🏋️

darren angle darrenangle

🏋️

computer programmer

32 followers · 119 following

Chicago, IL
11:27 (UTC -05:00)
darren.computer
@darrenangle

View GitHub Profile

Recently created

Least recently created

Recently updated

Least recently updated

darrenangle / grpo_demo.py

Created January 28, 2025 23:48 — forked from willccbb/grpo_demo.py

GRPO Llama-1B

	# train_grpo.py
	import re
	from datasets import load_dataset, Dataset
	from transformers import AutoTokenizer
	from peft import LoraConfig
	from trl import GRPOConfig, GRPOTrainer

	# Load and prep dataset

	SYSTEM_PROMPT = """

darrenangle / choice_tree.py

Created June 3, 2024 13:44 — forked from wassname/choice_tree.py

for huggingface transformers sometime you want to constrain output to json schema and record the probabilities on choices/enums. I use it when rating, judging. It's much more efficient than sampling multiple times.

	from jaxtyping import Float, Int
	import torch
	from torch.nn import functional as F
	from torch import Tensor
	from typing import List, Callable, Tuple, Dict, Optional
	import pandas as pd
	from transformers import AutoModelForCausalLM, AutoTokenizer


	def get_valid_next_choices(choices_tokens, current_tokens):

darrenangle / vllm.py

Last active March 20, 2024 16:56

vllm server wrapper with auto restart and vllm node script with exponential backoff for requests

	import subprocess
	import time
	import re
	import signal
	import sys
	import select
	import os

	def start_server():
	command = [

darrenangle / embed_and_load_arxiv.py

Created January 24, 2024 13:09

embed and load 2.5M arxiv abstracts into qdrant

	import json
	import numpy as np
	from FlagEmbedding import FlagModel
	from qdrant_client import QdrantClient
	from qdrant_client.http.models import Batch, VectorParams, Distance
	import uuid

	# Initialize the Qdrant client
	client = QdrantClient(host="localhost", port=6333)

darrenangle / tools.txt

Last active May 5, 2024 20:44

open LLM tool use, a prompt template, tested with Mistral, OpenChat, SOLAR.


	# WHO ARE YOU

	- You are an AI trained to call functions in order to solve problems.
	- You are an expert in XML outputs and well-formatted JSON.

	# YOUR TASK

	- You will be given a scenario that requires a decision.
	- After thinking quietly about the scenario and reflecting on all of your options, you will respond by using a tool.