Usage

Tutorial

from fastermodels.hf import *
from fastai.vision.all import *
from fastai.callback.all import *
from fasterai.core.criteria import *
import torch_pruning as tp
from torch_pruning.pruner import function
from torch_pruning.pruner.algorithms.scheduler import linear_scheduler
import timm

import torch
import torch.nn as nn
import torch.nn.functional as F

from fasterbench.benchmark import *
from fasterai.prune.all import *

hf_model_name = 'tutorial'
model_name = 'alexnet'

Get a model

model = globals()[model_name]()
model_size = get_model_size(model)

def get_dls(size, bs):
    path = URLs.IMAGENETTE_160
    source = untar_data(path)
    blocks=(ImageBlock, CategoryBlock)
    tfms = [RandomResizedCrop(size, min_scale=0.35), FlipItem(0.5)]
    batch_tfms = [Normalize.from_stats(*imagenet_stats)]

    csv_file = 'noisy_imagenette.csv'
    inp = pd.read_csv(source/csv_file)
    dblock = DataBlock(blocks=blocks,
               splitter=ColSplitter(),
               get_x=ColReader('path', pref=source),
               get_y=ColReader(f'noisy_labels_0'),
               item_tfms=tfms,
               batch_tfms=batch_tfms)

    return dblock.dataloaders(inp, path=source, bs=bs)

dls = get_dls(160, 32)
learn = Learner(dls, model, metrics = [accuracy])
learn.fit_one_cycle(5)

epoch	train_loss	valid_loss	accuracy	time
0	2.190084	2.294868	0.154650	00:05
1	1.957930	1.917212	0.328153	00:05
2	1.689020	1.605206	0.442293	00:05
3	1.430839	1.361510	0.538344	00:05
4	1.343451	1.291558	0.572994	00:05

loss, acc = learn.validate()

Upload the model

metadata = {
    "model_name": hf_model_name,
    "architecture": model_name,
    "framework": "PyTorch",
    "datasets": ["ImageNet"],
    "task": "Classification",
    "metrics": {"Model Accuracy": acc},
    "usage_example": """""",
    "description": "A simple torchvision model",
    "license": "mit",
    "tags": ["classification"],
}

# Generate the README content
readme_content = generate_readme(metadata)

# Save the README to a file
with open("README.md", "w") as f:
    f.write(readme_content)

print("README.md has been generated.")

README.md has been generated.

create_repository(f"Nathan12/{hf_model_name}")

save_and_upload_model(f"Nathan12/{hf_model_name}", model, f"{model_name}", readme_path="README.md")

Load the model from HuggingFace

from huggingface_hub import hf_hub_download
import torch

# Download the model from Hugging Face
model_file = hf_hub_download(
    repo_id=f"Nathan12/{hf_model_name}",
    filename=model_name
)

# Load the model (architecture + weights)
model = torch.load(model_file)

print("Model loaded successfully!")

/tmp/ipykernel_11566/2452638795.py:11: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  model = torch.load(model_file)

Model loaded successfully!

learn = Learner(dls, model, metrics = [accuracy])
learn.fit_one_cycle(5)

epoch	train_loss	valid_loss	accuracy	time
0	1.561073	1.590430	0.476688	00:05
1	1.508280	1.513178	0.504204	00:05
2	1.322130	1.321613	0.576815	00:05
3	1.143002	1.087752	0.644331	00:05
4	1.034787	1.018889	0.674140	00:05