HDCharles’s gists

HDCharles / gpucheck.sh

Created April 30, 2026 15:55

always know who is using the allocations

	#!/bin/bash

	# === Configuration ===
	SSH_CONFIG="$HOME/.ssh/config"
	STATE_FILE="$HOME/.gpucheck_state.json"
	LAST_SYNC_FILE="$HOME/.gpucheck_last_sync"
	HOSTS_URL="https://vigilant-happiness-o78jgy9.pages.github.io/hosts"
	SSH_USER="HDCharles"
	POLL_INTERVAL=300 # seconds between checks (5 minutes)
	# Three EMA alpha values for different time horizons

HDCharles / quantize.py

Created April 30, 2026 14:58

NVFP4 regression test suite: quantize.py and run_all_tests.sh

	import argparse
	import os
	import time

	import torch
	from compressed_tensors.offload import dispatch_model
	from compressed_tensors.quantization import preset_name_to_scheme
	from datasets import load_dataset
	from transformers import AutoModelForCausalLM, AutoTokenizer

HDCharles / quantize.py

Created April 28, 2026 20:02

LLM Compressor regression testing scripts

	import argparse
	import time

	import torch
	from compressed_tensors.offload import dispatch_model
	from compressed_tensors.quantization import preset_name_to_scheme
	from datasets import load_dataset
	from transformers import AutoModelForCausalLM, AutoTokenizer

	from llmcompressor import oneshot

HDCharles / extract_log_summary.py

Last active April 25, 2026 02:59

LLM Compressor Testing Setup - GPTQ actorder regression tests

	#!/usr/bin/env python3
	"""Extract summary data from AWQ DDP regression test log files.

	Parses the log output from run_all_tests.sh and produces a comparison table
	showing pre-DDP vs post-DDP results across models, schemes, and benchmarks.

	Usage:
	python extract_log_summary.py regression_results.log
	"""

HDCharles / bench_logprobs.py

Created April 23, 2026 20:49

vLLM full-vocab logprob timing benchmark (Llama-3-8B)

	import time, os

	model_path = "/mnt/data/engine/HDCharles/hf_hub/models--meta-llama--Meta-Llama-3-8B-Instruct/snapshots/8afb486c1db24fe5011ec46dfbe5b5dccdb575c2"

	from vllm import LLM, SamplingParams

	print("Loading model...")
	t0 = time.time()
	llm = LLM(model=model_path, gpu_memory_utilization=0.5, max_logprobs=-1)
	tokenizer = llm.get_tokenizer()

HDCharles / extract_log_summary.py

Created April 22, 2026 20:28

GPTQ actorder regression test suite for llm-compressor (FP8 block, W4A16, W8A16)

	#!/usr/bin/env python3
	"""Extract summary data from AWQ DDP regression test log files.

	Parses the log output from run_all_tests.sh and produces a comparison table
	showing pre-DDP vs post-DDP results across models, schemes, and benchmarks.

	Usage:
	python extract_log_summary.py regression_results.log
	"""

HDCharles / extract_log_summary.py

Created April 22, 2026 18:54

GPTQ actorder regression test suite for llm-compressor

	#!/usr/bin/env python3
	"""Extract summary data from AWQ DDP regression test log files.

	Parses the log output from run_all_tests.sh and produces a comparison table
	showing pre-DDP vs post-DDP results across models, schemes, and benchmarks.

	Usage:
	python extract_log_summary.py regression_results.log
	"""

HDCharles / extract_log_summary.py

Created April 20, 2026 15:55

AWQ DDP Regression Test Suite — compares pre-DDP vs post-DDP AWQ quantization quality across models, schemes, and benchmarks

	#!/usr/bin/env python3
	"""Extract summary data from AWQ DDP regression test log files.

	Parses the log output from run_all_tests.sh and produces a comparison table
	showing pre-DDP vs post-DDP results across models, schemes, and benchmarks.

	Usage:
	python extract_log_summary.py regression_results.log
	"""

HDCharles / extract_log_summary.py

Created April 15, 2026 03:54

AWQ DDP Regression Test Suite - quantization + eval across models, schemes, and pre/post-DDP commits

	#!/usr/bin/env python3
	"""Extract summary data from AWQ DDP regression test log files.

	Parses the log output from run_all_tests.sh and produces a comparison table
	showing pre-DDP vs post-DDP results across models, schemes, and benchmarks.

	Usage:
	python extract_log_summary.py regression_results.log
	"""

HDCharles / extract_log_summary.py

Created April 13, 2026 03:34

AWQ DDP Regression Test Suite - pre/post DDP quality comparison across models, formats, and benchmarks

	#!/usr/bin/env python3
	"""Extract summary data from AWQ DDP regression test log files.

	Parses the log output from run_all_tests.sh and produces a comparison table
	showing pre-DDP vs post-DDP results across models, schemes, and benchmarks.

	Usage:
	python extract_log_summary.py regression_results.log
	"""