README.md

metadata

license: apache-2.0
tags:
  - sentiment-analysis
  - distillation
  - small-model
  - smollm
  - nlp
model-index:
  - name: distilled-smollm-sentiment-analyzer
    results:
      - task:
          type: sentiment-analysis
        dataset:
          name: Custom Distillation Dataset
          type: text
        metrics:
          - name: Accuracy
            type: accuracy
            value: ~65% (Relative Accuracy - compared with Teacher model gemma3:12b)

Distilled SmolLM Sentiment Analyzer

This model is a distilled version of a larger sentiment analysis model, fine-tuned on custom datasets using the Hugging Face Transformers library. It is designed for efficient, lightweight sentiment analysis tasks in resource-constrained environments.

✅ Key Features:

• Compact model architecture (SmolLM)
• Distilled for speed and smaller size
• Fine-tuned for sentiment classification tasks
• Supports labels: negative, neutral, positive

---

🔍 Model Details

Model	Distilled SmolLM Sentiment Analyzer
Base Model	SmollM
Task	Sentiment Analysis (3-class: negative, neutral, positive)
Dataset	Custom Yelp Review + Distilled Dataset
Framework	Hugging Face Transformers
Distillation Method	Knowledge Distillation
Accuracy	~75% (Relative Accuracy - compared with Teacher model gemma3:12b)

---

🚀 Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("AhilanPonnusamy/distilled-smollm-sentiment-analyzer")
model = AutoModelForSequenceClassification.from_pretrained("AhilanPonnusamy/distilled-smollm-sentiment-analyzer")

inputs = tokenizer("The movie was amazing!", return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    predicted_class_id = logits.argmax().item()

label_map = {0: "negative", 1: "neutral", 2: "positive"}
print("Predicted sentiment:", label_map[predicted_class_id])