Where is SmolLM2_model.py???

#1
by MartialTerran - opened

Where on huggingface.com is the actual model.py (e.g., SmolLM2_model.py) to run these weights on local Windows machine in python script? Without a python model.py script, it cannot be modified nor exported as torchscript to run on other non-python hardware. I am not interested in using "autotransformer" which conceals the actual model.py script.

Where on(https://huggingface.co./vonjack/SmolLM2-1.7B-Merged/) is the actual model.py (e.g., SmolLM2-1.7B-Merged _model.py) to run these weights on local Windows machine in python script? Without a python model.py script, it cannot be modified nor exported as torchscript to run on other non-python hardware. I am not interested in using "autotransformer" which conceals the actual model.py script.

Where on huggingface.com is the actual train.py (e.g., SmolLM2_model.py) to train/finetune these weights on local Windows machine in python script?

Hi, I'm using LlamaForCausalLM, so I think no extra xxx_model.py is needed.

There are dozens of modifications to be made to a SmolLM2_model.py for improved local inference or edge operation on various hardware, or for experimentation, that cannot be implemented while using a cryptic and unobtainable Huggingface LlamaForCausalLM library.
And, the LlamaForCausalLM does not currently work with the provided parameters.
https://huggingface.co./HuggingFaceTB/SmolLM2-135M/discussions/4

If you have a readable, modifiable pytorch model.py for SmolLM2_model.py please provide a URL.

Sign up or log in to comment