4239 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
Divano	c1aa66df02	Revert "[Optim] Remove limitation of number of kvcache blocks (#5612)" (#5702) ... This reverts commit 9da89a374b.	2025-12-23 15:41:33 +08:00
Jiaxin Sui	0bef9b684f	[Metax][CI]fix CI bug (#5698) ... * Update run_ci_metax.sh * Fix pull request branch reference in CI workflow	2025-12-23 14:56:34 +08:00
RuohengMa	2c3c983b96	[XPU] modify speculate_verify (#5522)	2025-12-23 14:50:30 +08:00
MingkunZhang	945a1bc4e2	[Metax] update ci name (#5679) ... * [Metax] update ci name * Update CI_METAX workflow for pull request handling * Update ci_metax.yml * Update CI_METAX workflow for pull request handling * Remove commented-out code in run_ci_metax.sh * Add environment to Jenkins trigger job * Change trigger event from pull_request_target to pull_request * Fix environment name casing in CI workflow * Change environment name from Metax-ci to Metax_ci * Modify CI_METAX workflow for PR targeting and concurrency Updated workflow to use pull_request_target event and added concurrency settings. --------- Co-authored-by: Jiaxin Sui <95567040+plusNew001@users.noreply.github.com>	2025-12-23 14:00:48 +08:00
bukejiyu	6c36a17369	[Others]Prevent core dumps during Paddle version check (#5657)	2025-12-22 21:57:45 -08:00
Jiang-Jia-Jun	9da89a374b	[Optim] Remove limitation of number of kvcache blocks (#5612) ... * [Optim] Remove limitation of number of kvcache blocks * Update fastdeploy/envs.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update fastdeploy/worker/iluvatar_worker.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Add docs * Update fastdeploy/worker/worker_process.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * fix ci case --------- Co-authored-by: Jiang-Jia-Jun <jiangjiajun@baidu.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>	2025-12-23 11:18:29 +08:00
ddchenhao66	4a74f5ab9b	[XPU]Set top_p=0.0 by default on XPU to optimize performance (#5686) ... Co-authored-by: ddchenhao66 <dhaochen163.com>	2025-12-23 11:01:01 +08:00
xunyoyo	3aee5c4bf5	[CI] 【Hackathon 9th Sprint No.37】NO.37 功能模块单测补充 (#5059) ... * Add unit tests for TokenProcessor functionality * Add trace stubs for token processor tests * Increase token processor test coverage * Clean up imports in test_token_processor.py Remove unnecessary path manipulation in test file. * Cleanup: Remove unused imports in test_token_processor Removed unused imports from the test file. * Add trace_carrier to task in test cases Added trace_carrier attribute to task in multiple test cases to ensure proper handling of trace information. * Refine token processor tests for safe coverage * Expand postprocess coverage * Add ZMQ logprob parsing test --------- Co-authored-by: CSWYF3634076 <wangyafeng@baidu.com> Co-authored-by: Tao Luo <luotao02@baidu.com>	2025-12-23 10:35:16 +08:00
Jiaxin Sui	f16077a939	[XPU][CI] Xpu ci update (#5690) ... * Enhance run_ci_xpu.sh with caching and prefill options * Update model path and configuration in run_ci_xpu.sh * Add '北朝' keyword to assertion in run_45vl.py * Enhance process termination logic in run_ci_xpu.sh * Set timeout for CI_XPU job to 60 minutes * Remove extra newline in stop_processes function * Update paddlepaddle-xpu installation command Comment out the previous paddlepaddle-xpu installation command and replace it with a specific version installation due to EP parallel error. * Update PaddlePaddle installation command * Remove max_tokens from model response configuration Removed max_tokens parameter from the model response call.	2025-12-23 10:19:39 +08:00
xiaolei373	dfe8ea941c	[log]console log to llm log (#5680)	2025-12-23 10:05:45 +08:00
RAM	131defa122	Revert "Revert "[Feature] Use paddle.compat.enable_torch_proxy in `fastdepl…" (#5606) ... This reverts commit 021399f7c9.	2025-12-22 22:37:51 +08:00
ddchenhao66	a1535c7e7e	[XPU][CI] xpu add ci test for pd + TP2 (#5653) ... Co-authored-by: ddchenhao66 <dhaochen163.com>	2025-12-22 19:27:10 +08:00
Yuanle Liu	8beb0158fa	[BugFix] fix rl signal (#5681)	2025-12-22 00:35:54 -08:00
MingkunZhang	6ed9136a4e	[Metax] update ci yaml (#5674)	2025-12-22 16:00:25 +08:00
YuBaoku	b57deb671d	[CI] Update check_approval.sh	2025-12-22 15:52:04 +08:00
Sunny-bot1	04035e4ebf	support w4afp8 two stage (#5608)	2025-12-22 15:13:05 +08:00
Sunny-bot1	40f3897a4e	support w4afp8 moe offline permute & load (#5613)	2025-12-22 15:12:57 +08:00
ming1753	81384ef29e	[BugFix] fix download feature bug (#5669)	2025-12-22 13:46:39 +08:00
lizexu123	6d323769dd	fix w4afp8 (#5634)	2025-12-22 13:39:41 +08:00
freeliuzc	6eada4929d	[Speculative Decoding]Support multi-step mtp with cudagraph (#5624) ... * support multi-step mtp with cudagraph * fix usage * fix unit test	2025-12-22 11:34:04 +08:00
Yonghua Li	4f830aa505	[RL] provide options for whether shutdown comm group after weights cleared (#5663) ... * [rl] provide options for whether shutdown comm group after weights cleared * [fix] fix args hardcode * [fix] change args type * [fix] add worker process args	2025-12-19 07:06:48 -08:00
YuBaoku	fe55baae47	[CI] Fix unit_test error of unstable execution (#5660) ... * [CI] Fix unit_test error of unstable execution	2025-12-19 22:59:53 +08:00
chen	a32cb54d0b	[BugFix] Fix custom_all_reduce overflow (#5662) ... * check * check * code style	2025-12-19 18:24:21 +08:00
MingkunZhang	46d83be065	[Metax] update ci test (#5652)	2025-12-19 17:25:47 +08:00
xunyoyo	669dfe8dca	[CI] 【Hackathon 9th Sprint No.38】NO.38 功能模块单测补充 (#5060) ... * Add tests for global scheduler functionality This file contains tests for the global scheduler, including request handling, load accounting, and response management. * Enhance global scheduler unit tests * Clean up comments in test_global_scheduler.py Removed unnecessary comments and coverage instructions. * Update tests/scheduler/test_global_scheduler.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update tests/scheduler/test_global_scheduler.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update tests/scheduler/test_global_scheduler.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update tests/scheduler/test_global_scheduler.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Translate comments and docstrings to Chinese Updated comments and docstrings in Chinese to enhance clarity for native speakers. * Use monkeypatch for FD_ENABLE_MAX_PREFILL in tests Replaced direct assignment with monkeypatching for FD_ENABLE_MAX_PREFILL in test cases. * Correct test function definitions in test_global_scheduler Fix function definitions and update test cases for scheduler. * Fix typos in global scheduler tests * Translate comments to English in test_global_scheduler.py Updated comments and docstrings in test_global_scheduler.py from Chinese to English for better clarity and consistency. --------- Co-authored-by: CSWYF3634076 <wangyafeng@baidu.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Jiaxin Sui <95567040+plusNew001@users.noreply.github.com>	2025-12-19 16:28:16 +08:00
kevin	807e404369	[BugFix] fix eb5 mm prefix cache bug (#5638) ... * fix eb5 mm prefix cache bug * update code --------- Co-authored-by: Jiaxin Sui <95567040+plusNew001@users.noreply.github.com>	2025-12-19 14:57:37 +08:00
RichardWooSJTU	6bd772b93f	fix eplb weight updating (#5529) ... Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com>	2025-12-19 14:30:32 +08:00
Yuanle Liu	689f54f671	[RL] Update worker_process.py (#5651)	2025-12-18 20:07:58 -08:00
fmiao2372	a8fce47195	[Intel HPU] enable kv cache scheduler v1 for hpu (#5648) ... * [Intel HPU] enable kv cache scheduler v1 for hpu * fix copilt comments	2025-12-19 12:03:39 +08:00
bukejiyu	fc452c8e29	[RL]Fix RL load_weights (#5642) ... Co-authored-by: Jiaxin Sui <95567040+plusNew001@users.noreply.github.com>	2025-12-18 19:16:18 -08:00
lizan1999	ec6811f648	support token num = 0 (#5635) ... Co-authored-by: lizan1999 <lizan03@baidu.com> Co-authored-by: cmcamdy <1027740945@qq.com> Co-authored-by: Jiaxin Sui <95567040+plusNew001@users.noreply.github.com>	2025-12-19 10:20:38 +08:00
xunyoyo	d657455616	[CI] 【Hackathon 9th Sprint No.19】NO.19 功能模块单测补充 (#5063) ... * Add pooler unit tests * Refine pooler tests import handling * Refactor pooler tests to use real modules * Clean up test_pooler.py by removing docstring Removed unnecessary docstring and cleaned up code. * Clean up imports in test_pooler.py Removed unnecessary import of sys and related path adjustments. * Update model config and clean up test code * Update test_pooler.py --------- Co-authored-by: CSWYF3634076 <wangyafeng@baidu.com>	2025-12-18 21:32:44 +08:00
Yuanle Liu	b47674c796	[BugFix] fix rl model_weights_signal to support tp>1 (#5639)	2025-12-18 04:43:58 -08:00
Jiaxin Sui	d739af5e6e	Revert "[XPU][CI] xpu add ci test for pd (#5610)" (#5645) ... This reverts commit 80fb530ce2.	2025-12-18 19:59:09 +08:00
bukejiyu	4aa2c6871b	[RL]Support loading weights via the load_weights function for RL (#5549) ... * RL support load_weights * fix	2025-12-18 02:27:05 -08:00
yzwu	ac013803f3	[Iluvatar] Support V1_KVCACHE_SCHEDULER and paddleocr-vl rope mode (#5555)	2025-12-18 02:14:25 -08:00
Zhang Yulong	48f3e9797e	Update backend_request_func.py (#5633)	2025-12-18 16:21:34 +08:00
xunyoyo	2d2619d300	[CI] 【Hackathon 9th Sprint No.36】NO.36 功能模块单测补充 （修复） (#5609) ... * Implement unit tests for PrefixCacheManager * Update prefix cache manager tests * Handle get_all_visible_devices in prefix cache manager tests * Add repo root to prefix cache manager tests sys.path * Use pathlib for repo root in prefix cache manager tests * Refine repo root Path import in tests * Handle list-based visible device configuration * Refine PrefixCacheManager test stubs * Run pre-commit on prefix cache manager tests * Remove duplicate pytest import in cache manager tests * Add tests for visible device formatting * Revert * Simplify test stubs in prefix cache manager tests * Refine PrefixCacheManager tests * Adjust prefix cache manager tests per review * Remove ignored tests from coverage configuration * Make prefix cache manager tests runnable without paddle * Use real paddle import in prefix cache manager tests * Clean up imports in test_prefix_cache_manager.py Removed unnecessary import of 'os' and related path manipulation. * Update test_prefix_cache_manager.py * Replace pid_suffix with ipc_suffix in tests * Add local cache queue and RDMA ports to cache config	2025-12-18 16:08:42 +08:00
lizan1999	e1a9b282eb	fix bug for EP+MTP (#5605) ... Co-authored-by: lizan1999 <lizan03@baidu.com>	2025-12-18 14:34:54 +08:00
Longzhi Wang	d8587e987e	[Model] tp+ep support v1_loader (#5465) ... * [Model] tp+ep support v1_loader * fix * fix mtp_linear * fix mtp_linear * fix * fix * fix v0 loader * fix * Add get_tensor for ep * fix linear weight_loader * fix typo * fix	2025-12-18 14:31:54 +08:00
Zhang Yulong	c89a62e550	Update backend_request_func.py (#5631)	2025-12-18 14:20:17 +08:00
zhupengyang	8735cb5045	[XPU] refactor moe ffn (#5501) ... - remove BKCL_DISPATCH_ALL_GATHER - support sparse mode - support moe quant_method	2025-12-18 14:14:05 +08:00
MingkunZhang	d0a7834a17	[Metax] fix metax runner issue (#5629)	2025-12-17 21:32:54 -08:00
qw86972190	c606df59f5	[XPU]logprob bug (#5626)	2025-12-18 12:07:20 +08:00
kesmeey	d81341b9b3	[CI]【Hackathon 9th Sprint No.14】功能模块 fastdeploy/rl/rollout_model.py 单测补充 (#5552) ... * Add rollout model unit tests * test: update rl rollout_model tests * test: fix cache_type_branches unsupported platform case * test: fix rl rollout_model test indent * Delete tests/spec_decode/test_mtp_proposer.py * chore: format test_rollout_model * chore: translate rollout test comments to English * test: guard rollout_model import by disabling auto registry * chore: reorder imports in rl rollout test * test: isolate env for RL rollout tests * style: format rollout RL tests with black * update * test: remove RL rollout unit tests causing collection issues * test: add lightweight rollout_model RL unit tests * fix(coverage): filter test file paths and handle collection failures - Only extract real test file paths (tests/.../test_*.py) from pytest collect output - Filter out ERROR/collecting prefixes to prevent garbage in failed_tests.log - Add proper error handling for pytest collection failures - Exit early if no test files can be extracted - Preserve collection error output for debugging * update * style: fix code style issues in test_rollout_model.py - Remove unused 'os' import - Remove trailing blank lines --------- Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com>	2025-12-18 10:57:53 +08:00
Zhang Yulong	f45c131ddf	update (#5625)	2025-12-17 21:38:14 +08:00
Zhang Yulong	94be5ebdd1	[CI] Add CI case for MTP accept ratio (#5570) ... * Implement test for MTP accept ratio Add test for MTP accept ratio with assertions on results and metrics. * Update test_ernie_21b_mtp.py * Refactor test_mtp_accept_ratio for baseline comparison Refactor test_mtp_accept_ratio to compare results against baseline file and metrics. * Fix formatting issues in test_ernie_21b_mtp.py --------- Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com>	2025-12-17 21:35:02 +08:00
kesmeey	ac731653b3	[CI]【Hackathon 9th Sprint No.12】功能模块 fastdeploy/spec_decode/mtp.py 单测补充 (#5533) ... * Add unit tests for MTPProposer class in spec_decode/mtp.py * fix: remove non-existent QuantizationConfig import in test_mtp_proposer * fix: add logprobs_mode attribute to FakeModelConfig * fix: fix test failures in test_mtp_proposer - fix Mock setup, remove arrival_time, add missing keys * fix: add seq_lens_this_time initialization and kv_cache init before insert_tasks_v1 * fix: check pos_emb_type attribute existence before assertion * test: add minimal coverage for mtp cache type, mm init, preempted * test: fix cache_type_branches unsupported platform on 12 * test: refine MTPProposer tests for cache type, requests and chunked prefill * chore: remove stray spec_decode copy	2025-12-17 20:09:45 +08:00
megemini	111955ec0c	[BugFix] 移除重复的 PaddleOCRVLProcessor 初始化代码	2025-12-17 18:58:02 +08:00
fmiao2372	404cf0ece4	[Intel HPU] enable tensor_wise_fp8 (#5324) ... * [Intel HPU] enable tensor_wise_fp8 * update code based on comments * fix code style issue * fix bug about RP 5138 * mv kv_cache modifications to HPU backend * fix FP8 Precision Issues * fix FP8 Precision Issues * Add quantization UT --------- Co-authored-by: yanfeich <yanfei.cheng@intel.com> Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com>	2025-12-17 16:45:03 +08:00