trl has 4M monthly downloads and 19K GitHub stars. Version 1.7.1 supports Python 3.10+, first published in 2020.
It is actively maintained under a permissive license, last released in Jul 2026.
It is built on top of accelerate, datasets and jinja2. It is used by 172 tracked packages, including unsloth, unsloth-zoo and llamafactory.
Package Insights
((week_daily_avg - month_daily_avg) / month_daily_avg) * 100Weekly Downloads
GitHub Stars
Downloads by OS
Python Versions
Top Countries
Dependencies
- accelerate >=1.4.0
- datasets >=4.7.0
- jinja2
- packaging >20.0
- transformers >=4.56.2
47 optional dependencies
- aiohttp[vllm]
- bitsandbytes[quantization]
- bitsandbytes[dev]
- deepspeed[dev]
- deepspeed[deepspeed]
- fastapi[vllm]
- harbor[harbor]
- hf-doc-builder[quality]
- hf-doc-builder[dev]
- jmespath[dev]
- joblib[dev]
- joblib[bco]
- liger-kernel[dev]
- liger-kernel[liger]
- math-verify[math-verify]
- num2words[vlm]
- num2words[dev]
- openreward[dev]
- openreward[openreward]
- peft[dev]
- peft[peft]
- pillow[vlm]
- pillow[dev]
- pre-commit[dev]
- pre-commit[quality]
- pydantic[vllm]
- pytest[test]
- pytest[dev]
- pytest-cov[test]
- pytest-cov[dev]
- pytest-datadir[test]
- pytest-datadir[dev]
- pytest-rerunfailures[dev]
- pytest-rerunfailures[test]
- pytest-xdist[dev]
- pytest-xdist[test]
- requests[vllm]
- scikit-learn[dev]
- scikit-learn[bco]
- scikit-learn[scikit]
- torchvision[dev]
- torchvision[vlm]
- transformers[dev]
- transformers[deepspeed]
- transformers[kernels]
- uvicorn[vllm]
- vllm[vllm]
Used By
- unsloth
- unsloth-zoo
- llamafactory
- ms-swift
- wisent
- animica
- autotrain-advanced
- axolotl
- instructlab-training
- torchstudio
- fai-rl
- shadowlm
- instructlab
- text2text
- aitraining
- pruna-pro
- factory-sdk
- oumi
- modelforge-finetuning
- arbor-ai
- pruna
- backpropagate
- hpsv3
- finetune-cli
- fedops
- kailash-align
- superduper-transformers
- llmtune-local
- llm-forge-new
- llmtuner
- textrl
- modelco
- compressgpt-core
- konic
- erl-trainer
- document-language-model
- forgelm
- nexuss-transformer
- coreason-model-foundry
- questionnaire-mistral
- lfm-trainer
- geniusrise-vision
- fms-hf-tuning
- huggify-data
- roboreason
- engram-peft
- asft
- tunelabs-zoo
- geniusrise-text
- lean-dojo-v2