[BugFix] Add prefill restrictions for chunked_prefill+VL #2983
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
yuanlehome
approved these changes
Jul 23, 2025
yuanlehome
changed the title
luukunn
added a commit
to luukunn/FastDeploy
that referenced
this pull request
Jul 29, 2025
* [MTP Fix] Fix code and register cpp operators (PaddlePaddle#2965) * fix rl config local rank (PaddlePaddle#2957) * [FIX]fix rejection sampling when topp=0 using _SAMPLING_EPS (PaddlePaddle#2967) * fix rejection sampling when topp=0 * fix * [SOT] Add sot warmup (NVIDIA GPU Only) (PaddlePaddle#2929) * add sot warmup * fix code style * change batch_size list * add param to config * rm free_list settings && set sot_warmup_sizes * finish debug with dynamic dims by type annotations * add profile_run guard * rm sth useless * support chunk_prefill in fa3 * 【Infer】Improve the performance block_wise_fp8 of triton_moe_backend (PaddlePaddle#2942) * Update README.md * Update README.md * delete max-len (PaddlePaddle#2959) * [CI] add codestyle_check action (PaddlePaddle#2972) * [CI] add codestyle_check action * [CI] Integrate codestyle check via pre-commit in GitHub Actions * fix mtp bug in pd-split mode (PaddlePaddle#2970) * [BugFix] Add prefill restrictions for chunked_prefill+VL (PaddlePaddle#2983) * Fix performance degradation bug of custom_all_reduce (PaddlePaddle#2981) * FA3 fix bug (PaddlePaddle#2987) * polish code for prefill restrictions (PaddlePaddle#2991) * [Feature] Support block scheduler v1 for FD (PaddlePaddle#2928) * Support FD block scheduler v1 * Support FD block scheduler v1 * Support FD block scheduler v1 * Fix according to copilot review * Fix according to review * Remove is_dummy * Fix bug when real_bsz=1 * Fix infer first token cost time --------- Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com> * update (PaddlePaddle#2978) * [Code Simplification] fix init_distributed_environment() (PaddlePaddle#2982) * support c4 attn && fix cache * fix chunk_prefill * [benchmark] add quantization for benchmark yaml (PaddlePaddle#2995) * [Fix] fix mm ep empty run (PaddlePaddle#2999) * add ci reuse action (PaddlePaddle#2968) * add ci reuse action * fix code formatting * update * [Feature] multi-source download (PaddlePaddle#2986) * multi-source download * multi-source download * huggingface download revision * requirement * style * add revision arg * test * pre-commit * [LLM] update function name (PaddlePaddle#2985) * [LLM] update function name * [BugFix] fix multinode deployment (PaddlePaddle#2977) * Update benchmark tools (PaddlePaddle#3004) * update benchmark tools * update benchmark tools * update flake8 version to support pre-commit in python3.12 (PaddlePaddle#3000) * update flake8 version to support pre-commit in python3.12 * polish code * [Feature] multi source download (PaddlePaddle#3005) * multi-source download * multi-source download * huggingface download revision * requirement * style * add revision arg * test * pre-commit * Change default download * change requirements.txt * modify English Documentation * documentation * [GCU] Update to develop (PaddlePaddle#2988) * [Model] Provide clearer error for missing KV cache quantization scales (PaddlePaddle#3007) * [Feature] Support_eplb (PaddlePaddle#2997) * [Feature] support_eplb * [Feature] support_eplb * [Fix] fix mm ep * Update setup.py * [feat] add disable_chat_template in chat api as a substitute for previous raw_request (PaddlePaddle#3023) * [feat] add disable_chat_template in chat api as a substitute for previous raw_request * [fix] pre-commit code check --------- Co-authored-by: GoldPancake <56388518+Deleter-D@users.noreply.github.com> Co-authored-by: gaoziyuan <88373061+gzy19990617@users.noreply.github.com> Co-authored-by: Sunny-bot1 <68891411+Sunny-bot1@users.noreply.github.com> Co-authored-by: Ryan <zihaohuang@aliyun.com> Co-authored-by: lizhenyun01 <1500424927@qq.com> Co-authored-by: chen <103103266+ckl117@users.noreply.github.com> Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com> Co-authored-by: lizexu123 <39205361+lizexu123@users.noreply.github.com> Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com> Co-authored-by: freeliuzc <lzc842650834@gmail.com> Co-authored-by: Zero Rains <linjunlu@zerorains.top> Co-authored-by: zhink <33270771+zhink@users.noreply.github.com> Co-authored-by: chenjian <1435317881@qq.com> Co-authored-by: bukejiyu <52310069+bukejiyu@users.noreply.github.com> Co-authored-by: xiegegege <46314656+xiegegege@users.noreply.github.com> Co-authored-by: xiaoxiaohehe001 <49090790+xiaoxiaohehe001@users.noreply.github.com> Co-authored-by: YUNSHEN XIE <1084314248@qq.com> Co-authored-by: Yzc216 <101054010+Yzc216@users.noreply.github.com> Co-authored-by: ltd0924 <32387785+ltd0924@users.noreply.github.com> Co-authored-by: Zhang Yulong <35552275+ZhangYulongg@users.noreply.github.com> Co-authored-by: EnflameGCU <118410644+EnflameGCU@users.noreply.github.com> Co-authored-by: littledgg <61149469+littledgg@users.noreply.github.com> Co-authored-by: 李泳桦 <39643373+liyonghua0910@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
pcard-71500
当前VL和纯文本的执行流程已经合并到worker_process中，但合入时忽略了chunked_prefill场景下，纯文本与VL执行prefill的差异（VL每次只能有一个batch在做prefill，但纯文本可以有多个）。
本PR为执行流程添加了VL的限制，经过1000条数据的100并发测试，无报错内容且精度没问题。