Search results

GRPOConfig.max_prompt_length does not truncate prompts for a conversational dataset with vllm (only in the latest main branch)

🏋 GRPORelated to GRPORelated to GRPO

🐛 bugSomething isn't workingSomething isn't working

Status: Open.

#4358 In huggingface/trl;

· muupan opened on Oct 28, 2025

cuda oom when finetuning mllama with lora and sfttrainer

⚡ PEFTRelated to PEFTRelated to PEFT

🐛 bugSomething isn't workingSomething isn't working

🚀 deepspeedRelated to deepspeedRelated to deepspeed

Status: Open.

#4347 In huggingface/trl;

· osaidr opened on Oct 27, 2025

Issues exist with the settings for num generation and the number of model-generated responses (Completion). The number of Completion responses consistently equals per_device_batch_size multiplied by gradient_acc_steps.

❓ questionSeeking clarification or more informationSeeking clarification or more information

🐛 bugSomething isn't workingSomething isn't working

🚀 deepspeedRelated to deepspeedRelated to deepspeed

Status: Open.

#4346 In huggingface/trl;

· Haoyong23 opened on Oct 27, 2025

Refactoring out the logps and importance comptuation out of generate_and_score in GRPOTrainer

✨ enhancementNew feature or requestNew feature or request

🏋 GRPORelated to GRPORelated to GRPO

Status: Open.

#4343 In huggingface/trl;

· wirmius opened on Oct 27, 2025

The "report_to" parameter in GRPOConfig

❓ questionSeeking clarification or more informationSeeking clarification or more information

🏋 GRPORelated to GRPORelated to GRPO

📚 documentationImprovements or additions to documentationImprovements or additions to documentation

Status: Open.

#4341 In huggingface/trl;

· LWShowTime opened on Oct 26, 2025

Packing with VLMs

✨ enhancementNew feature or requestNew feature or request

🏋 SFTRelated to SFTRelated to SFT

Status: Open.

#4339 In huggingface/trl;

· jiosephlee opened on Oct 25, 2025

CI fails with dev dependencies: HFValidationError: Repo id must use alphanumeric chars

⚡ PEFTRelated to PEFTRelated to PEFT

🏋 RLOORelated to RLOORelated to RLOO

🏋 SFTRelated to SFTRelated to SFT

🐛 bugSomething isn't workingSomething isn't working

Status: Open.

#4323 In huggingface/trl;

· albertvillanova opened on Oct 22, 2025

Replace flash-attn with kernels in CI

✨ enhancementNew feature or requestNew feature or request

Status: Open.

#4319 In huggingface/trl;

· albertvillanova opened on Oct 22, 2025

torch.utils.checkpoint.CheckpointError when using LoRA + fsdp2 Context Parallel

⚡ PEFTRelated to PEFTRelated to PEFT

🏋 RLOORelated to RLOORelated to RLOO

Status: Open.

#4314 In huggingface/trl;

· zxgx opened on Oct 21, 2025

Turn off Qwen3's reasoning mode

✨ enhancementNew feature or requestNew feature or request

🏋 GRPORelated to GRPORelated to GRPO

Status: Open.

#4311 In huggingface/trl;

· Tuziking opened on Oct 20, 2025

Reference model for LoRA

⚡ PEFTRelated to PEFTRelated to PEFT

❓ questionSeeking clarification or more informationSeeking clarification or more information

Status: Open.

#4309 In huggingface/trl;

· SpaceHunterInf opened on Oct 20, 2025

CI fails for Python 3.9: ValueError: zero-size array to reduction operation maximum which has no identity

🏋 Online DPORelated to Online DPORelated to Online DPO

🐛 bugSomething isn't workingSomething isn't working

Status: Open.

#4308 In huggingface/trl;

· albertvillanova opened on Oct 20, 2025