Instructions to use Team-PIXEL/pixel-m4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Team-PIXEL/pixel-m4 with Transformers:
# Load model directly from transformers import AutoModelForPreTraining model = AutoModelForPreTraining.from_pretrained("Team-PIXEL/pixel-m4", dtype="auto") - Notebooks
- Google Colab
- Kaggle
`font_file` filename differs from the documented `GoNotoCurrent.ttf`
Hi! Thanks for releasing PIXEL-M4.
While reproducing the model, I noticed that the font_file entry in last-checkpoint/text_renderer_config.json points to a file with a long hexadecimal filename:
"font_file": "49e6dc219d1a1a1c9236acaf05a48b542002016a6dc877ee72baab085a84257b.3f28e7f4b38e1efe1b6da4a3732404c19d4c6a614ff32dce90a251e293d4ce58"
However, both the project repository and README mention using GoNotoCurrent.ttf as the renderer font.
To verify whether these were different fonts, I compared their SHA-256 hashes:
pixel-m4/<hashed_filename>
83ab5c39e2b1c34a955136275ce0db068cb20d9643ead033d6b8124a73ab4f64
pixel-base-bigrams/GoNotoCurrent.ttf
83ab5c39e2b1c34a955136275ce0db068cb20d9643ead033d6b8124a73ab4f64
The hashes are identical, so the files are byte-for-byte the same.
Could you clarify whether the hashed filename is intentional (for example, due to packaging or an internal storage process), or whether it was accidentally uploaded under this name?
If there is no specific reason for the hashed filename, would it make sense to rename it back to GoNotoCurrent.ttf (and update text_renderer_config.json accordingly)? It would make reproducing the model a bit more straightforward and avoid confusion for future users.
Thanks!
The hashed filename is due to the caching mechanism of Huggingface transformers package. I'll fix this issue in the near future and I'm leaving this issue open until we fix this.