ashok969/houdini-vex-project

Houdini VEX Coding Assistant — Project

Fine-tune of Qwen2.5-Coder-7B-Instruct on kelvincai/HoudiniVexBench (Houdini VEX code completion, doc-to-code, and code explanation).

Contents of this repo

• code/ — training notebook
• data/ — reformatted train/val JSONL used for SFT

Model weights (separate repos — see why in the training notebook)

• LoRA adapter: https://huggingface.co/ashok969/houdini-vex-lora
• Merged standalone model: https://huggingface.co/ashok969/houdini-vex-merged

Usage (LoRA adapter)

from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
    "ashok969/houdini-vex-lora", load_in_4bit=True
)

Usage (merged model)

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("ashok969/houdini-vex-merged")
tokenizer = AutoTokenizer.from_pretrained("ashok969/houdini-vex-merged")