22 Pull requests merged by 15 people

• [inference]Update doc

  #10278 merged Mar 28, 2025

• Add tokens_zip op to perform zip operation more efficiently, substituing unpermute

  #10300 merged Mar 28, 2025

• Implemented unzip_stable_op to fully substitute permute function

  #10288 merged Mar 28, 2025

• Update index.rst

  #10295 merged Mar 28, 2025

• Enhance RLHF

  #10123 merged Mar 28, 2025

• [Inference]migrate some moe file

  #10292 merged Mar 28, 2025

• [Serving] 支持配置version

  #10293 merged Mar 28, 2025

• Fp8 speed optmize

  #10272 merged Mar 28, 2025

• fix server

  #10286 merged Mar 27, 2025

• [Inference] Enhance fused_rotary_position_encoding and get_position_i…

  #10285 merged Mar 27, 2025

• 【Infer】MLA matrix absorption separation and fix

  #10283 merged Mar 27, 2025

• Fix typos in multiple files

  #10224 merged Mar 27, 2025

• Adding vectorized Fused dequant op

  #10276 merged Mar 27, 2025

• [fp8]zip and group gemm

  #10214 merged Mar 27, 2025

• add record_stream for dispatch and combine output

  #10270 merged Mar 27, 2025

• Add dynamic llama2-7b benchmark

  #10271 merged Mar 27, 2025

• [MLA] move compute_out_linear out and fix bug when q_lora_rank is None

  #10275 merged Mar 26, 2025

• 【Inference Optimize】support MOE EP Parallel 1.

  #10266 merged Mar 26, 2025

• 【Infer】MLA matrix absorption separation

  #10249 merged Mar 26, 2025

• optmize grouped gemm

  #10274 merged Mar 26, 2025

• Optimizing TokenDispatcherUtils ops' performance.

  #10262 merged Mar 26, 2025

• [LLM] fix openai client and stream output

  #10268 merged Mar 26, 2025

22 Pull requests opened by 17 people

• [LLM] fix openai client and stream output bug

  #10267 opened Mar 26, 2025

• Add record_stream for dispatch and combine output tensors

  #10269 opened Mar 26, 2025

• Dsv3 dev

  #10273 opened Mar 26, 2025

• [LLM INFER] Support dynamic graph server

  #10277 opened Mar 27, 2025

• [Deprecated] Delete few used model and tools.

  #10279 opened Mar 27, 2025

• Support Group Gemm Mask

  #10280 opened Mar 27, 2025

• Bf16 batch gemm dual gemm

  #10281 opened Mar 27, 2025

• [Unified Checkpoint] update moe

  #10282 opened Mar 27, 2025

• [INTEL_HPU] Enable Intel HPU fused multi transformer

  #10284 opened Mar 27, 2025

• Fix typos in multiple files

  #10289 opened Mar 28, 2025

• Fix typos

  #10290 opened Mar 28, 2025

• optimize memory init

  #10297 opened Mar 28, 2025

• Test manual fw

  #10298 opened Mar 28, 2025

• update pybind_H

  #10299 opened Mar 28, 2025

• [Inference]fix moe ops input name

  #10301 opened Mar 28, 2025

• support masked

  #10302 opened Mar 28, 2025

• [Tools] Add Script for Torch Parameters Conversion

  #10303 opened Mar 28, 2025

• support masked moe

  #10304 opened Mar 28, 2025

• Fix small bugs

  #10305 opened Mar 28, 2025

• Add debug tools

  #10306 opened Mar 28, 2025

• Fix oom

  #10307 opened Mar 28, 2025

• Adding probs recover functionality to tokens_zip_op

  #10308 opened Mar 29, 2025

1 Issue closed by 1 person

• [Question]: 如何更改模型加载路径

  #10294 closed Mar 28, 2025

2 Issues opened by 2 people

• [Question]: 静态图预置模型下载失败

  #10291 opened Mar 28, 2025

• [Question]: ernie-doc模型ernie-doc-base-en，ernie-doc-base-zh无法下载

  #10287 opened Mar 27, 2025

14 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.

• [generation]update generation

  #10080 commented on Mar 27, 2025 • 1 new comment

• PaddleNLP 快乐开源活动 (2025 H1) 🎉

  #9763 commented on Mar 27, 2025 • 0 new comments

• [Bug]: 服务化部署失败，ModuleNotFoundError: No module named 'predict'

  #10265 commented on Mar 27, 2025 • 0 new comments

• [Question]: pp-uie推理返回格式

  #10245 commented on Mar 27, 2025 • 0 new comments

• [Bug]: 训练后评估阶段报错

  #9908 commented on Mar 28, 2025 • 0 new comments

• [Tokenizer] Upgrade tokenizer

  #9683 commented on Mar 27, 2025 • 0 new comments

• [Feature] Sageattn write kv cache with BF16/FP16, INT8, FP8

  #10032 commented on Mar 27, 2025 • 0 new comments

• Test secrets

  #10087 commented on Mar 26, 2025 • 0 new comments

• [Inference] support deepseek_vl2

  #10182 commented on Mar 28, 2025 • 0 new comments

• 【Inference Optimize】Paddle supports MOE model EP parallel

  #10201 commented on Mar 28, 2025 • 0 new comments

• [DCU]support dcu PagedAttention prefix in compute_mla_absorb

  #10229 commented on Mar 28, 2025 • 0 new comments

• [WIP] Add deepep timer.

  #10232 commented on Mar 26, 2025 • 0 new comments

• 【PPO】support dataproto & fix dataflow

  #10259 commented on Mar 27, 2025 • 0 new comments

• [Docs] Move docs to docs/zh

  #10261 commented on Mar 28, 2025 • 0 new comments