71 Pull requests merged by 28 people

• [Executor] Fix set capture sizes bug

  #2903 merged Jul 18, 2025

• [Trace]fix opentelemetry can not work in uvicorn

  #2906 merged Jul 17, 2025

• [Trace]fix opentelemetry can not work in uvicorn

  #2907 merged Jul 17, 2025

• [Executor] Updated the documents related to cuda graph and static graph.

  #2899 merged Jul 17, 2025

• [Executor] Updated the documents related to cuda graph and static graph.

  #2898 merged Jul 17, 2025

• [LLM] delete fixed slot

  #2894 merged Jul 17, 2025

• [XPU][doc] pick xpu doc fix

  #2897 merged Jul 17, 2025

• [LLM] delete fixed slots

  #2893 merged Jul 17, 2025

• [XPU][doc] fix typo

  #2892 merged Jul 17, 2025

• [Inference, rename] remove padding_offsets from atten use batch_id_per_token

  #2880 merged Jul 17, 2025

• [Feature][MTP] Support cacheKV transfer in per_chunk mode

  #2890 merged Jul 17, 2025

• [Bug Fix] fix bug of prompt penalty

  #2888 merged Jul 17, 2025

• [Fix] remove misleading variables

  #2841 merged Jul 17, 2025

• enable CI workflow for pull requests targeting release/* branches

  #2886 merged Jul 17, 2025

• Enable CI workflow for pull requests targeting release/* branches

  #2887 merged Jul 17, 2025

• [Bug Fix]Fix config get

  #2883 merged Jul 17, 2025

• [LLM] Add parameter validation and exception throwing

  #2878 merged Jul 17, 2025

• [MM_PROCESS] add _extract_labels

  #2879 merged Jul 17, 2025

• Fix rollout_model init

  #2882 merged Jul 17, 2025

• Fix rollout_model init

  #2881 merged Jul 17, 2025

• [Feature] support prompt repetition_penalty

  #2806 merged Jul 17, 2025

• [XPU][doc] Update minimal fastdeploy required

  #2863 merged Jul 17, 2025

• [Feature] Add speculative decoding metrics.

  #2857 merged Jul 17, 2025

• [Trace] fix annotation when add opentelemetry

  #2869 merged Jul 17, 2025

• [Bug Fix] fix opentelemetry-bootstra

  #2875 merged Jul 16, 2025

• fix put opentelemetry-instrumentation-fastapi in requierment

  #2874 merged Jul 16, 2025

• [Sync Code] develop to release/2.0.3

  #2873 merged Jul 16, 2025

• [LLM] Update Multinode Deployment

  #2830 merged Jul 16, 2025

• [LLM] support send batch data and aggregate data

  #2860 merged Jul 16, 2025

• fix and refine vl

  #2866 merged Jul 16, 2025

• [Attention] remove cum_offsets from atten, and use cu_seqlens_q

  #2870 merged Jul 16, 2025

• [Trace] Support trace log

  #2864 merged Jul 16, 2025

• [Trace] add opentelemetry

  #2852 merged Jul 16, 2025

• rl update

  #2861 merged Jul 16, 2025

• [Fix] Fix FLAGS_max_partition_size

  #2854 merged Jul 16, 2025

• [Fix] Fix mm ep weight init.

  #2855 merged Jul 16, 2025

• [BugFxi]Fix Config

  #2858 merged Jul 16, 2025

• [Fix] Fix expert parallel config bug

  #2848 merged Jul 16, 2025

• [XPU] Update doc and add scripts for downloading dependencies

  #2845 merged Jul 16, 2025

• [BugFix]Fix Configs

  #2849 merged Jul 16, 2025

• [Executor] CUDA Graph support padding batch

  #2844 merged Jul 16, 2025

• refactor rl get_name_mappings_to_training

  #2847 merged Jul 15, 2025

• Merge vl execution path into normal execution path

  #2829 merged Jul 15, 2025

• [Docs] add enable_logprob parameter description

  #2850 merged Jul 15, 2025

• 适配vLLM框架

  #2851 merged Jul 15, 2025

• 【Inference Optimize】Support wint2 triton kernel about triton_utils_v2

  #2842 merged Jul 15, 2025

• [vl] Use top_k from config.json.

  #2831 merged Jul 14, 2025

• [Perf][MTP] Improve speculative decoding(MTP) efficiency

  #2840 merged Jul 14, 2025

• Simplify the Config code

  #2770 merged Jul 14, 2025

• 【Cherry Pick】CP PR#2820 to release/2.0.2

  #2839 merged Jul 14, 2025

• [Fix]fix 'force-reinstall all-depe-packages in build'

  #2837 merged Jul 14, 2025

• 【Update Docs】update supported_models doc

  #2836 merged Jul 14, 2025

• [Feature] Add DeepGEMM pre-compile tools.

  #2819 merged Jul 14, 2025

• [features][MTP]Support expert-parellel mode with MTP

  #2835 merged Jul 14, 2025

• fix spelling error

  #2826 merged Jul 14, 2025

• [Bug Fix] fix spelling error

  #2827 merged Jul 14, 2025

• [vl]remove duplicated load logic

  #2744 merged Jul 12, 2025

• [CI] add result save for ci

  #2824 merged Jul 12, 2025

• Feature/logprob bug fix

  #2817 merged Jul 12, 2025

• [Bug fix]fix num_blocks_local when small size model in TP2 running mode

  #2792 merged Jul 12, 2025

• [Feature]Support Qwen3-moe name_mapping

  #2820 merged Jul 12, 2025

• [FIX]fix topp default value

  #2814 merged Jul 11, 2025

• After enabling "top_logprobs supports passing 0 and fix max_completion_tokens", an incorrect finish_reason was returned.

  #2815 merged Jul 11, 2025

• [FIX 2.0.2]fix topp topk default value

  #2810 merged Jul 11, 2025

• [Feature] Support tensor-parallel-size>num_key_value_heads for qwen3

  #2799 merged Jul 11, 2025

• [BugFix] Fix rl config enable_logprob

  #2811 merged Jul 11, 2025

• [Feature] Global scheduler supports configuring hot updates

  #2812 merged Jul 11, 2025

• Global scheduler supports configuring hot updates

  #2807 merged Jul 11, 2025

• [XPU] Update docker file

  #2809 merged Jul 11, 2025

• fix enable_logprob not in rl_config

  #2808 merged Jul 11, 2025

• Delete Useless Files

  #2772 merged Jul 11, 2025

15 Pull requests opened by 12 people

• 【Fix】sync paddlenlp_ops build use_fast_math

  #2821 opened Jul 12, 2025

• speedup load

  #2859 opened Jul 16, 2025

• Unify server-side and model-side Config

  #2862 opened Jul 16, 2025

• [Feature] support min_p_sampling

  #2872 opened Jul 16, 2025

• [Iluvatar GPU] Add CI scripts

  #2876 opened Jul 17, 2025

• QA Test Use

  #2877 opened Jul 17, 2025

• [feat] adapt metax-gpu maca platform

  #2884 opened Jul 17, 2025

• [Fix]fix empty prompt_token_ids,update the parser's triggering condit…

  #2891 opened Jul 17, 2025

• [BugFix] Deepseek renaming cum_offsets to cu_seqlens_q

  #2895 opened Jul 17, 2025

• Enable ci release

  #2896 opened Jul 17, 2025

• 【Feature】support vl model ori_vacab_size parmameters

  #2900 opened Jul 17, 2025

• 【DCU】update top_p_sampling

  #2901 opened Jul 17, 2025

• [Executor] Fix set capture sizes bug

  #2902 opened Jul 17, 2025

• [BugFix]Fix sample rejection

  #2908 opened Jul 18, 2025

• [NEW] Support 45tVL EP FP8 Infer.

  #2909 opened Jul 18, 2025

6 Issues closed by 5 people

• 编译报错 _Z20MoEDeepGEMMDePermuteRKN6paddle6TensorES2_S2_S2_

  #2853 closed Jul 17, 2025

• piece id is out of range.

  #2843 closed Jul 15, 2025

• ERNIE-4.5-VL-28B-A3B-Paddle 加载卡主不动，无论是单卡4090 48b还是双卡4090 48g都不行

  #2739 closed Jul 14, 2025

• 需要fastdeploy1.0.7源码

  #2781 closed Jul 11, 2025

• 使用官网镜像+操作步骤启动报错

  #2651 closed Jul 11, 2025

• ERNIE-4.5-300B-A47B-2Bits-Paddle 双卡部署报错

  #2678 closed Jul 11, 2025

4 Issues opened by 3 people

• RuntimeError: [MoE Runner] Failed to run cutlass variable batched gemm. Error: Error Internal

  #2889 opened Jul 17, 2025

• 遇到报错：ValueError: Tokenizer class Ernie4_5_VLTokenizer does not exist or is not currently imported.

  #2846 opened Jul 15, 2025

• OrtBackend + CUDA 12.X (onnxruntime >1.17) 无法兼容fastdeploy1.1.0

  #2833 opened Jul 14, 2025

• Convolutional Coded Quantization

  #2828 opened Jul 13, 2025

11 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• [Feature] mm and thinking model support structred output

  #2749 commented on Jul 17, 2025 • 2 new comments

• [xpu] Error:solver cache filexpu3_fwd_solver.json not found

  #2803 commented on Jul 11, 2025 • 0 new comments

• 使用FastDeploy部署PPOCRv3模型存在精度问题

  #1121 commented on Jul 11, 2025 • 0 new comments

• [bug] 'CompletionOutput' object has no attribute 'tool_call_content'

  #2802 commented on Jul 13, 2025 • 0 new comments

• 官方有开发计划表么，何时发布2.0的C++库？

  #2626 commented on Jul 13, 2025 • 0 new comments

• fastdeploy 部署erine-21B

  #2704 commented on Jul 18, 2025 • 0 new comments

• [WIP] optimzie wint2 moe_group_gemm.

  #2661 commented on Jul 17, 2025 • 0 new comments

• Support use safetensors with paddle.MmapStorage to load model files

  #2730 commented on Jul 13, 2025 • 0 new comments

• Opt wint2

  #2741 commented on Jul 15, 2025 • 0 new comments

• [Feature] support c16 prefix_cache in flash_attention_v3

  #2766 commented on Jul 16, 2025 • 0 new comments

• [SOT] Mark dynamic dims by type annotations

  #2771 commented on Jul 17, 2025 • 0 new comments