Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,25 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
# 🧮 NanoCalc-1M
|
| 7 |
+
|
| 8 |
+
**NanoCalc-1M** is a ultra-compact, character-level Seq2Seq Transformer based on the T5 architecture, specifically trained to perform arithmetic operations with high precision.
|
| 9 |
+
|
| 10 |
+
* **Architecture:** T5-based Encoder-Decoder
|
| 11 |
+
* **Parameters:** 0.99M
|
| 12 |
+
* **Precision:** Mixed Precision (BF16/FP16)
|
| 13 |
+
* **Vocab:** Character-level (0-9, +, -, *, /, =)
|
| 14 |
+
* **Training Data:** 2,000,000 synthetic samples (3-digit arithmetic)
|
| 15 |
+
* **Max Input Length:** 20 tokens
|
| 16 |
+
* **Performance:** ~97% Accuracy on 4-operation math (Validation Set)
|
| 17 |
+
|
| 18 |
+
#### Performance Chart (Current Progress)
|
| 19 |
+
| Epoch | Training Loss | Val Accuracy | Status |
|
| 20 |
+
| :--- | :--- | :--- | :--- |
|
| 21 |
+
| 1 | 1.1420 | 54.89% | 🔴 Learnt Format |
|
| 22 |
+
| 2 | 0.3931 | 78.79% | 🟡 Learnt Digits |
|
| 23 |
+
| 5 | 0.1638 | 91.91% | 🟢 Learning subtleties |
|
| 24 |
+
| 9 | 0.1051 | 97.15% | 🔵 High Precision |
|
| 25 |
+
| 10 | 0.1004 | 97.73% | 🚀 **Near Perfect** |
|