Qwen3-32B-FP8-to-BF16 – AI Model by hassanshka | AlphaNeural AI

You can deploy this model and start earning money today!

hassanshka/
Qwen3-32B-FP8-to-BF16

like

0

text-generation

text-generation-inference

endpoints_compatible

Views

No views yet

Qwen3-32B-FP8-to-BF16 (Dequantized Bridge Model)

This is a dequantized BF16 version of Qwen3-32B-FP8. It has been converted from FP8 → BF16 format.

Purpose

This "Bridge Model" serves several purposes:

Fine-tuning base: Use as a starting point for LoRA or full fine-tuning
Research: Study quantization/dequantization quality recovery
Compatibility: Run on hardware without FP8 support

Dequantization Details

Parameter	Value
Source	Qwen3-32B-FP8
Target Dtype	BFloat16
Method	FP8 → BF16 dequantization

Important Notes

⚠️ This model is NOT identical to the original BF16 model.

The dequantization process recovers an approximation:

Some precision loss is expected from the quantization → dequantization roundtrip
Best used for fine-tuning where the loss can be recovered

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "hassanshka/Qwen3-32B-FP8-to-BF16",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("hassanshka/Qwen3-32B-FP8-to-BF16")

# Use for inference or fine-tuning

License

Apache 2.0

Citation

If you use this model, please cite the original Qwen3 model.