Cannot run bigcode example : access to source requires login credentials

[noahgift]

I love this project so far! Thanks everyone for working on it. I got several models to work but did run into an issue here. Hopefully, someone can help me with it.

When I run cargo run --example bigcode --release. I get a similar issue as issue 350.

codespace@codespaces-f226cf:/workspaces/rust-candle-demos/candle$ cargo run --example bigcode --release -- --prompt "hello world python function"
warning: some crates are on edition 2021 which defaults to `resolver = "2"`, but virtual workspaces default to `resolver = "1"`
note: to keep the current resolver, specify `workspace.resolver = "1"` in the workspace root's manifest
note: to use the edition 2021 resolver, specify `workspace.resolver = "2"` in the workspace root's manifest
    Finished release [optimized] target(s) in 3.68s
     Running `target/release/examples/bigcode --prompt 'hello world python function'`
Error: request error: https://huggingface.co/bigcode/starcoderbase-1b/resolve/main/tokenizer.json: status code 403

Caused by:
    https://huggingface.co/bigcode/starcoderbase-1b/resolve/main/tokenizer.json: status code 403
codespace@codespaces-f226cf:/workspaces/rust-candle-demos/candle$ 

Here are the steps that I tried:

• Authorized the form "Gated model
  You have been granted access to this model" for starcoder
• Setup HUGGING_FACE_HUB_TOKEN in GitHub Codespaces secret (Didn't work)
• echo $HUGGING_FACE_HUB_TOKEN > $HOME/.cache/huggingface/token (Didn't work)
• pip install huggingface_hub && huggingface-cli login (Didn't work)
• Tried to open up a vanilla python shell and ran >>> from transformers import pipeline then pipe = pipeline("text-generation", model="bigcode/starcoder")

FYI, this is my huggingface-cli env output

codespace@codespaces-f226cf:/workspaces/rust-candle-demos$ huggingface-cli env

Copy-and-paste the text below in your GitHub issue.

- huggingface_hub version: 0.16.4
- Platform: Linux-5.15.0-1041-azure-x86_64-with-glibc2.31
- Python version: 3.10.8
- Running in iPython ?: No
- Running in notebook ?: No
- Running in Google Colab ?: No
- Token path ?: /home/codespace/.cache/huggingface/token
- Has saved token ?: True
- Who am I ?: noahgift
- Configured git credential helpers: /.codespaces/bin/gitcredential_github.sh
- FastAI: N/A
- Tensorflow: N/A
- Torch: 2.0.1
- Jinja2: 3.1.2
- Graphviz: N/A
- Pydot: N/A
- Pillow: 10.0.0
- hf_transfer: N/A
- gradio: N/A
- tensorboard: N/A
- numpy: 1.25.2
- pydantic: N/A
- aiohttp: N/A
- ENDPOINT: https://huggingface.co
- HUGGINGFACE_HUB_CACHE: /home/codespace/.cache/huggingface/hub
- HUGGINGFACE_ASSETS_CACHE: /home/codespace/.cache/huggingface/assets
- HF_TOKEN_PATH: /home/codespace/.cache/huggingface/token
- HF_HUB_OFFLINE: False
- HF_HUB_DISABLE_TELEMETRY: False
- HF_HUB_DISABLE_PROGRESS_BARS: None
- HF_HUB_DISABLE_SYMLINKS_WARNING: False
- HF_HUB_DISABLE_EXPERIMENTAL_WARNING: False
- HF_HUB_DISABLE_IMPLICIT_TOKEN: False
- HF_HUB_ENABLE_HF_TRANSFER: False

[Narsil]

Try setting HF_HOME ?

By default it should indeed be in $HOME/.cache/huggingface/token as you tried.

https://github.com/huggingface/hf-hub/blob/main/src/lib.rs#L169-L181

I'm guessing something must be wrong in codespaces regarding HOME maybe ?

[LaurentMazare]

Looks like this uses starcoder/bigcodebase-1b but the link you've put is for starcoder/bigcode, maybe get access to https://huggingface.co/bigcode/starcoderbase-1b if you haven't already done so?

[noahgift]

your a miracle worker @LaurentMazare ! This was it:

codespace@codespaces-f226cf:/workspaces/rust-candle-demos/candle$ cargo run --example bigcode --release -- --prompt "hello world python function"
warning: some crates are on edition 2021 which defaults to `resolver = "2"`, but virtual workspaces default to `resolver = "1"`
note: to keep the current resolver, specify `workspace.resolver = "1"` in the workspace root's manifest
note: to use the edition 2021 resolver, specify `workspace.resolver = "2"` in the workspace root's manifest
    Finished release [optimized] target(s) in 4.43s
     Running `target/release/examples/bigcode --prompt 'hello world python function'`
tokenizer.json [00:00:00] [███████████████████████████████████████████████████████████████████] 1.96 MiB/1.96 MiB 19.97 MiB/s (0s)
model.safetensors [00:01:26] [████████████████████████████████████████████████████████████████] 4.24 GiB/4.24 GiB 49.91 MiB/s (0s)retrieved the files in 96.9321653s
Running on CPU, to run on GPU, build this example with `--features cuda`
loaded the model in 3.7673761s
starting the inference loop
hello world python function")

# +
# %%writefile hello_world.py

def hello_world():
    print("hello world python function")

[noahgift]

Thanks @Narsil Codespaces it a bit finicky. The issue turned out to be I didn't realize I need to ask for access to starcoderbase-1b.

Btw, I plan on making a reproducible .devcontainer project since I teach with this at Duke University and we get free GPU :). So I will share this config when done.

Thanks again for the help! HUGE fan of this project.