Bram Wasti bwasti

https://twitter.com/bwasti

bwasti / gt_diagram.md

Created November 3, 2025 22:45

bwasti / invariance_test.py

Created October 21, 2025 16:48

	from vllm import LLM, SamplingParams

	# Setup model (prefix caching disabled)
	llm = LLM(model="Qwen/Qwen3-1.7B", enable_prefix_caching=False, dtype="bfloat16")
	prompt = "Ok, this is an extremely long story. There once was a "
	params = SamplingParams(temperature=0.6, max_tokens=256, logprobs=1, seed=42)

	# Generate 256 tokens, extract token 256's logprob
	out1 = llm.generate([prompt], params)
	tokens = out1[0].outputs[0].token_ids

bwasti / gpu_tensorcore_monitor.py

Created September 23, 2025 15:33

	#!/usr/bin/env python3
	"""
	Real-time GPU Process Monitor with TensorCore Inference
	Monitors all GPU processes and infers TensorCore usage based on workload patterns
	"""

	import subprocess
	import json
	import time
	import psutil

bwasti / bleh.md

Created September 18, 2025 13:50

bwasti / test.md

Created September 17, 2025 23:21

bwasti / images.md

Last active September 16, 2025 16:10

bwasti / benchmark_oai.py

Created August 26, 2025 21:26

	#!/usr/bin/env python3
	"""
	OpenAI Prediction API Benchmark Tool

	Benchmarks latency and throughput for the OpenAI Completions API with prediction functionality.
	Supports custom endpoints (e.g., localhost:8000) for testing vLLM implementations.
	"""

	import asyncio
	import time

bwasti / test_bucketMul.py

Last active April 18, 2024 02:09

	# This is a test (not implementation) of the impact bucketMul has on matrix multiplications
	# https://kolinko.github.io/effort/bucketmul.html
	import torch
	import torch.nn.functional as F
	import math
	torch.manual_seed(1337)

	B = 2
	N = 8
	M = 16

bwasti / bun_sqlite.prompt.md

Last active September 17, 2023 18:50

Convert `sqlite3` to `bun:sqlite` ChatGPT prompt

Here's the API interface to bun:sqlite,

class Database {
  constructor(
    filename: string,
    options?:
      | number
      | {
 readonly?: boolean;

bwasti / lock_bench.py

Created August 29, 2023 19:52

	import time
	import multiprocessing

	def test_lock(lock, iterations, shared_value):
	for _ in range(iterations):
	with lock:
	shared_value.value += 1

	def benchmark(lock_type, num_processes, iterations_per_process):
	shared_value = multiprocessing.Value('i', 0)