ysdede’s gists

ysdede / Optimize_tk3.py

Last active November 4, 2022 18:54

Multi-objective NSGAII hyperparameters optimization with Optuna v3

	import contextlib
	import json
	import signal
	import sys
	from subprocess import PIPE, Popen, call
	import click

	import jessetk2.utils as utils
	from optuna import samplers
	import optuna

ysdede / gist:403ac8eab6f082739bc8c08b093b3135

Created March 29, 2023 15:53

Postgresql create/setup databases using the command line

	Linux shell:
	```bash
	sudo -u postgres psql -c "CREATE DATABASE optuna_db;"
	sudo -u postgres psql -c "CREATE USER optuna_user WITH PASSWORD 'password';"
	sudo -u postgres psql -c "GRANT ALL PRIVILEGES ON DATABASE optuna_db to optuna_user;"
	```

	Window:
	```cmd
	psql -U postgres -W -c "CREATE DATABASE optuna_db;"

ysdede / sha256.py

Created September 1, 2024 13:15

sha256.py Numba njit implementation

	"""This Python module is an implementation of the SHA-256 algorithm.
	From https://github.com/keanemind/Python-SHA-256"""

	'''
	MIT License

	Copyright (c) 2018 Keane Nguyen

	Permission is hereby granted, free of charge, to any person obtaining a copy
	of this software and associated documentation files (the "Software"), to deal

ysdede / phi-4-mm debug 1

Created March 5, 2025 20:25

	=== Critical Debug Info ===
	1. Top-level modules: ['base_model']
	2. All modules containing 'audio': ['base_model.model.base_model.model.model.embed_tokens_extend.audio_embed', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.drop', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed.out', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed.conv', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed.conv.0', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed.conv.1', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed.conv.2', 'base_model.model.base_model.model.model.embed_tokens_extend.audio_embed.encoder.embed.conv.3', 'base_model.model.base_mo

ysdede / Phi-4-logs

Last active March 5, 2025 20:39

	Training on 1 GPUs
	Loading checkpoint shards: 100%
	3/3 [00:02<00:00, 1.50it/s]
	Audio embed structure: Phi4MMAudioEmbedding(
	(drop): Dropout(p=0.0, inplace=False)
	(encoder): ConformerEncoder(
	(embed): NemoConvSubsampling(
	(out): Linear(in_features=10240, out_features=1024, bias=True)
	(conv): Sequential(
	(0): Conv2d(1, 1024, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))

ysdede / gist:da189c5239454de712c076d68f78f945

Last active March 6, 2025 14:47

Phi-4-mm Trainable params

	Trainable parameters: 1,296,888,448
	Unfrozen components:
	- model.embed_tokens_extend.audio_embed.encoder.embed.out.weight
	- model.embed_tokens_extend.audio_embed.encoder.embed.out.bias
	- model.embed_tokens_extend.audio_embed.encoder.embed.conv.0.weight
	- model.embed_tokens_extend.audio_embed.encoder.embed.conv.0.bias
	- model.embed_tokens_extend.audio_embed.encoder.embed.conv.2.weight
	- model.embed_tokens_extend.audio_embed.encoder.embed.conv.2.bias
	- model.embed_tokens_extend.audio_embed.encoder.embed.conv.3.weight
	- model.embed_tokens_extend.audio_embed.encoder.embed.conv.3.bias

Yunus Seyhan Dede ysdede