:51:24 +08:00
bukejiyu
37569cca86
[feat]add fast_weights_iterator ( #3258 )
...
Deploy GitHub Pages / deploy (push) Has been cancelled
* add fast_weights_iterator
* update
* update
2025-08-07 22:36:46 +08:00
bukejiyu
9408e667a5
[bugfix]fix blockwisefp8 and all_reduce ( #3243 )
...
* fix
* update
* fix linear for prequant loader
2025-08-06 23:54:33 +08:00
yangjianfengo1
3a15e0c53e
【Fix Bug】 修复 fa3 支持集中式bug ( #3235 )
...
Deploy GitHub Pages / deploy (push) Has been cancelled
* fix fa3 集中式bug
* 增加qknorm参数
2025-08-06 16:24:27 +08:00
lizexu123
afff4d37ea
[Feature] support seed parameter ( #3161 )
...
* support seed
* fix
* add SamplingMetadata seed test
* The next_tokens values are inconsistent!
* add air and rejection seed test
* fix
* add SamplingParams seed test
* fix seed=0
* Default to defualt
* fix
* fix args_utils
* fix review
* fix review
* fix
* fix
* add xpu,gcu,iluvatar support seed
* fix
2025-08-06 15:20:47 +08:00
bukejiyu
20839abccf
qwen3_moe ( #3084 )
2025-08-06 14:45:27 +08:00
Yuan Xiaolan
7ce00e597c
support qk norm ( #3145 )
2025-08-05 16:46:14 +08:00
RAM
4a10e29804
fix mla attention backend ( #3176 )
2025-08-05 16:43:15 +08:00
Yuan Xiaolan
af543b7f0f
revise get_moe_scores ( #3164 )
2025-08-05 16:43:07 +08:00
RichardWooSJTU
1e9a8e8cef
fix lm head bias ( #3185 )
...
Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com >
2025-08-05 15:40:24 +08:00
RichardWooSJTU
f5c64a074c
[EP] Refactor DeepEP Engine Organization for Mixed Mode & Buffer Management Optimization ( #3182 )
...
* Add support for mixed-ep across multi nodes
* code refine
---------
Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com >
2025-08-05 15:40:11 +08:00
lizhenyun01
fe540f6caa
[plugin] Custom model_runner/model support ( #3186 )
...
* support custom model&&model_runner
* fix merge
* add test && update doc
* fix codestyle
* fix unittest
* load model in rl
2025-08-04 18:52:39 -07:00
Yuan Xiaolan
1f8289e106
fix expertwise_scale ( #3181 )
2025-08-04 20:06:15 +08:00
gaoziyuan
4021d66ea5
【Feature】add fd plugins && rm model_classes ( #3123 )
...
* add fd plugins && rm model_classed
* fix reviews
* add docs
* fix
* fix unitest ci
2025-08-03 19:53:20 -07:00
bukejiyu
1582814905
fix load_pre_sharded_checkpoint ( #3152 )
...
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
2025-08-04 10:44:20 +08:00
ApplEOFDiscord
b71cbb466d
[Feature] remove dependency on enable_mm and refine multimodal's code ( #3014 )
...
* remove dependency on enable_mm
* fix codestyle check error
* fix codestyle check error
* update docs
* resolve conflicts on model config
* fix unit test error
* fix code style check error
---------
Co-authored-by: shige <1021937542@qq.com >
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
2025-08-01 20:01:18 +08:00
yangjianfengo1
64d7a3194d
集中式支持fa3 ( #3112 )
2025-08-01 18:03:36 +08:00
Ryan
94264bbf60
[Code Simplification] Refactor Post-processing in VL Model Forward Method ( #2937 )
...
* rm sth useless
* refactor model forward
* mv bool index to kernel
2025-08-01 17:28:07 +08:00
chen
a2f5cc54f8
moe preprocess op support 160 experts and fused_moe triton kernel name add K ( #3121 )
2025-08-01 10:46:20 +08:00
Yuan Xiaolan
5f56d289a7
fix is_permuted ( #3098 )
...
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
2025-07-31 19:58:05 +08:00
RAM
d850660872
[Executor] Refactor GetBlockShapeAndSplitKVBlock Kernel ( #2989 )
...
* reset decoder_block_shape_q buffer
* refactor GetBlockShapeAndSplitKVBlock Kernel and cudagraph padding batch
* update decode_max_tile_size
* fix pre-commit
* update block_multihead_attn_backend
* update flas attn backend
* update MLA Attention
* update XPU Attention
* update gcu,iluvatar model runner
* Update MTP
* fix MTP bug
2025-07-31 00:09:31 +08:00
bukejiyu
db698bda01
qwen loader ( #3057 )
2025-07-30 19:09:38 +08:00
zhink
d89b6dd43f
adapter qwen3 moe attr for init ( #3066 )
...
adapter qwen3 moe attr for init
2025-07-30 16:49:28 +08:00
bukejiyu
8e203666d9
w4a8 offline ( #3074 )
...
* w4a8 offline
* update
* update
* update
2025-07-30 16:33:30 +08:00
Yuan Xiaolan
35935da9e5
support W4A8 EPLB ( #3075 )
2025-07-30 14:34:12 +08:00
Zero Rains
b2f9a42d87
[Feature] Support repetition early stop ( #3024 )
...
* support repetition early stop and support user to set the parameter
* remove log
* fix codestyle
* add the early_stop_config to rollout_config
* update config and EarlyStopper class
* fix the bug for triton
* modify the stop method
* update description
* modify the usage for stop_flags
---------
Co-authored-by: Yuanle Liu <yuanlehome@163.com >
2025-07-29 22:42:54 +08:00
Yuan Xiaolan
3214fb5393
support model loading for w4a8 offline quant ( #3064 )
...
支持W4A8 EP 对离线量化权重的load
2025-07-29 21:54:37 +08:00
Longzhi Wang
be0a0f2bb2
fix arguement error in ep when pd ( #3060 )
2025-07-29 17:17:24 +08:00
YuanRisheng
502ee92a0a
Unify server-side and model-side Config (Part3) ( #3047 )
...
* merge model config
* fix arch
* fix rl
2025-07-29 17:07:44 +08:00
Longzhi Wang
907d561523
fix ep when paddle version mismatch ( #3056 )
2025-07-29 15:06:49 +08:00
JYChen
dafe02a7b9
[stop sequence] support stop sequence ( #3025 )
...
* stop seqs in multi-ends
* unittest for gpu stop op
* kernel tid==0
2025-07-29 14:17:37 +08:00
Yuan Xiaolan
b1d787a272
[fix] w4a8 model loading and hadamard config ( #3013 )
2025-07-28 18:17:59 +08:00
K11OntheBoat
83048bbe55
[Feature] Deepseekv3 supports cudagraph ( #3041 )
...
Co-authored-by: K11OntheBoat <“ruianmaidanglao@163.com ”>
2025-07-28 17:12:54 +08:00
AIbin
ec52d39e68
【Inference Optimize】Update wint2 weight n-dim reorder ( #3042 )
2025-07-28 16:31:56 +08:00
YuanRisheng
bddf403576
Unify server-side and model-side Config (Part2) ( #3035 )
...
* merge speculative and graph opt conifg
* add attr
2025-07-28 15:31:48 +08:00
Longzhi Wang
247010d298
fix arguement error ( #3030 )
2025-07-28 11:03:29 +08:00
YuanRisheng
6ccc10ad47
Unify server-side and model-side Config (Part1) ( #3018 )
...
* move cache config
* fix mtp
2025-07-28 10:51:52 +08:00
Longzhi Wang
0700c90caa
[Feat] support mixed ep ( #2969 )
...
Deploy GitHub Pages / deploy (push) Has been cancelled
* Support mixed ep
* fix comment
* fix comment
* update mixep
* fix conflict
* fix typo
* update
* fix typo
* fix code style
* fix conflict
2025-07-25 15:29:30 +08:00
chen
332154f504
[feature] Support FA2 ( #3009 )
2025-07-25 14:09:00 +08:00
xiaoxiaohehe001
2970b00dfa
[Feature] Support_eplb ( #2997 )
...
Deploy GitHub Pages / deploy (push) Has been cancelled
* [Feature] support_eplb
* [Feature] support_eplb
* [Fix] fix mm ep
2025-07-24 20:22:45 +08:00
