Spaces:
No application file
No application file
File size: 4,135 Bytes
7e7196e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 | import os
import torch
from PIL import Image
import matplotlib.pyplot as plt
from datasets import load_dataset
from transformers import AutoProcessor, Qwen2VLForConditionalGeneration, BitsAndBytesConfig
from peft import PeftModel
from evaluate import load
# =========================
# CONFIG
# =========================
MODEL_NAME = "Qwen/Qwen2-VL-2B-Instruct"
CHECKPOINT_PATH = "./qlora-vlm"
JSONL_PATH = "0508_clean.jsonl"
NUM_SAMPLES = 3
IMAGE_SIZE = 512
MAX_NEW_TOKENS = 1024
# =========================
# LOAD PROCESSOR
# =========================
processor = AutoProcessor.from_pretrained(MODEL_NAME)
# =========================
# LOAD 4-BIT CONFIG
# =========================
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.float16,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4"
)
# =========================
# LOAD ZERO-SHOT MODEL
# =========================
print("Loading zero-shot model...")
base_model_zero = Qwen2VLForConditionalGeneration.from_pretrained(
MODEL_NAME,
quantization_config=bnb_config,
device_map="auto",
)
base_model_zero.eval()
# =========================
# LOAD FINE-TUNED MODEL
# =========================
print("Loading fine-tuned model...")
base_model = Qwen2VLForConditionalGeneration.from_pretrained(
MODEL_NAME,
quantization_config=bnb_config,
device_map="auto",
)
model = PeftModel.from_pretrained(base_model, CHECKPOINT_PATH)
model.eval()
# =========================
# LOAD DATASET
# =========================
dataset = load_dataset("json", data_files=JSONL_PATH)["train"]
# =========================
# HELPER: GENERATION
# =========================
def build_inputs(image):
prompt = "Convert this document image into structured Markdown."
messages = [
{
"role": "user",
"content": [
{"type": "image"},
{"type": "text", "text": prompt}
]
}
]
text = processor.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
inputs = processor(
text=text,
images=image,
return_tensors="pt"
)
return inputs
def generate(model_obj, image_path):
image = Image.open(image_path).convert("RGB").resize((IMAGE_SIZE, IMAGE_SIZE))
inputs = build_inputs(image).to(model_obj.device)
with torch.no_grad():
output = model_obj.generate(
**inputs,
max_new_tokens=MAX_NEW_TOKENS,
do_sample=False
)
return processor.decode(output[0], skip_special_tokens=True)
# =========================
# OPTIONAL: METRIC
# =========================
def compute_rouge(pred, gt):
rouge = load("rouge")
score = rouge.compute(predictions=[pred], references=[gt])
return score
# =========================
# VISUALIZATION
# =========================
def visualize(sample, title="Sample", save=False):
image_path = sample["image"]
gt = sample["markdown"]
pred_zero = generate(base_model_zero, image_path)
pred_ft = generate(model, image_path)
image = Image.open(image_path).convert("RGB")
# show image
plt.figure(figsize=(8, 5))
plt.imshow(image)
plt.axis("off")
plt.title(title)
if save:
plt.savefig(f"{title.replace(' ', '_')}.png")
plt.show()
print("\n" + "="*100)
print("📌 GROUND TRUTH:\n")
print(gt[:1500])
print("\n" + "-"*100)
print("🤖 ZERO-SHOT OUTPUT:\n")
print(pred_zero[:1500])
print("\n" + "-"*100)
print("🚀 FINE-TUNED OUTPUT:\n")
print(pred_ft[:1500])
# optional metric
rouge_scores = compute_rouge(pred_ft, gt)
print("\n📊 ROUGE (Fine-tuned):", rouge_scores)
print("="*100)
# =========================
# RUN: TRAIN SAMPLES
# =========================
print("\n===== TRAIN SAMPLES =====\n")
for i in range(NUM_SAMPLES):
visualize(dataset[i], title=f"Train Sample {i+1}") |