Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,20 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
datasets:
|
| 4 |
+
- tokyotech-llm/swallow-code-v2
|
| 5 |
+
base_model:
|
| 6 |
+
- meta-llama/Llama-3.2-1B
|
| 7 |
+
- dudeperf3ct/codellm-tokenizer
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
Writeup: https://dudeperf3ct.github.io/projects/train_llm_part2/
|
| 11 |
+
|
| 12 |
+
Repo: https://github.com/dudeperf3ct/minicode-llm/tree/main/codellm_pretrain/torch_titan
|
| 13 |
+
|
| 14 |
+
This contains checkpoints every 5k steps for pretraining run 9.8B tokens using
|
| 15 |
+
|
| 16 |
+
- Custom tokenizer: https://dudeperf3ct.github.io/projects/train_llm_part1/
|
| 17 |
+
- Dataset: [`tokyotech-llm/swallow-code-v2`](https://huggingface.co/datasets/tokyotech-llm/swallow-code-v2)
|
| 18 |
+
- Model Architecture: Llama 3.2 1B (1 billion parameter)
|
| 19 |
+
|
| 20 |
+
The repository contains detailed step on how to run evaluation using PyTorch DCP checkpoints.
|