maxhirez/mdnaPlus-gemma-3-27b-it

maxhirez/mdna-gemma-3-27b-it

This model maxhirez/mdna-gemma-3-27b-it was converted to MLX format from google/gemma-3-27b-it using mlx-lm version 0.22.4.

Manager's Discussion and Analysis generator.

Trained on data set refined from all (public domain by definition) [SEC EDGAR] (https://www.sec.gov/edgar/search/) quarterly and annual report filings before May 2023 where the company's stock price appreciated in the period from 7 days prior to 7 days after the report filing. Batches of data were separated into prompt/completion pairs with the prompts generated by Llama3.2:3b. In contrast to the [Llama 3 8B instruct version](https://huggingface.co/maxhirez/MDNAPlus-Llama-3-8B-4b, the dataset was adopted to replace numbers such as dollar amounts, percentages, dates, etc. with tags <DOLLARS>, <PERCENTAGE>, <DATE>, <MONTH>, <YEAR>, etc. LoRA executed by MLX-LM.

(In the future, automated means may be implemented to replace tags with prompted values but for now it's manual.)

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("maxhirez/mdna-gemma-3-27b-it")

prompt = '''You are the CEO of a NASDAQ listed entity.  Generate the Q2 financial
          report for a quarter where the enterprise beat expectations by 2% ($4BN)
          but earnings are down from last quarter due to tarif related uncertainty.'''

if tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)

DISCLAIMER

THIS MODEL WAS CREATED FOR EDUCATIONAL PURPOSES. USERS TAKE FULL RESPONSIBILITY FOR THE ACCURACY OF GENERATIONS IF SHARED OR UTILIZED.