Jakelolipopp/Qwen3.5-9B-AltText-v4-merged

Qwen3.5-9B-AltText-v4 (Merged Weights)

This repository contains the full, merged bfloat16 safetensors weights of Jakelolipopp/Qwen3.5-9B-AltText-v4-LORA applied to base model unsloth/Qwen3.5-9B.

Model Details

• Base Model: unsloth/Qwen3.5-9B
• LoRA Fine-Tune: Jakelolipopp/Qwen3.5-9B-AltText-v4-LORA
• Data Type: bfloat16
• Format: Hugging Face SafeTensors

Usage with Transformers

import torch
from transformers import AutoProcessor, AutoModelForImageTextToText
from PIL import Image

model_id = "Jakelolipopp/Qwen3.5-9B-AltText-v4-merged"
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

image = Image.open("example.jpg")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": "Provide detailed alternative text describing this image:"}
        ]
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], images=[image], return_tensors="pt").to("cuda")

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=256)

generated_ids = outputs[:, inputs.input_ids.shape[1]:]
print(processor.batch_decode(generated_ids, skip_special_tokens=True)[0])