tlrmchlsmth

def decompress_2_4(metadata, values):
    """
    Decompress 2:4 sparse tensor
    Args:
        metadata: Tensor of shape (N, K/8) uint8.
                  Each byte stores indexes for 2 blocks
        values: Tensor of shape (N, K/2) bf16, storing non-zero values
    Returns:
        Dense tensor of shape (N, K) in bf16
    """

tlrmchlsmth

import numpy as np
import msgpack
import pickle
import time
import json
from msgspec import msgpack as msgspec_msgpack
import struct
from io import BytesIO

def benchmark_methods(arr, num_iterations=1000):

tlrmchlsmth

import os
import time
import asyncio
import argparse
import statistics
from typing import List

def benchmark_time_sleep(sleep_duration: float, iterations: int) -> List[float]:
    durations = []
    for _ in range(iterations):

tlrmchlsmth

import pdb
import traceback
import inspect


def function_in_stack(function_name):
    stack = traceback.extract_stack() 
    for frame in stack:
        if frame.name == function_name:
            return True

tlrmchlsmth

#!/bin/bash

# Default values
model="meta-llama/Meta-Llama-3-8B"
port=8192
num_prompts=500

# TODO: Add other datasets
dataset_path="ShareGPT_V3_unfiltered_cleaned_split.json"
dataset_url="https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/$dataset_path"