CoderDoge/synthfix-deepseek-coder-1.3b-pyrepair

1---2library_name: peft3base_model: deepseek-ai/deepseek-coder-1.3b-base4pipeline_tag: text-generation5tags:6- code7- program-repair8- vulnerability-repair9- peft10- lora11- synthfix12- router13- pyrepair14---15 16# SynthFix deepseek-coder-1.3b-base (pyrepair)17 18This repository contains the SynthFix LoRA repair-agent adapter and matching router checkpoint for **Python functional repair** on **pyrepair**.19 20SynthFix is an adaptive neuro-symbolic code repair framework. During training, a lightweight router chooses between supervised fine-tuning (SFT) and reward fine-tuning (RFT), while the reward combines compiler-, analyzer-, and test-derived symbolic evidence. At inference time, the same evidence supports best-of-K candidate selection with a greedy floor.21 22## Contents23 24- `adapter_model.safetensors`, `adapter_config.json`: PEFT/LoRA adapter for `deepseek-ai/deepseek-coder-1.3b-base`.25- tokenizer files copied from the training checkpoint.26- `router.pt`: PyTorch state dict for the SynthFix router associated with this adapter.27 28## Base Model29 30This adapter is intended to be loaded with `deepseek-ai/deepseek-coder-1.3b-base` using `peft.PeftModel.from_pretrained`.31 32```python33from transformers import AutoModelForCausalLM, AutoTokenizer34from peft import PeftModel35 36base_model = "deepseek-ai/deepseek-coder-1.3b-base"37adapter_id = "CoderDoge/synthfix-deepseek-coder-1.3b-pyrepair"38 39tokenizer = AutoTokenizer.from_pretrained(adapter_id, trust_remote_code=True)40base = AutoModelForCausalLM.from_pretrained(base_model, trust_remote_code=True)41model = PeftModel.from_pretrained(base, adapter_id)42```43 44The router architecture is implemented in the SynthFix artifact repository: https://github.com/CoderDoge1108/SynthFix and https://github.com/largehappygroup/SynthFix.45 46## Paper47 48SynthFix: Adaptive Neuro-Symbolic Code Vulnerability Repair49 50- arXiv: https://arxiv.org/abs/2604.1718451- GitHub artifact: https://github.com/CoderDoge1108/SynthFix52 53## Notes54 55These are adapter weights rather than full copies of the base models. Users should follow the license and usage terms of the corresponding base model.56