Ornith-1.5-9B-FP8 / README.md
Sohailhosseini's picture
Fix Python usage example: use llm.chat() with the chat template
d30f70d verified
|
Raw
History Blame Contribute Delete
1.8 kB
metadata
base_model: ornith-ai/Ornith-1.5-9B
base_model_relation: quantized
license: mit
pipeline_tag: image-text-to-text
tags:
  - compressed-tensors
  - fp8
  - quantized
  - vllm

Ornith-1.5-9B - FP8

ornith-ai/Ornith-1.5-9B quantized to FP8 (8-bit weights).

What this is

Near-lossless, no calibration data, and it halves every Linear weight. The safe default when you care about quality and have Ada/Hopper or newer.

Caveat. Needs compute capability >= 8.9 (Ada/Hopper+) to run fast.

Details

Source ornith-ai/Ornith-1.5-9B
Scheme FP8 (8-bit)
Format compressed-tensors
Parameters 9.4B
Size on disk 11.9 GB
Compression 1.58x smaller than the 18.8 GB source
Left unquantized lm_head, re:.*visual.*, re:.*vision_tower.*, re:.*vision_model.*, re:.*vision.*, re:.*multi_modal_projector.*, re:.*merger.*
Quantized on A40
Quantized by Sohailhosseini

Usage

vllm serve Sohailhosseini/Ornith-1.5-9B-FP8 \
  --max-model-len 32768
from vllm import LLM, SamplingParams

if __name__ == "__main__":
    llm = LLM("Sohailhosseini/Ornith-1.5-9B-FP8", max_model_len=32768)
    out = llm.chat(
        [{"role": "user", "content": "What is quantization? Answer in one sentence."}],
        SamplingParams(temperature=0.6, max_tokens=512),
    )
    print(out[0].outputs[0].text)

Provenance

Produced with HF-quantized. recipe.yaml in this repo is the exact modifier stack that was applied, and the scheme, ignored layers and hardware are in the table above.

Licence is inherited from the source model. Quantization does not change what you are permitted to do with the weights.