HardikJha/extractor-aea

Extractor (SFT warmup) — Adversarial Structured-Extraction Arena

This model repo hosts the SFT warmup LoRA adapter trained for the OpenEnv project Adversarial Structured-Extraction Arena: an adversary perturbs messy documents/schemas (under a budget) and the extractor must output valid JSON matching a target schema.

Links (submission)

• GitHub repo: https://github.com/Hardikjha09/openenv-adversarial-extraction-arena
• Runnable Space: https://huggingface.co/spaces/HardikJha/extraction-arena
• Colab (re-run training): https://colab.research.google.com/github/Hardikjha09/openenv-adversarial-extraction-arena/blob/main/notebooks/Train_Extractor_Colab.ipynb
• Paired adversary LoRA: https://huggingface.co/HardikJha/adversary-aea

Evidence (plots + logs)

• Training loss: https://huggingface.co/HardikJha/extractor-aea/blob/main/plots/sft_loss.png
• Eval reward (moving average): https://huggingface.co/HardikJha/extractor-aea/blob/main/plots/rewards.png
• Eval Elo: https://huggingface.co/HardikJha/extractor-aea/blob/main/plots/elo_ratings.png
• Eval metrics JSON: https://huggingface.co/HardikJha/extractor-aea/blob/main/eval_metrics.json
• SFT trainer log (raw JSON): https://huggingface.co/HardikJha/extractor-aea/blob/main/trainer_log_history.json

What this checkpoint is

• Base model: unsloth/Qwen2.5-1.5B-Instruct (4-bit Unsloth bundle: unsloth/qwen2.5-1.5b-instruct-unsloth-bnb-4bit)
• Adapter: LoRA (peft), saved from training/sft_warmup.py

Quick start (load base + adapter)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "unsloth/Qwen2.5-1.5B-Instruct"
adapter_id = "HardikJha/extractor-aea"

tokenizer = AutoTokenizer.from_pretrained(adapter_id, trust_remote_code=True)

base = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True,
)
model = PeftModel.from_pretrained(base, adapter_id)

Training procedure

• Objective: supervised JSON extraction formatting aligned to the repo’s extractor prompt (training/prompts.py)
• Framework: TRL SFTTrainer + Unsloth FastLanguageModel (see training/sft_warmup.py)

This model was trained with SFT.

Framework versions

• PEFT 0.18.1
• TRL: 0.23.0
• Transformers: 4.57.2
• Pytorch: 2.10.0+cu128
• Datasets: 4.3.0
• Tokenizers: 0.22.2

Citations

Cite TRL as:

@misc{vonwerra2022trl,
    title        = {{TRL: Transformer Reinforcement Learning}},
    author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
    year         = 2020,
    journal      = {GitHub repository},
    publisher    = {GitHub},
    howpublished = {\url{https://github.com/huggingface/trl}}
}