orchestra-research/ai-research-skills

simpo-training

Simple Preference Optimization for LLM alignment.

Voir la source

Mode d’emploi

  1. Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
  2. Consultez le document original pour les outils, identifiants et environnements requis.
  3. Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/orchestra-research/ai-research-skills --skill simpo-training

Avant de commencer

  • L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
  • Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Document Skill original

Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.

SimPO - Simple Preference Optimization

Quick start

SimPO is a reference-free preference optimization method that outperforms DPO without needing a reference model.

Installation:

bash
# Create environment
conda create -n simpo python=3.10 && conda activate simpo

# Install PyTorch 2.2.2
# Visit: https://pytorch.org/get-started/locally/

# Install alignment-handbook
git clone https://github.com/huggingface/alignment-handbook.git
cd alignment-handbook
python -m pip install .

# Install Flash Attention 2
python -m pip install flash-attn --no-build-isolation

Training (Mistral 7B):

bash
ACCELERATE_LOG_LEVEL=info accelerate launch \
  --config_file accelerate_configs/deepspeed_zero3.yaml \
  scripts/run_simpo.py \
  training_configs/mistral-7b-base-simpo.yaml

Common workflows

Workflow 1: Train from base model (Mistral 7B)

Config (mistral-7b-base-simpo.yaml):

yaml
# Model
model_name_or_path: mistralai/Mistral-7B-v0.1
torch_dtype: bfloat16

# Dataset
dataset_mixer:
  HuggingFaceH4/ultrafeedback_binarized: 1.0
dataset_splits:
  - train_prefs
  - test_prefs

# SimPO hyperparameters
beta: 2.0                  # Reward scaling (2.0-10.0)
gamma_beta_ratio: 0.5       # Target margin (0-1)
loss_type: sigmoid          # sigmoid or hinge
sft_weight: 0.0             # Optional SFT regularization

# Training
learning_rate: 5e-7         # Critical: 3e-7 to 1e-6
num_train_epochs: 1
per_device_train_batch_size: 1
gradient_accumulation_steps: 8

# Output
output_dir: ./outputs/mistral-7b-simpo

Launch training:

bash
accelerate launch --config_file accelerate_configs/deepspeed_zero3.yaml \
  scripts/run_simpo.py training_configs/mistral-7b-base-simpo.yaml

Workflow 2: Fine-tune instruct model (Llama 3 8B)

Config (llama3-8b-instruct-simpo.yaml):

yaml
model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct

dataset_mixer:
  argilla/ultrafeedback-binarized-preferences-cleaned: 1.0

beta: 2.5
gamma_beta_ratio: 0.5
learning_rate: 5e-7
sft_weight: 0.1             # Add SFT loss to preserve capabilities

num_train_epochs: 1
per_device_train_batch_size: 2
gradient_accumulation_steps: 4
output_dir: ./outputs/llama3-8b-simpo

Launch:

bash
accelerate launch --config_file accelerate_configs/deepspeed_zero3.yaml \
  scripts/run_simpo.py training_configs/llama3-8b-instruct-simpo.yaml

Workflow 3: Reasoning-intensive tasks (lower LR)

For math/code tasks:

yaml
model_name_or_path: deepseek-ai/deepseek-math-7b-base

dataset_mixer:
  argilla/distilabel-math-preference-dpo: 1.0

beta: 5.0                   # Higher for stronger signal
gamma_beta_ratio: 0.7       # Larger margin
learning_rate: 3e-7         # Lower LR for reasoning
sft_weight: 0.0

num_train_epochs: 1
per_device_train_batch_size: 1
gradient_accumulation_steps: 16

When to use vs alternatives

Use SimPO when:

  • Want simpler training than DPO (no reference model)
  • Have preference data (chosen/rejected pairs)
  • Need better performance than DPO
  • Limited compute resources
  • Single-node training sufficient

Algorithm selection:

  • SimPO: Simplest, best performance, no reference model
  • DPO: Need reference model baseline, more conservative
  • PPO: Maximum control, need reward model, complex setup
  • GRPO: Memory-efficient RL, no critic

Use alternatives instead:

  • OpenRLHF: Multi-node distributed training, PPO/GRPO
  • TRL: Need multiple methods in one framework
  • DPO: Established baseline comparison

Common issues

Issue: Loss divergence

Reduce learning rate:

yaml
learning_rate: 3e-7  # Reduce from 5e-7

Reduce beta:

yaml
beta: 1.0  # Reduce from 2.0

Issue: Model forgets capabilities

Add SFT regularization:

yaml
sft_weight: 0.1  # Add SFT loss component

Issue: Poor preference separation

Increase beta and margin:

yaml
beta: 5.0            # Increase from 2.0
gamma_beta_ratio: 0.8  # Increase from 0.5

Issue: OOM during training

Reduce batch size:

yaml
per_device_train_batch_size: 1
gradient_accumulation_steps: 16  # Maintain effective batch

Enable gradient checkpointing:

yaml
gradient_checkpointing: true

Advanced topics

Loss functions: See references/loss-functions.md for sigmoid vs hinge loss, mathematical formulations, and when to use each.

Hyperparameter tuning: See references/hyperparameters.md for beta, gamma, learning rate selection guide, and model-size-specific recommendations.

Dataset preparation: See references/datasets.md for preference data formats, quality filtering, and custom dataset creation.

Hardware requirements

  • GPU: NVIDIA A100/H100 recommended
  • VRAM:
  • 7B model: 1× A100 40GB (DeepSpeed ZeRO-3)
  • 8B model: 2× A100 40GB
  • 70B model: 8× A100 80GB
  • Single-node: DeepSpeed ZeRO-3 sufficient
  • Mixed precision: BF16 recommended

Memory optimization:

  • DeepSpeed ZeRO-3 (default config)
  • Gradient checkpointing
  • Flash Attention 2

Resources

  • Paper: https://arxiv.org/abs/2405.14734 (NeurIPS 2024)
  • GitHub: https://github.com/princeton-nlp/SimPO
  • Models: https://huggingface.co/princeton-nlp
  • Alignment Handbook: https://github.com/huggingface/alignment-handbook

Tâches connexes

du même dépôt

Autres Skills

Tous les Skills
orchestra-research
Communauté

stable-diffusion-image-generation

State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.

Installations / 1h
2
GitHub Stars
14,7 k
Mis à jour
11 oct.
orchestra-research
Communauté

evaluating-cosmos-policy

Evaluates NVIDIA Cosmos Policy on LIBERO and RoboCasa simulation environments. Use when setting up cosmos-policy for robot manipulation evaluation, running headless GPU evaluations with EGL rendering, or profiling inference latency on cluster or local GPU machines.

Installations / 1h
1
GitHub Stars
14,7 k
Mis à jour
11 oct.
orchestra-research
Communauté

fine-tuning-openvla-oft

Fine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation with continuous action heads, LoRA adaptation, and FiLM conditioning on LIBERO simulation and ALOHA real-world setups. Use when reproducing OpenVLA-OFT paper results, training custom VLA action heads (L1 or diffusion), deploying server-client inference for ALOHA, or debugging normalization, LoRA merge, and cross-GPU issues.

Installations / 1h
1
GitHub Stars
14,7 k
Mis à jour
11 oct.
orchestra-research
Communauté

transformer-lens-interpretability

Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.

Installations / 1h
1
GitHub Stars
14,7 k
Mis à jour
11 oct.