BartaLens-E2B

Paper Coming Soon

Paper Dataset Models

BartaLens-E2B is a Bangla multimodal headline generation model fine-tuned from gemma-4-E2B-it. Given a Bengali news article (and optionally an accompanying image), it generates a concise, accurate newspaper-style headline. Trained with LoRA on the dipta007/shironam-pro-max dataset with 50% image supervision โ€” the model is robust to both image-present and text-only inputs.

Highlights

  • ROUGE-1: 0.3851 | ROUGE-L: 0.3551 | BLEU-4: 10.43 | BERTScore: 0.8969 on own test split (with images)
  • ROUGE-1: 0.3840 | ROUGE-L: 0.3553 | BLEU-4: 11.02 | BERTScore: 0.8970 on Shironam test split (text-only)
  • Outperforms zero-shot Gemma4-E2B, Qwen3.5-4B, Ministral-3 3B, and prior Bangla headline systems
  • Robust to missing images: trained with 50% text-only supervision so performance doesn't degrade without images

Model Overview

Property Value
Model Type Vision-Language Model (Causal LM + Vision Encoder)
Base Model unsloth/gemma-4-E2B-it
Training SFT + LoRA (r=32, alpha=32)
LoRA Targets all-linear (vision + language + attention + MLP)
Max Sequence Length 4,096 tokens
Language Bengali (เฆฌเฆพเฆ‚เฆฒเฆพ)
Image Supervision 50% (model sees image for half of training samples)
Effective Batch Size 64 (8 per device x 8 gradient accumulation)

Quickstart

from unsloth import FastVisionModel, get_chat_template
import torch
from PIL import Image

model_name = "dipta007/BartaLens-E2B"

model, processor = FastVisionModel.from_pretrained(
    model_name,
    max_seq_length=4096,
    load_in_4bit=False,
    dtype=torch.bfloat16,
)
processor = get_chat_template(processor, "gemma-4")
FastVisionModel.for_inference(model)

INSTRUCTION = (
    "เฆ†เฆชเฆจเฆฟ เฆเฆ•เฆœเฆจ เฆ…เฆญเฆฟเฆœเงเฆž เฆฌเฆพเฆ‚เฆฒเฆพ เฆธเฆ‚เฆฌเฆพเฆฆ เฆธเฆฎเงเฆชเฆพเฆฆเฆ•เฅค เฆจเฆฟเฆšเง‡เฆฐ เฆธเฆ‚เฆฌเฆพเฆฆ เฆจเฆฟเฆฌเฆจเงเฆงเฆŸเฆฟเฆฐ เฆœเฆจเงเฆฏ เฆเฆ•เฆŸเฆฟ เฆ‰เฆชเฆฏเงเฆ•เงเฆค เฆถเฆฟเฆฐเง‹เฆจเฆพเฆฎ เฆคเงˆเฆฐเฆฟ เฆ•เฆฐเงเฆจเฅค\n"
    "\n"
    "เฆจเฆฟเฆฏเฆผเฆฎเฆพเฆฌเฆฒเง€:\n"
    "- เฆจเฆฟเฆฌเฆจเงเฆงเง‡เฆฐ เฆฎเง‚เฆฒ เฆ˜เฆŸเฆจเฆพ เฆ“ เฆคเฆฅเงเฆฏ เฆธเฆ เฆฟเฆ•เฆญเฆพเฆฌเง‡ เฆชเงเฆฐเฆ•เฆพเฆถ เฆ•เฆฐเงเฆจ; เฆ•เฆพเฆฒเงเฆชเฆจเฆฟเฆ• เฆคเฆฅเงเฆฏ เฆฏเง‹เฆ— เฆ•เฆฐเฆฌเง‡เฆจ เฆจเฆพเฅค\n"
    "- เฆธเฆ‚เฆฌเฆพเฆฆเฆชเฆคเงเฆฐเง‡เฆฐ เฆธเฆพเฆงเฆพเฆฐเฆฃ เฆถเฆฟเฆฐเง‹เฆจเฆพเฆฎเง‡เฆฐ เฆถเงˆเฆฒเง€เฆคเง‡, เฆธเฆ‚เฆ•เงเฆทเฆฟเฆชเงเฆค เฆ“ เฆ†เฆ•เฆฐเงเฆทเฆฃเง€เฆฏเฆผเฆญเฆพเฆฌเง‡ เฆฒเฆฟเฆ–เงเฆจเฅค\n"
    "- เฆ‰เฆฆเงเฆงเงƒเฆคเฆฟ เฆšเฆฟเฆนเงเฆจ, เฆฎเฆพเฆฐเงเฆ•เฆกเฆพเฆ‰เฆจ, เฆ‡เฆฎเง‹เฆœเฆฟ, เฆคเฆพเฆฒเฆฟเฆ•เฆพ เฆšเฆฟเฆนเงเฆจ (*, -), เฆ…เฆฅเฆฌเฆพ เฆจเฆฎเงเฆฌเฆฐ (เงง., 1.) เฆฌเงเฆฏเฆฌเฆนเฆพเฆฐ เฆ•เฆฐเฆฌเง‡เฆจ เฆจเฆพเฅค\n"
    "- เฆ•เง‹เฆจเง‹ เฆญเง‚เฆฎเฆฟเฆ•เฆพ, เฆฌเงเฆฏเฆพเฆ–เงเฆฏเฆพ, เฆเฆ•เฆพเฆงเฆฟเฆ• เฆฌเฆฟเฆ•เฆฒเงเฆช เฆฌเฆพ เฆ…เฆคเฆฟเฆฐเฆฟเฆ•เงเฆค เฆฎเฆจเงเฆคเฆฌเงเฆฏ เฆฏเง‹เฆ— เฆ•เฆฐเฆฌเง‡เฆจ เฆจเฆพเฅค\n"
    "- เฆถเงเฆงเง เฆถเฆฟเฆฐเง‹เฆจเฆพเฆฎเฆŸเฆฟ เฆเฆ•เฆŸเฆฟ เฆฒเฆพเฆ‡เฆจเง‡ เฆฒเฆฟเฆ–เงเฆจ, เฆ…เฆจเงเฆฏ เฆ•เฆฟเฆ›เง เฆจเฆฏเฆผเฅค\n"
)


def generate_headline(article: str, image: Image.Image | None = None):
    """Generate a Bengali headline for a news article."""
    user_text = f"{INSTRUCTION}\nเฆจเฆฟเฆฌเฆจเงเฆง:\n{article}\n\nเฆถเฆฟเฆฐเง‹เฆจเฆพเฆฎ:"

    if image is None:
        image = Image.new("RGB", (224, 224), color="black")

    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image"},
                {"type": "text", "text": user_text},
            ],
        }
    ]

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

    with torch.no_grad():
        out = model.generate(
            **inputs,
            max_new_tokens=64,
            use_cache=True,
            do_sample=False,
        )
    headline = processor.tokenizer.decode(
        out[0][inputs["input_ids"].shape[1]:],
        skip_special_tokens=True,
    ).strip()
    return headline


# Usage: text-only
article = """เฆฌเฆพเฆ‚เฆฒเฆพเฆฆเง‡เฆถ เฆœเฆพเฆคเง€เฆฏเฆผ เฆ•เงเฆฐเฆฟเฆ•เง‡เฆŸ เฆฆเฆฒเง‡เฆฐ เฆ…เฆงเฆฟเฆจเฆพเฆฏเฆผเฆ• เฆจเฆพเฆœเฆฎเงเฆฒ เฆนเง‹เฆธเง‡เฆจ เฆถเฆพเฆจเงเฆค เฆ†เฆœ 
เฆธเฆ‚เฆฌเฆพเฆฆ เฆธเฆฎเงเฆฎเง‡เฆฒเฆจเง‡ เฆœเฆพเฆจเฆฟเฆฏเฆผเง‡เฆ›เง‡เฆจ, เฆฆเฆฒ เฆ†เฆ—เฆพเฆฎเง€ เฆŸเง‡เฆธเงเฆŸ เฆธเฆฟเฆฐเฆฟเฆœเง‡เฆฐ เฆœเฆจเงเฆฏ เฆชเงเฆฐเง‹เฆชเงเฆฐเฆฟ เฆชเงเฆฐเฆธเงเฆคเงเฆคเฅค 
เฆคเฆฟเฆจเฆฟ เฆฌเฆฒเง‡เฆจ, เฆคเฆฐเงเฆฃ เฆ–เง‡เฆฒเง‹เฆฏเฆผเฆพเฆกเฆผเฆฆเง‡เฆฐ เฆชเฆพเฆฐเฆซเฆฐเฆฎเงเฆฏเฆพเฆจเงเฆธ เฆฆเฆฒเง‡เฆฐ เฆถเฆ•เงเฆคเฆฟ เฆฌเฆพเฆกเฆผเฆฟเฆฏเฆผเง‡เฆ›เง‡เฅค"""

headline = generate_headline(article)
print(headline)

# Usage: with image
# image = Image.open("news_image.jpg")
# headline = generate_headline(article, image=image)

Performance

Comparison with zero-shot baselines (Own eval, n=15,000)

Model ROUGE-1 ROUGE-2 ROUGE-L BLEU-4 BERTScore METEOR
Qwen3.5-0.8B 0.1546 0.0541 0.1429 2.14 0.8373 0.1002
Qwen3.5-2B 0.2029 0.0613 0.1821 1.83 0.8498 0.1191
Ministral-3 3B 0.2892 0.0903 0.2445 2.48 0.8725 0.1625
Qwen3.5-4B 0.2984 0.1065 0.2618 3.98 0.8729 0.1924
Gemma4-E2B (zero-shot) 0.3484 0.1507 0.3127 6.37 0.8874 0.2531
BartaLens-E2B (ours) 0.3851 0.1807 0.3551 10.43 0.8969 0.2644

Cross-dataset generalization (Shironam eval, text-only, n=15,012)

Model ROUGE-1 ROUGE-2 ROUGE-L BLEU-4 BERTScore METEOR
Gemma4-E2B (zero-shot) 0.3535 0.1526 0.3190 6.36 0.8883 0.2620
BartaLens-E2B (ours) 0.3840 0.1809 0.3553 11.02 0.8970 0.2590

Training Details

  • Dataset: dipta007/shironam-pro-max (train split)
  • Supervision: 50% of training samples include the news image; 50% use a black placeholder (text-only). This trains the model to be robust when no image is available.
  • Early stopping: patience=3 on validation loss, eval every 200 steps
  • Optimizer: AdamW, LR=5e-4, cosine schedule, warmup 5%
  • Hardware: NVIDIA L40S (46 GB)
  • Metrics: csebuetnlp multilingual ROUGE (Bengali stemmer), HF BLEU, BanglaBERT BERTScore, METEOR

Intended Use

  • In-scope: generating concise Bengali news headlines from article text (optionally with an image), headline suggestion for editors, summarization benchmarks.
  • Out-of-scope: generating headlines in other languages, creative/clickbait headline generation, summarization of non-news content.

Citation


License

Released under the Gemma License.

Downloads last month
18
Safetensors
Model size
5B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for dipta007/BartaLens-E2B

Adapter
(64)
this model
Adapters
2 models

Collection including dipta007/BartaLens-E2B