simpo-training
Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.
Install this skill
Works with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode, and 70+ other harnesses.
npx skills add KalarisLabs/research-agent-skills --skill simpo-traininguvx research-agent-skills install simpo-trainingDocumentation & Instructions
SimPO - Simple Preference Optimization
Quick start
SimPO is a reference-free preference optimization method that outperforms DPO without needing a reference model.
Installation:
# 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):
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):
# 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:
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):
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:
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:
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:
learning_rate: 3e-7 # Reduce from 5e-7
Reduce beta:
beta: 1.0 # Reduce from 2.0
Issue: Model forgets capabilities
Add SFT regularization:
sft_weight: 0.1 # Add SFT loss component
Issue: Poor preference separation
Increase beta and margin:
beta: 5.0 # Increase from 2.0
gamma_beta_ratio: 0.8 # Increase from 0.5
Issue: OOM during training
Reduce batch size:
per_device_train_batch_size: 1
gradient_accumulation_steps: 16 # Maintain effective batch
Enable gradient checkpointing:
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
Agent operating procedure
- Check the environment. Check GPU type, memory and driver/CUDA versions (
nvidia-smi), framework versions, and dataset location and size. - Pin down the inputs. Confirm formats, identifiers and parameters from the data or the user. Ask rather than guess any value that changes the result.
- Run a small version first. Do a smoke run: tiny model or subset, few steps, and confirm loss decreases and checkpoints save.
- Execute the full task using the instructions and references above.
- Validate the result. Track metrics on held-out data, compare against a baseline, and record seeds, configs and hardware.
- Report. State what was run (versions, commands, parameters), what was checked, and what is still uncertain.
| If this happens | Do this |
|---|---|
| CUDA out-of-memory | Reduce batch size, enable gradient accumulation/checkpointing or mixed precision, or shard the model. |
| Loss is NaN or diverges | Lower the learning rate, check data for invalid values, and enable gradient clipping. |
| A function, flag or endpoint in these instructions is missing in the installed version | Check the installed version's own documentation (help(), --help, official docs), adapt, and tell the user. Never invent an API. |
| A required input, identifier or parameter is ambiguous | Ask the user, or state the assumption explicitly before running. |
Integrity rules
- Never fabricate results, parameters, identifiers, citations or statistics. If something cannot be run or verified, say so plainly.
- Never claim training results without logs; estimate compute cost before launching large jobs.
- Treat version-specific details here as possibly outdated: confirm them against the official documentation for the installed version.
- Ask before actions that cost money, consume shared GPUs or cloud quota, touch personal or patient data, or cannot be undone.
Related skills
fine-tuning-with-trl: Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimi…openrlhf-training: High-performance RLHF framework with Ray+vLLM acceleration.axolotl: Provides guidance for fine-tuning large language models with Axolotl, covering YAML training configs, LoRA and QLoRA, preference training w…
Trigger Prompts & Evaluation Corpus
These evaluation prompts test that the agent correctly selects simpo-training for relevant tasks and leaves it out for near misses.
Should trigger (3)
- should trigger
“How do I train a model with SimPO instead of DPO?”
expects Explains SimPO is reference-free, so no reference model is loaded. Gives the install and launch steps from the skill (config YAML and accelerate/DeepSpeed ZeRO-3 launch), and mentions the key hyperparameters (beta, gamma, learning rate). Checks the environment and the preference data format first.
- should trigger
“I have 60k chosen/rejected pairs in ultrafeedback_binarized.jsonl and a single node with 2x A100 40GB. I want to align Llama-3-8B-Instruct. My last run with lr 1e-5 had the loss go to NaN after about 200 steps. Can you set up the config and help me stop it from diverging?”
expects Builds a Llama 3 8B instruct preference-optimization config using the SimPO workflow, with ZeRO-3, BF16 and gradient checkpointing, which fits the 2x A100 40GB. For the divergence it lowers the learning rate (much lower than 1e-5 for the instruct model), possibly reduces beta, checks the data for invalid values, and enables gradient clipping. It runs a small smoke test first and doesn't claim results without logs.
- should trigger
“I've got a Mistral 7B that's already been SFT'd, plus a pile of pairwise human preference labels. I want to push it toward preferred responses without the memory cost of keeping a frozen copy of the model around, and I don't want to build a reward model or run PPO. What's the simplest approach and how do I run it?”
expects Recommends a reference-free preference optimization method (SimPO) because it needs no frozen reference model and no reward model. Describes the Mistral 7B workflow with hardware and VRAM expectations and notes the trade-offs against DPO. Asks about or checks the data format and GPU setup, and suggests a small trial run before the full job.
Near misses (2)
- near miss
“Using TRL's DPOTrainer, help me fine-tune Qwen2.5-7B on my preference dataset with a frozen reference model. I need the trainer config and the beta setting for the DPO loss.”
expects Uses the TRL DPOTrainer workflow with a reference model and the DPO beta. It does not switch to SimPO, because the user explicitly asked for DPO in TRL.
- near miss
“I need to run PPO on a 70B model across 4 nodes with Ray and vLLM generation, using a trained reward model. How should I set up the cluster and launch it?”
expects Points to a multi-node distributed RLHF setup (Ray, vLLM, ZeRO-3, PPO) and does not use SimPO's single-node, reference-free approach. Covers the actor, critic, reward and reference placement across GPUs.