This repository contains full BF16 weights for a Widget2Code SFT model based on Qwen/Qwen3-VL-32B-Instruct. A rank-32 LoRA adapter (lora_alpha=64) was merged into the base model with PEFT merge_and_unload(safe_merge=True).

Artifact details

Base model: Qwen/Qwen3-VL-32B-Instruct
Format: Hugging Face Transformers safetensors
Saved parameters: 33,357,390,064
Weight dtype: BF16
Adapter training dataset name: widget2code_sft

merge_provenance.json records the base revision, adapter configuration, adapter hashes, merge environment, and audited tensor inventory. The saved weights contain no remaining LoRA tensors.

Usage

python

from transformers import AutoModelForImageTextToText, AutoProcessor

model_id = "Djanghao/widget2code-qwen3-vl-32b-size-tools-merged-bf16"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
    model_id,
    dtype="auto",
    device_map="auto",
)

Limitations

No standalone benchmark results are included in this model card. Users should evaluate the model on their target screenshot-to-code distribution before deployment.