Zaid Khan codezakh

codezakh / rstar.py

Created January 21, 2025 17:02

A sketch of the MCTS implementation used in rStar.

codezakh / generic_search.py

Last active January 16, 2025 03:43

A single file implementation of a generic, modular search algorithm for natural language reasoning.

codezakh / blip2_confidence_snippet.py

Created February 4, 2024 14:32

Get the answers+confidence from a BLIP-2 model for a VQA task.

	outputs = blip2.generate(
	pixel_values=pixel_values,
	input_ids=input_ids,
	attention_mask=attention_mask,
	do_sample=False,
	num_beams=5,
	max_new_tokens=10,
	min_length=1,
	length_penalty=-1,
	return_dict_in_generate=True,

codezakh / homogenous_batching.py

Created June 26, 2020 19:24

	class CustomBatchSampler(Sampler):
	def __init__(self, batch_size, dataset):
	self.sampler = SequentialSampler(dataset)
	self.batch_size = batch_size
	self.drop_last = False
	self.dataset = dataset
	def __iter__(self):
	batch = []
	for idx in self.sampler:
	batch.append(idx)

codezakh / random_prediction.py

Last active November 23, 2019 04:39

Code to predict random labels on the validation pairs in Phase 1 of the RFIW 2020 Verification Track,

	import pandas as pd
	import numpy as np

	val_pairs = pd.read_csv('val_pairs.csv').set_index('index')

	# Randomly predict {0, 1} for each pair.
	random_predictions = np.random.randint(0,2, size=len(val_pairs))

	# Add the predictions as a column named "label".
	val_pairs['label'] = pd.Series(random_predictions)

codezakh / microsoftOCRsample.json

Created August 17, 2017 21:03

An example response returned by the microsoft cognition api when using OCR.