李泳桦
ec499a0104
[Cherry-pick] fix requests & block metrics ( #4500 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* [fix] fix requests & block metrics
* [chore] rename variables
2025-10-21 10:43:33 +08:00
ltd0924
3cd9d3060a
[Fearture] Support mm model close prefix cache ( #4502 )
...
* support mm prefix cache close
* add
* fix
* fix
* fix
---------
Co-authored-by: ltd0924 <luotingdan@baidu.com >
2025-10-21 09:56:47 +08:00
GoldPancake
9c7187998c
[Feature] support mtp logprob ( #4457 )
...
* support logprob in mtp
* remove debug code
* fix
* feat: add draft_logprobs for Speculative Decode MTP
* Revert "feat: add draft_logprobs for Speculative Decode MTP"
This reverts commit d5a3c5c933sunlei5788@gmail.com >
Co-authored-by: sunlei18 <sunlei18@sunlei18deMacBook-Pro.local >
2025-10-20 10:18:00 +08:00
RAM
920df5be5a
[Graph Optimization][Speculative Decoding] Fix the bug of CUDAGraph + MTP + EP ( #4430 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* Fix MTP dummy run bug
* Target Model and Draft Model using the same flag
* aovid moe bug in cudagraph padding
* In mtp replace use_cudagraph as step_use_cudagraph
2025-10-17 14:22:05 +08:00
guozhuangzhuang
cfd93c0966
fix: image token output ( #4399 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* fix: image token output
* fix: code style
* fix: CompletionOutput.decode_type
2025-10-16 14:51:32 +08:00
Yuanle Liu
83f97d1196
support speculate_limit_thinking_content_length_v2 ( #4428 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* support speculate_limit_thinking_content_length_v2
* fix
* fix import
2025-10-16 13:23:16 +08:00
gaoziyuan
74ae214f1a
fix ep perf ( #4381 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-10-15 18:38:20 +08:00
freeliuzc
c3499875bd
[MTP]support mtp chunk_prefill_v1 ( #4365 )
...
* support mtp chunk_prefill_v1
* fix mtp chunkprefill output
* fix mtp chunkprefill output, fix unit test
* fix save_output
---------
Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com >
2025-10-15 15:33:59 +08:00
ltd0924
dd425b89ed
[BugFix] fix cache port and zmq close bugs ( #4371 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* Update common_engine.py
* Update zmq_client.py
* Update expert_service.py
---------
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
2025-10-15 10:29:30 +08:00
yangjianfengo1
bc7193f21d
增加4合一视频选择 ( #4372 )
2025-10-15 09:58:24 +08:00
ltd0924
fa9a3eef4f
Update token_processor.py ( #4395 )
2025-10-15 09:43:28 +08:00
Jundong Liu
0b7a5778ab
[Executor]CUDAGraph support Speculate Decode ( #4258 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* [Executor]CUDAGraph support Speculate Decode
* fix problem
* solve problem
* fix
* fast compile
* CUDAGraph + mtp support eb5(only target model)
* Revert "fast compile"
This reverts commit 3cfe8373edgstain5555@outlook.com >
Co-authored-by: gongshaotian <gstian5555@outlook.com >
2025-10-13 15:21:41 +08:00
Zero Rains
07db281647
[Cherry-Pick][BugFix]fix the bug for prefilled_step_idx signal of cache_messager in cudagraph and PD ( #4252 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* fix the bug for prefilled_step_idx signal of cache_messager in cudagraph and PD
* support dp
2025-10-13 10:18:53 +08:00
freeliuzc
8d629568f2
[MTP]fix speculate-decoding in dpep mode ( #4351 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-10-11 17:16:57 +08:00
gaoziyuan
0c4c28d799
Update rollout_model.py ( #4349 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-10-11 11:30:05 +08:00
ltd0924
28aa18bfc1
[Feature] support prefix cache + dp ( #4356 )
...
* fix
* fix
* fix
* [Feature] support clear data
* update
* fix
* fix
* fix
* fix
* [BugFix] fix clear data
* Update api_server.py
* Update api_server.py
* [Feature] support fd decode response
* Update engine.py
* Update envs.py
* Update expert_service.py
* Update common_engine.py
* [Feature] support prefix cache + ep
* fix
* add
* Update expert_service.py
* Update common_engine.py
* Update common_engine.py
* Update common_engine.py
* Update common_engine.py
* Update common_engine.py
---------
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
Co-authored-by: ltd0924 <luotingdan@baidu.com >
2025-10-11 10:02:08 +08:00
Yuanle Liu
3c9eedd562
Simplify CUDAGraph creation logic ( #4298 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* Simplify CUDAGraph creation logic
Refactor CUDAGraph initialization to always use unique memory pool if configured.
* Conditionally import CUDAGraph based on CUDA compilation
2025-10-10 10:46:16 +08:00
ltd0924
c35a21a99a
[Feature] support fd return decode response ( #4300 )
...
* fix
* fix
* fix
* [Feature] support clear data
* update
* fix
* fix
* fix
* fix
* [BugFix] fix clear data
* Update api_server.py
* Update api_server.py
* [Feature] support fd decode response
* Update engine.py
* Update envs.py
* Update expert_service.py
* Update common_engine.py
---------
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
Co-authored-by: ltd0924 <luotingdan@baidu.com >
2025-09-28 16:11:50 +08:00
freeliuzc
c8985727a6
support mtp in hybird-dp-tp mode ( #4299 )
2025-09-28 15:58:45 +08:00
GoldPancake
076c30cb0f
fix top_p_candidates and support separate setting of sampling params for mtp ( #4189 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* fix top_p_candidates
* For separate setting params for mtp
* delete print
* fix
2025-09-28 11:41:20 +08:00
ltd0924
f8c6a354a1
[BUGFIX] clear request ( #4286 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* fix
* fix
* fix
* [Feature] support clear data
* update
* fix
* fix
* fix
* fix
* [BugFix] fix clear data
* Update api_server.py
* Update api_server.py
---------
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
2025-09-27 14:08:48 +08:00
freeliuzc
b176cba474
support mtp in ep64 ( #4280 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-09-26 15:38:03 +08:00
Yuanle Liu
dcf633c4d9
delete default value reasoning_max_tokens ( #4250 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* delete default value reasoning_max_tokens
* Adjust max_tokens and reasoning_max_tokens logic
2025-09-26 10:42:27 +08:00
Zhong Hui
213f15ef55
fix ernie vl distributed attr. ( #4259 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-09-25 20:06:29 +08:00
lizhenyun01
bab779011c
[CudaGraph] support cudagraph use shared pool ( #4199 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* support cudagraph use shared pool
* add envs
* change CUDAGRAPH_POOL_ID to int
* change CUDAGRAPH_POOL_ID to use_memory_pool
* unify use_unique_memory_pool
* fix use_unique_memory_pool
2025-09-24 21:32:04 +08:00
freeliuzc
e2b68b33c9
fix mtp in rl ( #4234 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-09-24 16:59:24 +08:00
ltd0924
1aab1c8d06
[BugFix] fix clear data ( #4227 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* [Feature] support adapter
* fix
* fix
* fix
* fix
* fix
* fix
* [BugFix] fix clear data
* Update api_server.py
2025-09-24 11:23:44 +08:00
freeliuzc
94b6e7a341
[MTP][RL]support rl reshard wenxin-tools-145 ( #4173 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* support mtp reshard in rl mode
* fix function
2025-09-23 20:40:26 +08:00
Yuanle Liu
389c5dd3a2
Each module should have its own plugins_loaded ( #4149 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-09-23 15:44:46 +08:00
Yuanle Liu
361104508e
support reasoning_max_tokens ( #4207 )
2025-09-23 15:44:41 +08:00
ltd0924
f489c9f8ef
[Feature] support adapter ( #4180 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* [Feature] support adapter
* fix
* fix
* fix
* fix
* fix
* fix
2025-09-22 19:32:24 +08:00
lzy
be98f6e950
supports internode_ll_two_stage ( #4143 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* supports internode_ll_two_stage
* supports internode_ll_two_stage
* supports internode_ll_two_stage
* supports internode_ll_two_stage
2025-09-22 14:55:06 +08:00
ltd0924
f75697c2d1
[Feature] support clear data ( #4185 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* fix
* fix
* fix
* [Feature] support clear data
* update
* fix
* fix
* fix
* fix
2025-09-21 20:41:27 +08:00
gaoziyuan
5027ed7239
【BugFif】fix ep decode ( #4138 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* suppoort expert num 3 pre rank
* fix ep decode
2025-09-17 14:18:31 +08:00
Yuan Xiaolan
25aa2d94aa
cp dynamic Cfp8 ( #4120 )
...
* supports dynamic Cfp8
* add unittest
* fix dynamic Cfp8 computing error
* fix Cfp8 for RL load
---------
Co-authored-by: carryyu <569782149@qq.com >
2025-09-17 11:55:47 +08:00
Yuanle Liu
d381fa8194
fix reasoning parsers plugin ( #4104 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-09-15 22:30:16 +08:00
freeliuzc
d2ab369427
[MTP]Support RL reshard ( #4074 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* support rl reshard
* modify model name
2025-09-15 11:47:06 +08:00
Yuanle Liu
2883746132
fix model_weights_signal ( #4092 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* fix model_weights_signal
2025-09-13 11:55:25 +08:00
chen
2485333f71
ep support logprob ( #4089 )
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
2025-09-12 21:11:16 +08:00
gaoziyuan
10768a4d79
[NewFeture]add ep rollout model init and update/clear ep buffer ( #3927 )
...
* add ep rollout model init && add deep update/clear
* fix test
2025-09-12 14:15:13 +08:00
gaoziyuan
447297a7b5
fix gid ( #4054 )
...
Co-authored-by: Divano <dddivano@outlook.com >
2025-09-11 16:08:00 +08:00
RAM
63d24b2210
[Executor] Adjust signal sending order in RL training ( #3773 ) ( #4066 )
...
* Adjust processing order
* fix bug
* fix update_parameters bug
* refine code
2025-09-11 15:41:32 +08:00
Yuanle Liu
48f2ab3fb3
support cuda graph ( #4056 )
...
* support cuda graph
* upstate
2025-09-11 11:38:32 +08:00
ltd0924
749f074e44
Update multi_api_server.py ( #4023 )
2025-09-10 17:15:01 +08:00
guozhuangzhuang
f06e3ee1fc
Use uuid to name the metrics shared folder ( #4025 )
...
* Use uuid to name the metrics shared folder
* Use uuid to name the metrics shared folder test case
2025-09-10 16:58:13 +08:00
freeliuzc
2f473ba966
[Feature][MTP]Support MTP for rl-model ( #4009 )
...
* qk norm for speculate decode C16
* support mtp in v1_scheduler mode
* support mtp rope_3d
* support mtp features
* add unit test && del some log
---------
Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com >
Co-authored-by: xiaoxiaohehe001 <hiteezsf@163.com >
2025-09-10 13:34:37 +08:00
Yuanle Liu
cce2410fad
Fix parameter shape for down projection weight ( #4028 )
2025-09-09 17:28:04 +08:00
Zero Rains
d8985a7a21
get org_vocab_size from args ( #3985 )
...
Co-authored-by: Yuanle Liu <yuanlehome@163.com >
2025-09-09 15:08:58 +08:00
Yuanle Liu
71a9127e13
Update args_utils.py
2025-09-08 01:41:43 -07:00
lizhenyun01
d40a1046de
[Feature] support rl_tp_degree ( #3934 )
...
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* [Feature] support rl_tp_degree
* add rl_tp_degree in lmhead
* add rl_tp_degree in bias
* fix split_axis=0 in bias
* fix split_axis in weight
* fix bias rl_tp_degree
* fix bias rl_tp_degree
* change attr to dict
---------
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com >
2025-09-08 16:20:32 +08:00
