hermes-agent/skills/mlops/simpo/SKILL.md
teknium1 14e59706b7 Add Skills Hub — universal skill search, install, and management from online registries
Implements the Hermes Skills Hub with agentskills.io spec compliance,
multi-registry skill discovery, security scanning, and user-driven
management via CLI and /skills slash command.

Core features:
- Security scanner (tools/skills_guard.py): 120 threat patterns across
  12 categories, trust-aware install policy (builtin/trusted/community),
  structural checks, unicode injection detection, LLM audit pass
- Hub client (tools/skills_hub.py): GitHub, ClawHub, Claude Code
  marketplace, and LobeHub source adapters with shared GitHubAuth
  (PAT + gh CLI + GitHub App), lock file provenance tracking, quarantine
  flow, and unified search across all sources
- CLI interface (hermes_cli/skills_hub.py): search, install, inspect,
  list, audit, uninstall, publish (GitHub PR), snapshot export/import,
  and tap management — powers both `hermes skills` and `/skills`

Spec conformance (Phase 0):
- Upgraded frontmatter parser to yaml.safe_load with fallback
- Migrated 39 SKILL.md files: tags/related_skills to metadata.hermes.*
- Added assets/ directory support and compatibility/metadata fields
- Excluded .hub/ from skill discovery in skills_tool.py

Updated 13 config/doc files including README, AGENTS.md, .env.example,
setup wizard, doctor, status, pyproject.toml, and docs.
2026-02-18 16:09:05 -08:00

222 lines
5.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

---
name: simpo-training
description: 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.
version: 1.0.0
author: Orchestra Research
license: MIT
dependencies: [torch, transformers, datasets, trl, accelerate]
metadata:
hermes:
tags: [Post-Training, SimPO, Preference Optimization, Alignment, DPO Alternative, Reference-Free, LLM Alignment, Efficient Training]
---
# 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](references/loss-functions.md) for sigmoid vs hinge loss, mathematical formulations, and when to use each.
**Hyperparameter tuning**: See [references/hyperparameters.md](references/hyperparameters.md) for beta, gamma, learning rate selection guide, and model-size-specific recommendations.
**Dataset preparation**: See [references/datasets.md](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