chore: add Moshi #914
chore: add Moshi #914
Conversation
Vaibhavs10
requested review from
SBrandeis,
Wauplin,
gary149,
julien-c,
osanseviero and
pcuenca
as code owners
| repoName: "Moshi", | ||
| repoUrl: "https://github.com/kyutai-labs/moshi", | ||
| filter: false, | ||
| countDownloads: `path:"tokenizer-e351c8d8-checkpoint125.safetensors"`, |
There was a problem hiding this comment.
Yes! Since there's a mic of safetensors and GGUFs this points to the tokenizer for now - this is the same across all the repos, irrespective of the backend (candle, mlx, PyTorch) - it's the safest way I found without double counting.
There was a problem hiding this comment.
Any thoughts about having
| countDownloads: `path:"tokenizer-e351c8d8-checkpoint125.safetensors"`, | |
| countDownloads: `path:"model.safetensors" OR path:"model.q4.safetensors" OR path:"model.q8.safetensors" OR path:"model.gguf" OR path:"model.q4.gguf" OR path:"model.q8.gguf"`, |
?
It's long and ugly but at least if a new tokenizer is pushed, it won't break the download count. And it wouldn't duplicate counts if I checked correctly.
Otherwise I'm fine with the current solution, there is no perfect one anyway.
Yes! I was thinking about it, but wasn't sure how best to support all three backends, since each has a different setup and installation 😅 |
if i'm not mistaken there's a way to specify multiple snippets now, even with titles/collapsible sections etc |
|
can be in a separate PR though, if you prefer |
Is it possible to add (agree can be done in a follow-up PR) |
There was a problem hiding this comment.
Thanks for handling it @Vaibhavs10 :)
Approving "as is" (comments above are optional)
|
Cool! ah yeah! Good shout re: MLX + Candle tags, I'll add them tomorrow/ later today and make a follow-up PR 🤗 Thank you for the review and suggestions! ❤️ |
ref: https://huggingface.co/models?other=moshi