mitchellnw

import numpy as np
import torch

# This is the unfused version of StableAdamW. It is slower than the fused version (coming).

class StableAdamWUnfused(torch.optim.Optimizer):

    def __init__(self, params, lr=0.002, weight_decay=0.2, betas=(0.9, 0.99), eps=1e-6, clip_thresh=1., precision='amp_bfloat16', custom_scalar=65536):
        beta1, beta2 = betas[0], betas[1]
        defaults = dict(lr=lr, weight_decay=weight_decay, beta1=beta1, beta2=beta2)

cat-state

to use

HF_API_TOKEN=<token> sbatch hf-infer.sbatch

then run

HF_API_TOKEN=<token> HOSTNAME=<hostname of infernce server> python gradio-tgl.py

setup env following hf inference server instructions but chance /usr/local to path to conda env instead.