PyTorch
ONNX
Safetensors
English
nanochat
Eval Results (legacy)

GGUF support

#12
by Twobombs - opened

thanks for converting this to HF format

with the idea to pull this model even more into mainstream a GGUF conversion seems in place
however, even with the special 5.0dev HF transformer one cannot build the GGUF with llama.cpp because its format is not recognised

python3 convert_hf_to_gguf.py nanochat-d20/ --outfile nanochat-d20-f16.gguf
INFO:hf-to-gguf:Loading model: nanochat-d20
INFO:hf-to-gguf:Model architecture: NanoChatForCausalLM
ERROR:hf-to-gguf:Model NanoChatForCausalLM is not supported

would be great if one could actually convert and have the luxury of all the related tooling of GGUF [?]

opened a request at llama.cpp https://github.com/ggml-org/llama.cpp/discussions/16720 support for this would be appreciated

closed - enough GGUF support in the ecosystem

Twobombs changed discussion status to closed

Where is the enough GGUF support in the ecosystem? I can't find any script to do it.
What I'm missing?

huggingface has many d34 and other iterations of karpathy's repo

Twobombs changed discussion status to open

huggingface has many d34 and other iterations of karpathy's repo

If the repo is this one, i didnt find any sugestive name about the gguf.
https://github.com/karpathy/nanochat

I'm a newbie, could you be kind to give a link to some script or tutorial that explain how to do it.

The only script i found was this one, from nanochat to hf format, but its for an old version, and it doesnt work well with the last version of nanochat checkpoints.

https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/nanochat/convert_nanochat_checkpoints.py

thank you for your reply.
i'm sorry for my bad english

Sign up or log in to comment