Skip to content

Instantly share code, notes, and snippets.

View entrpn's full-sized avatar

Juan Acevedo entrpn

  • Google
  • Lake Forest, CA
View GitHub Profile
@karpathy
karpathy / nes.py
Last active June 7, 2025 14:26
Natural Evolution Strategies (NES) toy example that optimizes a quadratic function
"""
A bare bones examples of optimizing a black-box function (f) using
Natural Evolution Strategies (NES), where the parameter distribution is a
gaussian of fixed standard deviation.
"""
import numpy as np
np.random.seed(0)
# the function we want to optimize
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@skye
skye / tpu_topology_env_vars.py
Last active June 17, 2025 12:51
You can use these environment variables to run a Python process on a subset of the TPU cores on a Cloud TPU VM. This allows running multiple TPU processes at the same time, since only one process can access a given TPU chip at a time. Note that on TPU v2 and v3, 1 TPU chip = 2 TpuDevice as reported by `jax.devices()` (8 devices total). On v4, 1 …
# ==== Non-communicating processes
# 4x 1 chip per process:
os.environ["TPU_CHIPS_PER_PROCESS_BOUNDS"] = "1,1,1"
os.environ["TPU_PROCESS_BOUNDS"] = "1,1,1"
# Different per process:
os.environ["TPU_VISIBLE_DEVICES"] = "0" # "1", "2", "3"
# 1-liner for bash: TPU_CHIPS_PER_PROCESS_BOUNDS=1,1,1 TPU_PROCESS_BOUNDS=1,1,1 TPU_VISIBLE_DEVICES=0 TPU_MESH_CONTROLLER_ADDRESS=localhost:8476 TPU_MESH_CONTROLLER_PORT=8476
@kohya-ss
kohya-ss / forward_of_sdxl_original_unet.py
Created November 14, 2023 03:39
SDXLで高解像度での構図の破綻を軽減する
def forward(self, x, timesteps=None, context=None, y=None, **kwargs):
# broadcast timesteps to batch dimension
timesteps = timesteps.expand(x.shape[0])
hs = []
t_emb = get_timestep_embedding(timesteps, self.model_channels) # , repeat_only=False)
t_emb = t_emb.to(x.dtype)
emb = self.time_embed(t_emb)
assert x.shape[0] == y.shape[0], f"batch size mismatch: {x.shape[0]} != {y.shape[0]}"