import torch
from peft import PeftModel
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration

from fovis import FOVISConfig, patch_qwen3vl_model, register_fovis

register_fovis(
    FOVISConfig(fovis_chunk_size=512, fovis_topk=3),
    model_type="qwen3vl",
)

base_model = Qwen3VLForConditionalGeneration.from_pretrained(
    "Qwen/Qwen3-VL-8B-Instruct",
    torch_dtype=torch.bfloat16,
    device_map="cuda:0",
    attn_implementation="flash_attention_2",
)
patch_qwen3vl_model(base_model)

model = PeftModel.from_pretrained(
    base_model,
    "kiki041126/fovis",
    torch_dtype=torch.bfloat16,
)
model = model.merge_and_unload()
model.eval()

processor = AutoProcessor.from_pretrained("kiki041126/fovis")