kushalicious/research-slm-360m-lora

ResearchSLM-360M LoRA

LoRA adapter fine-tuned on SmolLM2-360M-Instruct for structured research skills (JSON outputs).

Training

Parameter	Value
Base model	SmolLM2-360M-Instruct
Method	LoRA (r=16, alpha=32) via Unsloth
Data	research-slm-dataset — 15k train / 500 eval
Hardware	Google Colab free T4
Steps	250 (3k examples subsampled)

Evaluation (rule-based, 30 examples)

Model	Overall
Base	66.1%
This adapter	67.8%

Usage

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

base = "HuggingFaceTB/SmolLM2-360M-Instruct"
adapter = "kushalicious/research-slm-360m-lora"

tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(model, adapter)

Or use the full research loop from the GitHub repo:

huggingface-cli download kushalicious/research-slm-360m-lora --local-dir lora_adapter
python -m runtime.main "Your research question" --adapter lora_adapter

GitHub

Full code, eval scripts, and Colab notebook: github.com/kushalicious/research-slm