About this model

What it is

A 125,848,320-parameter decoder-only transformer (Llama-style), pretrained from scratch with a fresh 16,384-token tokenizer on US case law, SEC filings and educational web text. It is a base model: it continues text and does not follow instructions.

What it isn't

• Not a source of facts. It invents case names, citations and figures fluently.
• Not legal, financial or any other professional advice.
• Not safety-tuned. It may reproduce passages from public filings and opinions, including names.

Get it

• Weights and model card: huggingface.co/Deepanshu027/slm-125m-base
• Inference API (OpenAPI docs): https://mdeepanshu2706--slm-125m-serve-api.modal.run/docs

The API runs on CPU and scales to zero when idle, so the first request after a quiet period takes about 12 seconds. Requests are rate-limited to 30 per minute.

Training data

• US case law: HFforLegal/case-law
• SEC filings: PleIAs/SEC
• Educational web: HuggingFaceFW/fineweb-edu

Excerpts shown on this site come from these public datasets; names of private parties are redacted. The model's license has not been chosen yet.