File size: 2,330 Bytes
0770ced
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import gradio as gr
import torch
import os
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from huggingface_hub import hf_hub_download

# This points to the repo where you uploaded the .pt files in Phase 1
MODEL_REPO = "ibz18/Model_D_weights" 
BASE_MODEL = "csebuetnlp/banglat5"

hf_token = os.environ.get("HF_TOKEN")

print("1. Downloading .pt file...")
abstracter_rl_path = hf_hub_download(
    repo_id=MODEL_REPO, 
    filename="abstracter_rl.pt",
    token=hf_token
)

print("2. Loading tokenizer and base model...")
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
model = AutoModelForSeq2SeqLM.from_pretrained(BASE_MODEL)

print("3. Resizing embeddings...")
model.resize_token_embeddings(len(tokenizer))

print("4. Injecting .pt weights into memory...")
checkpoint = torch.load(abstracter_rl_path, map_location="cpu", weights_only=True)

if isinstance(checkpoint, dict) and "state_dict" in checkpoint:
    state_dict = checkpoint["state_dict"]
else:
    state_dict = checkpoint

model.load_state_dict(state_dict, strict=False)
model.eval()

def generate_summary(text):
    if not text.strip():
        return "Please enter Bangla text."
    
    try:
        inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
        
        with torch.no_grad():
            output_ids = model.generate(
                **inputs, 
                max_new_tokens=128,
                do_sample=False,
                num_beams=2,
                repetition_penalty=2.5,
                early_stopping=True,
                decoder_start_token_id=tokenizer.pad_token_id,
                eos_token_id=tokenizer.eos_token_id,
                pad_token_id=tokenizer.pad_token_id
            )
            
        summary = tokenizer.decode(output_ids[0], skip_special_tokens=True)
        return summary if summary.strip() else "ERROR: Empty string"

    except Exception as e:
        return f"CRASH ERROR: {str(e)}"

# --- INTERFACE WITH NEW NAME ---
demo = gr.Interface(
    fn=generate_summary,
    inputs=gr.Textbox(lines=8, label="Input Bangla Text", placeholder="এখানে আপনার বাংলা টেক্সট দিন..."),
    outputs=gr.Textbox(label="Generated Summary"),
    title="Model_D",
    description="Live testing interface for Model_D"
)

demo.launch()