Sohailhosseini/Ornith-1.5-9B-FP8

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.