8 Pull requests merged by 7 people

• support bw split

  #10915 merged Aug 9, 2025

• Support fp8 weight quant cache

  #10914 merged Aug 7, 2025

• Support magic send for mtp

  #10916 merged Aug 7, 2025

• Update ppo

  #10912 merged Aug 7, 2025

• Fix mtp bug when send_mtp_embed=True

  #10909 merged Aug 7, 2025

• fix ordered_save func

  #10896 merged Aug 5, 2025

• token_dispatcher support expert_num 64

  #10905 merged Aug 5, 2025

• Fix 0 size bug

  #10906 merged Aug 5, 2025

5 Pull requests opened by 4 people

• Revert dsv3_dev to runnable version

  #10907 opened Aug 5, 2025

• Add vapo

  #10908 opened Aug 5, 2025

• Update vapo

  #10918 opened Aug 8, 2025

• [Auto-parallel] Improve usability of auto_dy pipeline parallel

  #10920 opened Aug 8, 2025

• fp8_quant_cache_and_using_fp8_gemm

  #10923 opened Aug 10, 2025

3 Issues closed by 2 people

• [Question]: 模型不能从本地目录加载吗？

  #10917 closed Aug 8, 2025

• [Question]:

  #10635 closed Aug 5, 2025

• [Bug]: paddle3.0如何导出2.x一样的静态模型

  #10630 closed Aug 4, 2025

5 Issues opened by 5 people

• [Question]: No module named 'paddle.distributed.auto_parallel.intermediate'

  #10922 opened Aug 10, 2025

• [Question]: 飞腾+昆仑R200,飞浆官网没有提供paddlepaddle3.0的昆仑芯 XPU 的 ARM 架构安装包

  #10921 opened Aug 9, 2025

• [Question]: 关于开启stage3后，checkpoint保存失败的问题

  #10919 opened Aug 8, 2025

• [Question]:尝试推理模型提示Failed to initialize paddl enlp_ops.so: undefined symbol: _ZN5baidu3xpu3api6plugin15rebuild_paddingI8bfloat16EEiPNS1_7ContextEPT_PKS7_PKiSC_SC_iii

  #10913 opened Aug 6, 2025

• [Bug]: 单机运行DeepSeek_R1动态图 + MTP 动态图出现no attribute 'output_via_mq'

  #10911 opened Aug 6, 2025

55 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.

• [Bug]: error_msgs = _load_state_dict_into_model(model, state_dict, "") TypeError: _load_state_dict_into_model() missing 1 required positional argument: 'model_to_load_state_dict'

  #10690 commented on Aug 4, 2025 • 0 new comments

• [Bug]: Deepseekr1 推理 接口报错

  #10894 commented on Aug 4, 2025 • 0 new comments

• [Bug]: 在用Taskflow推理的时候，指定本地模型路径没生效

  #10660 commented on Aug 8, 2025 • 0 new comments

• [Question]: pp-uie使用taskflow无法加载自定义路径模型，只能加载默认路径：C:\Users\.paddlenlp\models\paddlenlp/PP-UIE-0.5B

  #10409 commented on Aug 9, 2025 • 0 new comments

• [Bug]: 安装完paddlenlp_ops运行不起来

  #10719 commented on Aug 10, 2025 • 0 new comments

• Add model VisualBERT

  #1283 commented on Aug 4, 2025 • 0 new comments

• Add byt5 Model

  #1742 commented on Aug 4, 2025 • 0 new comments

• Add question generation example

  #2944 commented on Aug 4, 2025 • 0 new comments

• Refactor training loop

  #6098 commented on Aug 4, 2025 • 0 new comments

• Moe

  #6515 commented on Aug 4, 2025 • 0 new comments

• [DO NOT MERGE][TaskFlow]Support llama & bloom taskflow

  #6586 commented on Aug 4, 2025 • 0 new comments

• [NPU] adaptation for LLaMA

  #7262 commented on Aug 4, 2025 • 0 new comments

• [AutoParallel] Test 3d SP acc

  #7677 commented on Aug 4, 2025 • 0 new comments

• [DO NOT Merge] Test dynamic auto parallel 3d sp acc

  #7683 commented on Aug 4, 2025 • 0 new comments

• vocabulary_expansion_pretraining

  #7755 commented on Aug 4, 2025 • 0 new comments

• [DO NOT MERGE] Dynamic Auto parallel Performance Test.

  #7804 commented on Aug 4, 2025 • 0 new comments

• ceval_quant_eval

  #8220 commented on Aug 7, 2025 • 0 new comments

• xxx. fix_ceval_quant_eval

  #8221 commented on Aug 6, 2025 • 0 new comments

• support Baichuan13B inference

  #8287 commented on Aug 6, 2025 • 0 new comments

• fix bug for gpt args

  #8530 commented on Aug 4, 2025 • 0 new comments

• Allow to pre alloc memory for pretraining for better memory use.

  #8600 commented on Aug 4, 2025 • 0 new comments

• [Trainer] Add metrics dumper in background

  #9112 commented on Aug 4, 2025 • 0 new comments

• [DON'T NEED REVIEW] Mthreads llama 13 b 128 pp16

  #9193 commented on Aug 5, 2025 • 0 new comments

• Sci/benchmark iluvatar

  #9369 commented on Aug 5, 2025 • 0 new comments

• [DON'T NEED REVIEW] Mthreads llama 13 b 64 pp8

  #9557 commented on Aug 4, 2025 • 0 new comments

• [DON'T NEED REVIEW] Mthreads llama 13 b 64 pp16

  #9558 commented on Aug 4, 2025 • 0 new comments

• [LLM] Add fused attention in Qwen2MoE

  #9767 commented on Aug 8, 2025 • 0 new comments

• Support deepseek v3

  #9835 commented on Aug 4, 2025 • 0 new comments

• [Feature] Sageattn write 8 bit kv-cache

  #10032 commented on Aug 8, 2025 • 0 new comments

• 2:4 sparse for int8/fp8/bf16/fp16 gemm

  #10081 commented on Aug 4, 2025 • 0 new comments

• [custom device]: replace broadcast kernel with custom api for sdaa.

  #10082 commented on Aug 4, 2025 • 0 new comments

• Add fused_topk_to_multihot / fused_multihot_prob_backto_topk(grad) custom op to prevent CPU stall.

  #10125 commented on Aug 4, 2025 • 0 new comments

• [LLM] fix openai client and stream output bug

  #10267 commented on Aug 7, 2025 • 0 new comments

• Add record_stream for dispatch and combine output tensors

  #10269 commented on Aug 4, 2025 • 0 new comments

• Dsv3 dev

  #10273 commented on Aug 9, 2025 • 0 new comments

• Bf16 batch gemm dual gemm

  #10281 commented on Aug 8, 2025 • 0 new comments

• update pybind_H

  #10299 commented on Aug 10, 2025 • 0 new comments

• Deepseek xpu

  #10340 commented on Aug 4, 2025 • 0 new comments

• add_pybind for dyGraph predictor running

  #10374 commented on Aug 10, 2025 • 0 new comments

• int8 train test

  #10451 commented on Aug 9, 2025 • 0 new comments

• add append_attn ut

  #10527 commented on Aug 6, 2025 • 0 new comments

• Add llama-13b dynamic auto benchmark

  #10654 commented on Aug 5, 2025 • 0 new comments

• 【Hackathon 8th No.30】在 PaddleNLP 中复现 Gemma2 模型

  #10684 commented on Aug 4, 2025 • 0 new comments

• 【Hackathon 8th No.29】在 PaddleNLP 中复现 ModernBERT 模型

  #10686 commented on Aug 4, 2025 • 0 new comments

• 【Hackathon 8th No.28】在 PaddleNLP 中复现 Phi3

  #10688 commented on Aug 9, 2025 • 0 new comments

• Tmp llama13b benchmark

  #10694 commented on Aug 10, 2025 • 0 new comments

• 【Hackathon 8th No.31】在 PaddleNLP 中复现 Apollo 精调算法

  #10703 commented on Aug 4, 2025 • 0 new comments

• 【Hackathon 8th No.32】在 PaddleNLP 中复现 Adam-mini 精调算法

  #10704 commented on Aug 4, 2025 • 0 new comments

• Support wint2 unzip

  #10706 commented on Aug 5, 2025 • 0 new comments

• [Auto-Parallel] Add benchmark for fast_rms_norm in llama13b N4C32 dy_auto

  #10713 commented on Aug 9, 2025 • 0 new comments

• Add forward unittest for fused_transpose_split_quant

  #10717 commented on Aug 10, 2025 • 0 new comments

• [Auto Parallel] Align the meaning of configuration items, decouple auto-parallel pipeline networking and trainer

  #10775 commented on Aug 6, 2025 • 0 new comments

• [CI]add workflow for llm&unittest-gpu

  #10878 commented on Aug 7, 2025 • 0 new comments

• add ppo

  #10884 commented on Aug 6, 2025 • 0 new comments

• implement of DISCO

  #10904 commented on Aug 4, 2025 • 0 new comments