dmx-mistral-7b-m5

DMX M=7 compressed version of mistralai/Mistral-7B-Instruct-v0.3.

Stats

• Source: mistralai/Mistral-7B-Instruct-v0.3 (FP16)
• Format: DMX BFP M=7 (7 mantissa bits, block floating point)
• File size: 5.01 GB (65% smaller than FP16)
• Quality: Within GPU variance of FP16 (BF16-equivalent precision)

Usage

pip install dmx-compress dmx-runtime

from dmx_runtime import from_dmx_compressed

model = from_dmx_compressed(
    "model.dmx",
    model_id="mistralai/Mistral-7B-Instruct-v0.3"
)

Compressed with dmx-compress.