# The download of a model is too slow

**URL:** <https://discuss.huggingface.co/t/the-download-of-a-model-is-too-slow/53279>\
**Category:** Beginners\
**Created:** [September 2, 2023, 6:01pm UTC](https://discuss.huggingface.co/t/the-download-of-a-model-is-too-slow/53279 "2023-09-02T18:01:30Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![gabifg](https://avatars.discourse-cdn.com/v4/letter/g/bbce88/32.png) [@gabifg](https://discuss.huggingface.co/u/gabifg)\
**Post date:** [September 2, 2023, 6:01pm UTC](https://discuss.huggingface.co/t/the-download-of-a-model-is-too-slow/53279/1 "2023-09-02T18:01:30Z")

</div>

Hi everyone! 👋

I am trying to download the Falcon-7B model from its [repo](https://huggingface.co/tiiuae/falcon-7b) into my local machine and the download speed is very slow. I get an average download speed of 2.50 Mbps and it takes nearly 1h to download the 15GB weight files (I have good internet connection with at least 15Mbps)

When I try downloading the same model in Google Colab the download speed is way faster. The first time I tried to download the model it took less than 4 minutes, and the next ones ~10 minutes (I suppose Hugging Face servers slow down repetitive calls from the same user)

Here is the code I run to download the model (I don’t have GPU)

```auto
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoTokenizer

model_id = "tiiuae/falcon-7b"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    device_map="auto", 
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
    cache_dir="cache_models")

```

I’ve also tried cloning the repo through `git lfs` but after some minutes waiting I get a warning stating that the weight models could not be downloaded correctly. When I try to load the model from the cloned repo I get the error: _“Unable to load weights from pytorch checkpoint file for ‘./cache\_models/falcon-7b\pytorch\_model-00001-of-00002.bin’ at ‘./cache\_models/falcon-7b\pytorch\_model-00001-of-00002.bin’. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from\_tf=True.”_

Appending the option `from_tf=True` yields the error: _‘list’ object has no attribute ‘endswith’_ which I don’t know how to fix, but I suspect that the original error comes from the faulty download of the weights by `git lfs`

**Does anyone have faced a similar problem or might know why is this happening? Is there any chance to speed up the download process?**

  
PS. When I try to download other models like [MTP-7B](https://huggingface.co/mosaicml/mpt-7b) the download times remain the same. PC specifications: 
- i7-1065G7 CPU @ 1.30GHz
- 12GB RAM
- Windows 11
- 200GB free disk space