Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,30 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
tags:
|
| 6 |
+
- ablations
|
| 7 |
+
- datasets
|
| 8 |
+
- 1m
|
| 9 |
+
- llama
|
| 10 |
+
- custom_tokenizer
|
| 11 |
---
|
| 12 |
+
|
| 13 |
+
# DataBench
|
| 14 |
+
|
| 15 |
+
Rather than training one model, we trained multiple models, each on a different dataset. Everything else, though, was kept the same.
|
| 16 |
+
|
| 17 |
+
## Model Architecture
|
| 18 |
+
|
| 19 |
+
- Base Architecture: `LlamaForCasualLM`
|
| 20 |
+
- Tokenizer: [`Harley-ml/Dillionv2-1.3M`](https://huggingface.co/Harley-ml/Dillionv2-1.3M)
|
| 21 |
+
- Hidden Size: `128`
|
| 22 |
+
- Vocab Size: 2564
|
| 23 |
+
- Number of Layers: `7`
|
| 24 |
+
- Number of Heads: `4`
|
| 25 |
+
- Number of KV Heads: `1`
|
| 26 |
+
- Intermediate Size: `313`
|
| 27 |
+
- Max Position Embeddings: `256`
|
| 28 |
+
- RoPe Theta: `2125.0`
|
| 29 |
+
- Hidden Activation: `silu`
|
| 30 |
+
-
|