Skip to content

Instantly share code, notes, and snippets.

View HarshVasisht's full-sized avatar
🍇

Harsh Vasisht HarshVasisht

🍇
View GitHub Profile
@HarshVasisht
HarshVasisht / grpo_demo.py
Created January 31, 2025 15:58 — forked from willccbb/grpo_demo.py
GRPO Llama-1B
# train_grpo.py
import re
import torch
from datasets import load_dataset, Dataset
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import LoraConfig
from trl import GRPOConfig, GRPOTrainer
# Load and prep dataset
def transition (input_char, replace_with, move, new_state):
global tapehead, state
if tape[tapehead] == input_char:
tape[tapehead] = replace_with
state = new_state
if move == 'L':
tapehead -= 1
else:
tapehead += 1
return True