costinstroie/smollm2-135m-clinical-note-GGUF

smollm2-135m-clinical-note-GGUF

GGUF conversion of fahmiaziz/SmolLM2-135M-Instruct-Clinical-Note, an SFT (TRL) fine-tune of HuggingFaceTB/SmolLM2-135M-Instruct for clinical notes. Converted with llama.cpp (b8733); all credit for the fine-tune goes to the original author.

Usage

The embedded chat template is the ChatML template from the source repo.

llama-cli -m smollm2-135m-clinical-note-Q8_0.gguf

Limitations

• —The source repository has no description of its training data or prompt format.
• —In my tests the model writes in a discharge-summary style but invents details that are not in the input (imaging findings, medications, vitals) and emits ___ de-identification placeholders, which suggests training on de-identified hospital notes. This is expected of a 135M-parameter model and is not caused by quantization.
• —Not for clinical use. Outputs are frequently wrong and must never be relied on for patient care.

License

Apache-2.0, inherited from the SmolLM2 base model. The source fine-tune does not state its own license or training data; verify with the original author before commercial use.