Talkie 1930 in transformers format

#2284
by vasya100 - opened

There's 2 versions of Talkie 1930 that were converted to transformers format, already with config.json, ninja template and .safetensors files

Is it possible to convert one of those to gguf?

https://huggingface.co/lewtun/talkie-1930-13b-it-hf

https://huggingface.co/Abstract4700/talkie-1930-13b-it-hf-transformers

wonderful, someone requested not that long ago. since they seem to be identical, I will queue https://huggingface.co/lewtun/talkie-1930-13b-it-hf
hopefully it's supported

You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#talkie-1930-13b-it-hf-GGUF for quants to appear.

TalkieForCausalLM not compatible =(

please remind me in a week or something, it might be merged into main llama cpp and work =)
you can also try to talk to the author about helping adding compatibility for their model in llamacpp

Someone published a q5 gguf of talkie 1930
https://huggingface.co/sol-wy/talkie-1930-13b-it-q5
said it's not compatible with stock llama.cpp or stock LM Studio because Talkie uses a custom architecture.

but they provide custom patched llama.cpp version to run it
https://github.com/solwyc/talkie-1930-13b-it-q5

no information about those changes in main llama cpp yet

And here's the patch that was applied to llama.cpp to make it work with llama
https://github.com/solwyc/talkie-1930-13b-it-q5/tree/main/patches

we sadly dont support the custom forks, we need it to be merged into main llamacpp

Sign up or log in to comment