arvindcr4/llama-3.2-1b-distillation-offpolicy-lora

Llama 3.2 1B - Distillation Off-Policy LoRA

LoRA adapter trained with Tinker (by Thinking Machines) using off-policy distillation on OpenThoughts3 dataset.

Training Details

• —Base model: meta-llama/Llama-3.2-1B
• —Method: Off-policy distillation (SFT on OpenThoughts3)
• —LoRA rank: 32, alpha: 32
• —Target modules: all-linear
• —Checkpoint: batch 700

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-1B")
model = PeftModel.from_pretrained(base, "arvindcr4/llama-3.2-1b-distillation-offpolicy-lora")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-1B")

Platform

Trained using Tinker - hosted fine-tuning service for open-source LLMs.