---
base_model:
- Darkhn/Magistral-Small-2509-Text-Only
---

fp8 w8a8 quant of Darkhn/Magistral-Small-2509-Text-Only b/c vllm seems to take issue with the pixtral vision setup for me.

All thanks to Darkhn/Magistral-Small-2509-Text-Only for uploading the no vision checkpoint.


Recipe:
```python
from transformers import AutoTokenizer, AutoModelForCausalLM

MODEL_ID = "Darkhn/Magistral-Small-2509-Text-Only"

model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier

# Configure the simple PTQ quantization
recipe = QuantizationModifier(
  targets="Linear", scheme="FP8_DYNAMIC", ignore=["lm_head"])

# Apply the quantization algorithm.
oneshot(model=model, recipe=recipe)

# Save the model.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-Dynamic"
model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)
```