please, make the model inference-able , any format - just make so people can run infernce please, gguf / normal format whatever, just make it inferenceable

#3075
by Phase-Technologies - opened
Phase-Technologies changed discussion title from please, make the model inference-able , any format - just make so people can run infernce please, gguf / nornal format whatever, just make it inferenceable to please, make the model inference-able , any format - just make so people can run infernce please, gguf / normal format whatever, just make it inferenceable

Phase-Technologies/FORGE-500M-Chat: no architectures entry

please add architectures entry to your model. and of course convert it to safetensors

okay , thanks for the directive

Hi @RichardErkhov ,

The blockers have been resolved:

  1. Added config.json with "architectures": ["ForgeForCausalLM"], "model_type": "forge", and auto_map for Transformers integration.
  2. Converted and uploaded model.safetensors (1.35 GB, verified 100% bitwise parity with original weights).
  3. Added Hugging Face remote code integration (configuration_forge.py, modeling_forge.py).
  4. Removed the redundant .pt file so all security and picklescan checks are completely clean.

The repository is ready for conversion:
https://huggingface.co/Phase-Technologies/FORGE-500M-Chat

Could you please re-run the conversion? Thank you!

Is that custom architecture? Im not sure that will work...

yeah custom arch, you can give it a try

ERROR:hf-to-gguf:Model ForgeForCausalLM is not supported

oh okay thanks for giving it a try

Sign up or log in to comment