training & fine-tuning: 16 curated resources
training & fine-tuning
16 resources · page 1 of 1
LLaMA-Factory
by Yaowei Zheng
LLaMA-Factory GitHub repo: unified framework for fine-tuning over 100 LLMs and VLMs.
hiyouga/LlamaFactory
by Yaowei Zheng
LlamaFactory GitHub repo (ACL 2024): unified fine-tuning toolkit for over 100 language and vision-language models.
open-instruct (Ai2)
by Ai2
Ai2 open-instruct GitHub repo: post-training codebase for instruction tuning and RL of language models.
Unsloth notebooks
by Unsloth AI
Unsloth notebooks GitHub repo: Colab notebooks for fine-tuning and RL of text, vision and audio models.
Unsloth
by Unsloth AI
Unsloth GitHub repo: memory-efficient fine-tuning and local running of LLMs.
Axolotl docs
by Axolotl AI
Axolotl documentation: open-source, config-driven tool for fine-tuning and post-training LLMs.
HF smol course
by Hugging Face
Hugging Face smol course GitHub repo: practical course on fine-tuning and aligning small language models.
PyTorch tutorials
Official PyTorch tutorials covering basics, training, deployment and performance, with runnable code.
RLHF Book (Nathan Lambert)
Nathan Lambert's free online RLHF book covering preference tuning, reward models, RLVR and post-training of language models.
TensorFlow tutorials
Official TensorFlow tutorials on training and deploying machine-learning models.
Engineering fast, performant AI agents through fine-tuning (Decagon)
by Decagon
Decagon engineering post on using several fine-tuned smaller models to cut latency in customer-service agents.
Direct Preference Optimization
by Rafael Rafailov et al. (Stanford University)
arXiv paper (2023) introducing Direct Preference Optimization (DPO), preference training without a separate reward model.
LoRA
by Edward J. Hu et al. (Microsoft)
arXiv paper (2021) "LoRA: Low-Rank Adaptation of Large Language Models", introducing LoRA, low-rank adaptation for fine-tuning large models.
Training language models to follow instructions with human feedback
by Long Ouyang et al. (OpenAI)
arXiv paper (2022) "Training language models to follow instructions with human feedback", describing InstructGPT and the RLHF post-training pipeline.
TRL documentation
by Hugging Face
Hugging Face TRL documentation: trainers for SFT, DPO, GRPO and other post-training methods.
Tülu 3
by Nathan Lambert et al. (Ai2)
arXiv paper (2024) describing Tülu 3, a fully open post-training recipe that introduced RLVR.