---
tags:
- fp8
- vllm
---

# Gemma-2-9B-It-SPPO-Iter3-Q8

## Model Overview
[Gemma-2-9B-It-SPPO-Iter3](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3) quantized to FP8 weights using dynamic activation scheme, ready for inference with vLLM >= 0.5.0.

## Usage and Creation
Produced using [AutoFP8](https://github.com/neuralmagic/AutoFP8/).

```python
from transformers import AutoTokenizer
from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig

pretrained_model_dir = "UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3"
quantized_model_dir = "/quantized/Gemma-2-9B-It-SPPO-Iter3_Q8"

tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir)

quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="dynamic")

model = AutoFP8ForCausalLM.from_pretrained(
    pretrained_model_dir, quantize_config=quantize_config
)

model.save_quantized(quantized_model_dir)
```

## How to run FP8 quantized models
vLLM has full support for FP8 models quantized with this package. Install vLLM with: `pip install vllm>=0.5.0`

Then simply pass the quantized checkpoint directly to vLLM's entrypoints! It will detect the checkpoint format using the quantization_config in the config.json.
```python
from vllm import LLM
model = LLM("tranhoangnguyen03/Gemma-2-9B-It-SPPO-Iter3_Q8")

outputs = model.generate("Once upon a time,")
print(outputs[0].outputs[0].text)
```

## Benchmark Results
||| TBA |||