to uv

README.md

@@ -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)
 
 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.

pyproject.toml

@@ -1,34 +1,30 @@
-[tool.poetry]
+[project]
 name = "simpo"
 version = "0.1.0"
 description = ""
-authors = ["wassname <1103714+wassname@users.noreply.github.com>"]
 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]
-requires = ["poetry-core"]
-build-backend = "poetry.core.masonry.api"
+requires = ["hatchling"]
+build-backend = "hatchling.build"
 
-[[tool.poetry.source]]
-name = "pytorch"
-url = "https://download.pytorch.org/whl/cu121"
-priority = "explicit"
+[tool.hatch.build.targets.wheel]
+packages = ["simpo"]
 
-[tool.poetry.dependencies]
-python = "^3.10.9,<4.0"
-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
+[tool.uv.sources]
+alignment-handbook = { git = "https://github.com/huggingface/alignment-handbook.git" }