Update README.md
Browse files
README.md
CHANGED
|
@@ -20,6 +20,12 @@ use full ultrachat and cot data, about 3B tokens
|
|
| 20 |
[@JL-er](https://huggingface.co/JL-er)
|
| 21 |
[@Remixa](https://huggingface.co/Remixa)
|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
# Training details
|
| 24 |
[wandb.ai](https://wandb.ai/one-/one-rwkv-64k)
|
| 25 |
|
|
|
|
| 20 |
[@JL-er](https://huggingface.co/JL-er)
|
| 21 |
[@Remixa](https://huggingface.co/Remixa)
|
| 22 |
|
| 23 |
+
# Design of experiment
|
| 24 |
+
this model lose multi-turn chat ability,cause from using whole ultrachat datasets.
|
| 25 |
+
so i continue tuned multi-turn datasets with 2 aspects
|
| 26 |
+
1.role play
|
| 27 |
+
2.novel multiturn instruction
|
| 28 |
+
|
| 29 |
# Training details
|
| 30 |
[wandb.ai](https://wandb.ai/one-/one-rwkv-64k)
|
| 31 |
|