Original Model Link : https://huggingface.co/common-pile/comma-v0.1-2t

name: comma-v0.1-2t-MLX-Q8
base_model: common-pile/comma-v0.1-2t
license: apache-2.0
tags:
- Comma
- MLX
- EleutherAI
- Llama
- Meta
pipeline_tag: text-generation
size:
- 7440657364
- 7.35GB
datasets: common-pile/comma_v0.1_training_dataset
tasks :
 - text-generation
 - text-to-text
 - text2text-generation
language: en
library_name: mlx
funded_by:
- Mozilla Foundation
- Sutter Hill Ventures
- Natural Sciences and Engineering Research Council of Canada (NSERC-CSE)
hardware_type: 50% 64 H100s / 50% LLNL El Capitan
training_time: 18 days @ 700W
co2_emitted: 17102 lb/7757kg CO2eq
get_started_code: uvx --from mlx-lm mlx_lm.generate --model "darkshapes/comma-v0.1-2t-MLX-Q8" --prompt 'Test prompt'

comma-v0.1-2t-MLX-Q8

Comma v0.1-2T is a 7 billion parameter transformer model trained on openly licensed text from the Common Pile. This repo is a quantized model that lowers memory requirement from 14GB to 7GB, increasing speed and availability at the sacrifice of some accuracy.

MLX is a framework for METAL graphics supported by Apple computers with ARM M-series processors (M1/M2/M3/M4)

Generation using uv https://docs.astral.sh/uv/**:

uvx --from mlx-lm mlx_lm.generate --model “darkshapes/comma-v0.1-2t-ML-Q8" --prompt 'Test prompt'

Generation using pip:

pipx --from mlx-lm mlx_lm.generate --model “darkshapes/comma-v0.1-2t-ML-Q8" --prompt 'Test prompt'