kdubovikov

🤗 Hugging Face Downloader

Note

(2026-03-10) Ported CAMD reliability updates: retry/backoff controls, network timeout controls, and proxy-prefix fallback support. (2025-01-08) Added revision selection via --revision, contributed by @Bamboo-D. (2024-12-17) Switched to the metadata/file-list flow for faster startup and resume without git clone.

This script downloads models and datasets from Hugging Face with curl plus either aria2c or wget. It keeps the repo's fast metadata-based startup and resume behavior, and now also includes the CAMD script's retry/backoff and timeout controls.

Features

kdubovikov

apiVersion: v1
kind: Service
metadata:
  labels:
    app: chopstick-classifier
  name: chopstick-classifier
spec:
  type: LoadBalancer # we need a load balancer to distribute our requests between all replicas in the Deployment
  ports:
  - port: 8500  # run this 

kdubovikov

apiVersion: apps/v1
kind: Deployment
metadata:
  name: chopstick-classifier
spec:
  replicas: 3 # Here we tell Kubernetes to keep 3 replicas up. This will help us with reliability and scalability. The value can be changed online and Kubernetes will update the number of replicas to required quantity
  selector:
    matchLabels: # all pods labeled "app: chopstick-classifier will be in scope of this deployment"
      app: chopstick-classifier
  template: # this template will be allied to each replica in the set

kdubovikov

from __future__ import absolute_import, division, print_function

import argparse

import tensorflow as tf
from colorama import Back, Fore, Style, init
from grpc.beta import implementations

from tensorflow_serving.apis import predict_pb2, prediction_service_pb2

kdubovikov

from __future__ import absolute_import, division, print_function

import argparse

import tensorflow as tf
from colorama import Back, Fore, Style, init
from grpc.beta import implementations

from tensorflow_serving.apis import classification_pb2, prediction_service_pb2

kdubovikov

"""Train chopstick efficiency classifier and export it as Servable"""
import argparse

import numpy as np
import pandas as pd
import tensorflow as tf


def preprocess_data(filename, val_rows_count=20):
    """Load data and split it to train and validation sets"""

kdubovikov

"""Train chopstick efficiency classifier and export it as Servable"""
import argparse

import pandas as pd
import tensorflow as tf


def preprocess_data(filename, val_rows_count=20):
    """Load data and split it to train and validation sets"""

kdubovikov

import seaborn as sns
import matplotlib.pyplot as plt
%matplotlib inline

fig, ax = plt.subplots(figsize=(12,8),sharey=True)
sns.despine(left=True)
sns.distplot(choice_num, ax=ax)
sns.distplot(fast_num, ax=ax)
ax.set_title("Collision count distributions")

kdubovikov

from scipy.stats import ttest_rel 

# We must do depentent t-test since all statistics were collected from the 
# same array
ch_fast_test = ttest_rel(choice_num, fast_num)
print(f"choice sampling vs fast sampling significant: {ch_fast_test.pvalue < 0.05}, p = {ch_fast_test.pvalue}")

kdubovikov

def collect_collisions(arr, sample_size, n_samples, 
                       fast=False):
    """Collect total number of collisions made for each sample of arr.
    
    Parameters
    ----------
    arr: np.array
        array to sample from.
    sample_size: int
        sample size.