1 Release published by 1 person

• rl-v1.0.0 Stable RL v1.0.0

  published May 21, 2025

21 Pull requests merged by 14 people

• [FIX] fix the bug that some gemm config did not be handled

  #10627 merged May 22, 2025

• Add sp_async_reduce_scatter in dygraph auto mode

  #10620 merged May 22, 2025

• update DeepGEMM

  #10429 merged May 21, 2025

• [Typos] Fix

  #10486 merged May 21, 2025

• 【PaddleNLP No.5】Update simcse to apply PIR

  #10396 merged May 21, 2025

• [Taskflow] Fix the recognition bug of json format with both PIR suffix and id2label

  #10487 merged May 21, 2025

• [PIR] Update paddle.inference infer example for Ernie-vil2.0

  #10500 merged May 21, 2025

• 【PaddleNLP No.1】 add pretrain.md

  #10506 merged May 21, 2025

• Fix typo get_dispached_metadata

  #10533 merged May 21, 2025

• [infer]support output length truncated

  #10614 merged May 20, 2025

• Fix fp8 in dtype_byte_size

  #10618 merged May 20, 2025

• [LLM] fix sharding v1 reshard

  #10610 merged May 19, 2025

• [LLM] support disable monkey patch

  #10617 merged May 19, 2025

• [INTEL_HPU] Enable FusedBlockMultiTransformerHPU

  #10514 merged May 19, 2025

• Update ppo_model_utils.py

  #10593 merged May 19, 2025

• [RL] Add reshard controller

  #10606 merged May 19, 2025

• Fix a small bug of sharding reshard

  #10613 merged May 19, 2025

• [LLM] fix normal save of rng states

  #10609 merged May 18, 2025

• Add offload opt util

  #10608 merged May 17, 2025

• Add offload optimizer function

  #10607 merged May 17, 2025

• [RL] logprob compute use the same method

  #10596 merged May 15, 2025

20 Pull requests opened by 13 people

• [Auto-Parallel] Add ci for tensor_fusion and overlap in auto-dy

  #10598 opened May 15, 2025

• Add benchmark for tensor_fusion and overlap in llama7b N4C32 dy_auto

  #10599 opened May 16, 2025

• Add fused_transpose_quant op

  #10601 opened May 16, 2025

• Test the improve in hand_dy llama7b tensor_fusion and sharding overlap

  #10602 opened May 16, 2025

• Adding fused_swiglu_probs_bwd op

  #10604 opened May 16, 2025

• [Docs] Add llm pretrain tutorial

  #10611 opened May 18, 2025

• [llm] support tensorwise fp8/int8 training

  #10612 opened May 19, 2025

• [RL] reinforce learning benchmark framework

  #10619 opened May 19, 2025

• run llama13b custom op perf

  #10622 opened May 20, 2025

• Add arbitrary padding to Unzip to handle extreme inbalance case.

  #10623 opened May 20, 2025

• Remove baichuan13b hidden layers in dynamic auto mode

  #10624 opened May 21, 2025

• 【PaddleNLP No.2】Add finetune tutorial

  #10625 opened May 21, 2025

• llama 自动并行 pp组网

  #10626 opened May 21, 2025

• [infer] update_input_v2 op will not return eos when env_var is set

  #10628 opened May 21, 2025

• add merge_all_horizontal_groups cinn flag for llama

  #10629 opened May 21, 2025

• 【PaddleNLP No.3】Add alignment tutorial

  #10631 opened May 21, 2025

• 【Inference optimize】fix config bug

  #10632 opened May 22, 2025

• [Template] add unified template.

  #10633 opened May 22, 2025

• LLaMa vpp 组网

  #10634 opened May 22, 2025

• 自动并行 LLaMa组网 with DP

  #10636 opened May 22, 2025

4 Issues opened by 3 people

• [Question]:

  #10635 opened May 22, 2025

• [Bug]: paddle3.0如何导出2.x一样的静态模型

  #10630 opened May 21, 2025

• [Question]: 使用910B训练好ernie3的分类模型后，模型无法进行推理

  #10621 opened May 20, 2025

• [Question]: 请问如何使用paddlenlp微调Llama-3.3-70B-Instruct的模型？

  #10600 opened May 16, 2025

51 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.

• [Integration] add swanlab logger

  #10594 commented on May 16, 2025 • 4 new comments

• [Bug]: UIE信息抽取预测时，请求多次，每次返回结果不固定

  #10150 commented on May 17, 2025 • 0 new comments

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

  #9193 commented on May 22, 2025 • 0 new comments

• [NPU] Add chatglmv3-6b

  #9213 commented on May 19, 2025 • 0 new comments

• [MLU] taskflow mlu

  #9635 commented on May 18, 2025 • 0 new comments

• [LLM] Add Aquila2 model

  #9692 commented on May 19, 2025 • 0 new comments

• Add galvatron search engine & hardware bandwidth profile for paddle.

  #9731 commented on May 18, 2025 • 0 new comments

• [Infer] Add the custom_ops for paddlenlp

  #9794 commented on May 18, 2025 • 0 new comments

• mla's todo

  #10096 commented on May 21, 2025 • 0 new comments

• Split predictor.

  #10136 commented on May 18, 2025 • 0 new comments

• Added regroup padded op to perform token regroup using 1D expert_idx with max-token-per-expert padding

  #10164 commented on May 18, 2025 • 0 new comments

• Add custom op for Tokens zip and unzip, in preparation of using groupgemm

  #10169 commented on May 18, 2025 • 0 new comments

• [Infer] Change groupwise weight quant from cpu to gpu for deepseek_v2 model

  #10174 commented on May 16, 2025 • 0 new comments

• [LLM] remove distutils in pdc sdk

  #10176 commented on May 19, 2025 • 0 new comments

• [LLM] remove distutils in pdc sdk

  #10179 commented on May 18, 2025 • 0 new comments

• optimize expert memory

  #10215 commented on May 20, 2025 • 0 new comments

• Add fp8 support for regroup op.

  #10231 commented on May 20, 2025 • 0 new comments

• [DCU]add op reshape_and_cache for dcu paged_attention

  #10233 commented on May 21, 2025 • 0 new comments

• MTP Linear support fp8 gemm

  #10244 commented on May 21, 2025 • 0 new comments

• Dsv3 dev

  #10273 commented on May 21, 2025 • 0 new comments

• [Feature] Sage Attention Support Varlen & write kv cache BF16/FP16

  #10360 commented on May 21, 2025 • 0 new comments

• [Typos] Fix

  #10494 commented on May 21, 2025 • 0 new comments

• optimize a8w8

  #10516 commented on May 15, 2025 • 0 new comments

• [Inference] Add new wint2.75/wint2.5 quant type and support DeepseekV3

  #10578 commented on May 22, 2025 • 0 new comments

• support_dynamic_server for wintx

  #10589 commented on May 16, 2025 • 0 new comments

• Integrate DataProto into the GRPO

  #10597 commented on May 22, 2025 • 0 new comments

• [Question]: Error loading layoutlmv2-base-uncased: Missing model_state.pdparams file

  #9868 commented on May 17, 2025 • 0 new comments

• [Question]: doccano=1.6.2，autolabeling无法实现，一直转圈

  #9744 commented on May 17, 2025 • 0 new comments

• [Question]: paddlenlp 3.0.0b1版本使用uie-m-base报错找不到static/inference.pdmodel

  #9646 commented on May 17, 2025 • 0 new comments

• [Question]: ernie-3-tiny 按文档步骤运行例子报错

  #9896 commented on May 18, 2025 • 0 new comments

• [Question]: LogitsProcessorList 缺少__iter__和extend方法

  #9926 commented on May 19, 2025 • 0 new comments

• [Question]: Will open dataset for ppuie model

  #10556 commented on May 19, 2025 • 0 new comments

• [Bug]: paddle_ops cpu 编译报错

  #10199 commented on May 20, 2025 • 0 new comments

• [Bug]: taskflow uie 动转静报错

  #10159 commented on May 21, 2025 • 0 new comments

• [Question]:多个预训练模型无法下载

  #10042 commented on May 21, 2025 • 0 new comments

• [Question]: 我这边想部署个离线的微型nlp模型到我的exe中

  #10238 commented on May 22, 2025 • 0 new comments

• [Bug]: BERT蒸馏至BiLSTM的数据增强是不是有问题

  #9816 commented on May 22, 2025 • 0 new comments

• PaddleNLP 快乐开源活动 (2025 H1) 🎉

  #9763 commented on May 22, 2025 • 0 new comments

• Dygraph auto

  #7609 commented on May 22, 2025 • 0 new comments

• MP2-PP2 hack shared layer to non-sharded layer to Step Alignment

  #7614 commented on May 22, 2025 • 0 new comments

• [AutoParallel] Test 3d SP acc

  #7677 commented on May 22, 2025 • 0 new comments

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

  #7683 commented on May 22, 2025 • 0 new comments

• [WIP] Implement fast_rms_norm as custom operators.

  #8073 commented on May 22, 2025 • 0 new comments

• ceval_quant_eval

  #8220 commented on May 22, 2025 • 0 new comments

• xxx. fix_ceval_quant_eval

  #8221 commented on May 22, 2025 • 0 new comments

• [AutoParallel] Fit job schedule profiler tool for to static mode

  #8488 commented on May 17, 2025 • 0 new comments

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

  #8600 commented on May 22, 2025 • 0 new comments

• support Llama2 fp8c16 inference

  #8762 commented on May 22, 2025 • 0 new comments

• support qwen2 inferece model

  #8806 commented on May 22, 2025 • 0 new comments

• add fp8 cutlass gemm fused op for llama2 inference

  #8809 commented on May 22, 2025 • 0 new comments

• support llama2 w8a8 fp8 inference

  #8827 commented on May 22, 2025 • 0 new comments