Add notebooks to Examples docs and restructure (#4317)

Co-authored-by: Kashif Rasul <[email protected]>

Author: sergiopaniego

docs/source/example_overview.md

@@ -1,39 +1,51 @@
 # Examples
 
-## Introduction
+This directory contains a collection of examples that demonstrate how to use the TRL library for various applications. We provide both **scripts** for advanced use cases and **notebooks** for an easy start and interactive experimentation.
 
-The examples should work in any of the following settings (with the same script):
+The notebooks are self-contained and can run on **free Colab**, while the scripts can run on **single GPU, multi-GPU, or DeepSpeed** setups.
 
-- single GPU
-- multi GPUs (using PyTorch distributed mode)
-- multi GPUs (using DeepSpeed ZeRO-Offload stages 1, 2, & 3)
-- fp16 (mixed-precision), fp32 (normal precision), or bf16 (bfloat16 precision)
+**Getting Started**
 
-To run it in each of these various modes, first initialize the accelerate configuration with `accelerate config`.
-
-To train with a 4-bit or 8-bit model, please run:
+Install TRL and additional dependencies as follows:
 
 ```bash
 pip install --upgrade trl[quantization]
 ```
 
-## Accelerate Config
+Check for additional optional dependencies [here](https://github.com/huggingface/trl/blob/main/pyproject.toml).
 
-For all the examples, you'll need to generate a 🤗 Accelerate config file with:
+For scripts, you will also need an 🤗 Accelerate config (recommended for multi-gpu settings):
 
-```shell
+```bash
 accelerate config # will prompt you to define the training configuration
 ```
 
-Then, it is encouraged to launch jobs with `accelerate launch`!
+This allows you to run scripts with `accelerate launch` in single or multi-GPU settings.
+
+## Notebooks
+
+These notebooks are easier to run and are designed for quick experimentation with TRL. The list of notebooks can be found in the [`trl/examples/notebooks/`](https://github.com/huggingface/trl/tree/main/examples/notebooks/) directory.
+
+
+| Notebook | Description | Open in Colab |
+|----------|-------------|---------------|
+| [`sft_trl_lora_qlora.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/sft_trl_lora_qlora.ipynb) | Supervised Fine-Tuning (SFT) using QLoRA on free Colab | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/trl/blob/main/examples/notebooks/sft_trl_lora_qlora.ipynb) |
+| [`sft_qwen_vl.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/sft_qwen_vl.ipynb) | Supervised Fine-Tuning (SFT) Qwen3-VL with QLoRA using TRL on free Colab | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/trl/blob/main/examples/notebooks/sft_qwen_vl.ipynb) |
+| [`grpo_qwen3_vl.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/grpo_qwen3_vl.ipynb) | GRPO Qwen3-VL with QLoRA using TRL on free Colab | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/trl/blob/main/examples/notebooks/grpo_qwen3_vl.ipynb) |
+
+Legacy / Older Notebooks
 
-## Maintained Examples
+- [`best_of_n.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/best_of_n.ipynb): This notebook demonstrates how to use the "Best of N" sampling strategy using TRL when fine-tuning your model with PPO.
+- [`gpt2-sentiment.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/gpt2-sentiment.ipynb): This notebook demonstrates how to reproduce the GPT2 imdb sentiment tuning example on a jupyter notebook.
+- [`gpt2-sentiment-control.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/gpt2-sentiment-control.ipynb): This notebook demonstrates how to reproduce the GPT2 sentiment control example on a jupyter notebook.
 
-Scripts can be used as examples of how to use TRL trainers. They are located in the [`trl/scripts`](https://github.com/huggingface/trl/blob/main/trl/scripts) directory. Additionally, we provide examples in the [`examples/scripts`](https://github.com/huggingface/trl/blob/main/examples/scripts) directory. These examples are maintained and tested regularly.
+## Scripts
 
-| File | Description |
+Scripts are maintained in the [`trl/scripts`](https://github.com/huggingface/trl/blob/main/trl/scripts) and [`examples/scripts`](https://github.com/huggingface/trl/blob/main/examples/scripts) directories. They show how to use different trainers such as `SFTTrainer`, `PPOTrainer`, `DPOTrainer`, `GRPOTrainer`, and more.
+
+ File | Description |
 | --- | --- |
-| [`examples/scripts/bco.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/bco.py) | This script shows how to use the [`KTOTrainer`] with the BCO loss to fine-tune a model to increase instruction-following, truthfulness, honesty and helpfulness using the [openbmb/UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback) dataset. |
+| [`examples/scripts/bco.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/bco.py) | This script shows how to use the [`KTOTrainer`] with the BCO loss to fine-tune a model to increase instruction-following, truthfulness, honesty, and helpfulness using the [openbmb/UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback) dataset. |
 | [`examples/scripts/cpo.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/cpo.py) | This script shows how to use the [`CPOTrainer`] to fine-tune a model to increase helpfulness and harmlessness using the [Anthropic/hh-rlhf](https://huggingface.co/datasets/Anthropic/hh-rlhf) dataset. |
 | [`trl/scripts/dpo.py`](https://github.com/huggingface/trl/blob/main/trl/scripts/dpo.py) | This script shows how to use the [`DPOTrainer`] to fine-tune a model. |
 | [`examples/scripts/dpo_vlm.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/dpo_vlm.py) | This script shows how to use the [`DPOTrainer`] to fine-tune a Vision Language Model to reduce hallucinations using the [openbmb/RLAIF-V-Dataset](https://huggingface.co/datasets/openbmb/RLAIF-V-Dataset) dataset. |
@@ -52,38 +64,28 @@ Scripts can be used as examples of how to use TRL trainers. They are located in
 | [`examples/scripts/ppo/ppo.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/ppo/ppo.py) | This script shows how to use the [`PPOTrainer`] to fine-tune a model to improve its ability to continue text with positive sentiment or physically descriptive language. |
 | [`examples/scripts/ppo/ppo_tldr.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/ppo/ppo_tldr.py) | This script shows how to use the [`PPOTrainer`] to fine-tune a model to improve its ability to generate TL;DR summaries. |
 | [`examples/scripts/prm.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/prm.py) | This script shows how to use the [`PRMTrainer`] to fine-tune a Process-supervised Reward Model (PRM). |
-| [`examples/scripts/reward_modeling.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/reward_modeling.py) | This script shows how to use the [`RewardTrainer`] to train a Outcome Reward Model (ORM) on your own dataset. |
+| [`examples/scripts/reward_modeling.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/reward_modeling.py) | This script shows how to use the [`RewardTrainer`] to train an Outcome Reward Model (ORM) on your own dataset. |
 | [`examples/scripts/rloo.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/rloo.py) | This script shows how to use the [`RLOOTrainer`] to fine-tune a model to improve its ability to solve math questions. |
 | [`examples/scripts/sft.py`](https://github.com/huggingface/trl/blob/main/trl/scripts/sft.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a model. |
 | [`examples/scripts/sft_gemma3.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/sft_gemma3.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a Gemma 3 model. |
 | [`examples/scripts/sft_video_llm.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/sft_video_llm.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a Video Language Model. |
-| [`examples/scripts/sft_vlm.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/sft_vlm.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a Vision Language Model in a chat setting. The script has only been tested with [LLaVA 1.5](https://huggingface.co/llava-hf/llava-1.5-7b-hf), [LLaVA 1.6](https://huggingface.co/llava-hf/llava-v1.6-mistral-7b-hf), and [Llama-3.2-11B-Vision-Instruct](https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct) models so users may see unexpected behaviour in other model architectures. |
+| [`examples/scripts/sft_vlm.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/sft_vlm.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a Vision Language Model in a chat setting. The script has only been tested with [LLaVA 1.5](https://huggingface.co/llava-hf/llava-1.5-7b-hf), [LLaVA 1.6](https://huggingface.co/llava-hf/llava-v1.6-mistral-7b-hf), and [Llama-3.2-11B-Vision-Instruct](https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct) models, so users may see unexpected behaviour in other model architectures. |
 | [`examples/scripts/sft_vlm_gemma3.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/sft_vlm_gemma3.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a Gemma 3 model on vision to text tasks. |
 | [`examples/scripts/sft_vlm_smol_vlm.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/sft_vlm_smol_vlm.py) | This script shows how to use the [`SFTTrainer`] to fine-tune a SmolVLM model. |
 | [`examples/scripts/xpo.py`](https://github.com/huggingface/trl/blob/main/examples/scripts/xpo.py) | This script shows how to use the [`XPOTrainer`] to fine-tune a model. |
 
-Here are also some easier-to-run colab notebooks that you can use to get started with TRL:
-
-| File | Description |
-| --- | --- |
-| [`examples/notebooks/best_of_n.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/best_of_n.ipynb) | This notebook demonstrates how to use the "Best of N" sampling strategy using TRL when fine-tuning your model with PPO. |
-| [`examples/notebooks/gpt2-sentiment.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/gpt2-sentiment.ipynb) | This notebook demonstrates how to reproduce the GPT2 imdb sentiment tuning example on a jupyter notebook. |
-| [`examples/notebooks/gpt2-control.ipynb`](https://github.com/huggingface/trl/tree/main/examples/notebooks/gpt2-control.ipynb) | This notebook demonstrates how to reproduce the GPT2 sentiment control example on a jupyter notebook. |
-
-## Distributed training
+## Distributed Training (for scripts)
 
-All the scripts can be run on multiple GPUs by providing the path of an 🤗 Accelerate config file when calling `accelerate launch`. To launch one of them on one or multiple GPUs, run the following command (swapping `{NUM_GPUS}` with the number of GPUs in your machine and `--all_arguments_of_the_script` with your arguments).
+You can run scripts on multiple GPUs with 🤗 Accelerate:
 
 ```shell
 accelerate launch --config_file=examples/accelerate_configs/multi_gpu.yaml --num_processes {NUM_GPUS} path_to_script.py --all_arguments_of_the_script
 ```
 
-You can also adjust the parameters of the 🤗 Accelerate config file to suit your needs (e.g. training in mixed precision).
-
-### Distributed training with DeepSpeed
-
-Most of the scripts can be run on multiple GPUs together with DeepSpeed ZeRO-{1,2,3} for efficient sharding of the optimizer states, gradients, and model weights. To do so, run the following command (swapping `{NUM_GPUS}` with the number of GPUs in your machine, `--all_arguments_of_the_script` with your arguments, and `--deepspeed_config` with the path to the DeepSpeed config file such as `examples/deepspeed_configs/deepspeed_zero1.yaml`):
+For DeepSpeed ZeRO-{1,2,3}:
 
 ```shell
 accelerate launch --config_file=examples/accelerate_configs/deepspeed_zero{1,2,3}.yaml --num_processes {NUM_GPUS} path_to_script.py --all_arguments_of_the_script
 ```
+
+Adjust `NUM_GPUS` and `--all_arguments_of_the_script` as needed.