mirror of
https://github.com/wassname/SimPO.git
synced 2026-08-12 11:40:25 +08:00
to uv
This commit is contained in:
@@ -1,3 +1,22 @@
|
|||||||
|
In this form I SFT some base models, so they are suitable for testing alternatives to DPO. It's for my project: https://github.com/wassname/repr-preference-optimization
|
||||||
|
|
||||||
|
I train:
|
||||||
|
- [ ] HuggingFaceTB/SmolLM2-135M
|
||||||
|
- [ ] HuggingFaceTB/SmolLM2-360M
|
||||||
|
- [ ] HuggingFaceTB/SmolLM2-1.7B
|
||||||
|
- [ ] Qwen/Qwen3-0.6B-Base
|
||||||
|
- [ ] Qwen/Qwen2-4B-Base
|
||||||
|
- [x] https://huggingface.co/wassname/llama-3-2-1b-sft
|
||||||
|
|
||||||
|
Other peoples base->sft models
|
||||||
|
- allenai/OLMo-2-0425-1B-SFT
|
||||||
|
- allenai/OLMoE-1B-7B-0924-SFT
|
||||||
|
- https://huggingface.co/allenai/OLMo-2-0425-1B-SFT
|
||||||
|
- https://huggingface.co/allenai/OLMo-7B-SFT-hf
|
||||||
|
- https://huggingface.co/princeton-nlp/Llama-3-Base-8B-SFT
|
||||||
|
|
||||||
|
Original readme
|
||||||
|
----
|
||||||
# Simple Preference Optimization (SimPO)
|
# Simple Preference Optimization (SimPO)
|
||||||
|
|
||||||
This repository contains the code and released models for our paper [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://arxiv.org/abs/2405.14734). We propose a simpler and more effective preference optimization algorithm than DPO (Direct Preference Optimization) without using a reference model. SimPO outperforms DPO and its latest variants across AlpacaEval 2, MT-Bench, and Arena-Hard benchmarks under various settings.
|
This repository contains the code and released models for our paper [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://arxiv.org/abs/2405.14734). We propose a simpler and more effective preference optimization algorithm than DPO (Direct Preference Optimization) without using a reference model. SimPO outperforms DPO and its latest variants across AlpacaEval 2, MT-Bench, and Arena-Hard benchmarks under various settings.
|
||||||
|
|||||||
+22
-26
@@ -1,34 +1,30 @@
|
|||||||
[tool.poetry]
|
[project]
|
||||||
name = "simpo"
|
name = "simpo"
|
||||||
version = "0.1.0"
|
version = "0.1.0"
|
||||||
description = ""
|
description = ""
|
||||||
authors = ["wassname <1103714+wassname@users.noreply.github.com>"]
|
|
||||||
readme = "README.md"
|
readme = "README.md"
|
||||||
package-mode = false
|
requires-python = ">=3.10.9,<4.0"
|
||||||
|
|
||||||
|
dependencies = [
|
||||||
|
"torch>=2.1.2",
|
||||||
|
"deepspeed>=0.14",
|
||||||
|
"wandb>=0.18.2",
|
||||||
|
# git‐based dep
|
||||||
|
"alignment-handbook",
|
||||||
|
]
|
||||||
|
|
||||||
|
[dependency-groups]
|
||||||
|
dev = [
|
||||||
|
"pytest",
|
||||||
|
"ipykernel>=6.29.5",
|
||||||
|
]
|
||||||
|
|
||||||
[build-system]
|
[build-system]
|
||||||
requires = ["poetry-core"]
|
requires = ["hatchling"]
|
||||||
build-backend = "poetry.core.masonry.api"
|
build-backend = "hatchling.build"
|
||||||
|
|
||||||
[[tool.poetry.source]]
|
[tool.hatch.build.targets.wheel]
|
||||||
name = "pytorch"
|
packages = ["simpo"]
|
||||||
url = "https://download.pytorch.org/whl/cu121"
|
|
||||||
priority = "explicit"
|
|
||||||
|
|
||||||
[tool.poetry.dependencies]
|
[tool.uv.sources]
|
||||||
python = "^3.10.9,<4.0"
|
alignment-handbook = { git = "https://github.com/huggingface/alignment-handbook.git" }
|
||||||
torch = { version = ">=2.1.2+cu121", source = "pytorch" }
|
|
||||||
alignment-handbook = {git = "https://github.com/huggingface/alignment-handbook.git"}
|
|
||||||
#huggingface-hub = "~0.23.0"
|
|
||||||
# trl = "~0.9.0"
|
|
||||||
deepspeed = "^0.14"
|
|
||||||
# transformers="~4.39.3"
|
|
||||||
# python -m pip install flash-attn --no-build-isolation
|
|
||||||
wandb = "^0.18.2"
|
|
||||||
|
|
||||||
[tool.poetry.group.dev.dependencies]
|
|
||||||
pytest = "*"
|
|
||||||
ipykernel = "^6.29.5"
|
|
||||||
|
|
||||||
[virtualenvs]
|
|
||||||
in-project = true
|
|
||||||
|
|||||||
Reference in New Issue
Block a user