Create model_card.md
Browse files- model_card.md +44 -0
model_card.md
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Model Card: Chatbot Caca Retrieval
|
| 2 |
+
|
| 3 |
+
## Model Description
|
| 4 |
+
|
| 5 |
+
Lightweight retrieval-based QA system untuk Bahasa Indonesia.
|
| 6 |
+
|
| 7 |
+
### Training Data
|
| 8 |
+
|
| 9 |
+
- **Source:** datasets-caca-3500
|
| 10 |
+
- **Size:** 3,500 conversational QA pairs
|
| 11 |
+
- **Language:** Indonesian
|
| 12 |
+
- **Format:** User-Assistant conversations
|
| 13 |
+
|
| 14 |
+
### Architecture
|
| 15 |
+
|
| 16 |
+
- **Algorithm:** Hybrid scoring system
|
| 17 |
+
- BM25 (40% weight) - keyword matching
|
| 18 |
+
- TF-IDF + Cosine Similarity (50% weight) - semantic matching
|
| 19 |
+
- Fuzzy String Matching (10% weight) - typo tolerance
|
| 20 |
+
|
| 21 |
+
### Performance
|
| 22 |
+
|
| 23 |
+
| Metric | Value |
|
| 24 |
+
|--------|-------|
|
| 25 |
+
| Model Size | 2.69 MB |
|
| 26 |
+
| Query Latency | <10 ms |
|
| 27 |
+
| Memory Usage | ~5 MB RAM |
|
| 28 |
+
| Paraphrase Accuracy | High |
|
| 29 |
+
|
| 30 |
+
### Limitations
|
| 31 |
+
|
| 32 |
+
- Only works for questions in dataset or similar paraphrases
|
| 33 |
+
- No generative capability
|
| 34 |
+
- Limited to Indonesian language
|
| 35 |
+
|
| 36 |
+
### Ethical Considerations
|
| 37 |
+
|
| 38 |
+
- Responses reflect training data (datasets-caca-3500)
|
| 39 |
+
- Personality may include sarcasm/humor
|
| 40 |
+
- Not suitable for critical applications
|
| 41 |
+
|
| 42 |
+
### License
|
| 43 |
+
|
| 44 |
+
MIT License
|