Kohaku-Blueleaf KohakuBlueleaf

Benchmark for Xmrig

This is a python script that run a customized stratum server to specific the algorigthm combinations of 6 CN algos(used in ghostrider) And it calculate the Expected Average Hashrate automatically(All the weights are as same as cpuminer-gr's)

How to run

You need to install python3.8↑ with pip first! And download the script to xmrig's path

linux

Sd-WebUI standalone script

just throw the script into the root folder of sd-webui. and then modify the content to what you want to run. And then run it.

If you want to use some extension, just use my implementation as example (to modify the args for extensions)

	import torch
	import numpy as np
	import timeit

	#could be "cpu" "cuda" "opencl" "mps"...
	DEVICE = "mps"
	ROUNDS = 500
	NUM = (2048, 2048)

	# Input for benchmarking

	import os, sys

	from time import time_ns
	from json import load
	from io import BytesIO
	import math

	from PIL import Image

	import asyncio

	import torch
	import torch.nn as nn
	import torch.nn.functional as F

	from einops import rearrange


	def parallel_retention(
	q, k, v, # bsz, heads, seq_len, dim
	decay_mask = None # heads, seq_len, seq_len

	import asyncio
	import cv2
	import json
	import traceback
	from concurrent.futures import ProcessPoolExecutor
	from pathlib import Path

	import numpy as np
	from httpx import AsyncClient, HTTPError, Timeout
	from tqdm import tqdm


	import torch
	import torch.nn as nn
	import torch.nn.functional as F


	class TimestepEmbedding(nn.Module):
	def __init__(self, n_time=1024, n_emb=320, n_out=1280) -> None:
	super().__init__()
	self.emb = nn.Embedding(n_time, n_emb)

	## Note
	## if use vllm in same gpu, remember to set a low gpu_memory_utilization to avoid OOM
	## For larger model please consider to use multi-GPU or CPU offloading
	## AnySchedule: https://github.com/KohakuBlueleaf/AnySchedule
	## LyCORIS: https://github.com/KohakuBlueleaf/LyCORIS
	## Following code can perform reasonable training on Llama-3.2-1B-Instruct model with GSM8K dataset
	## With noticable improvement on each reward function
	from itertools import chain
	import re

	"""
	An Minimal Implementation of IMM (Inductive Moment Matching)
	"""

	import math
	import torch
	import torch.nn.functional as F


	def compute_mmd_loss_fully_vectorized(