Commit
•
41d97e2
1
Parent(s):
bd00b25
move dataset composition info
Browse files
README.md
CHANGED
|
@@ -68,6 +68,18 @@ This model was trained on:
|
|
| 68 |
- [shahules786/orca-chat](https://huggingface.co/datasets/shahules786/orca-chat)
|
| 69 |
- [togethercomputer/RedPajama-Data-1T-Sample](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)
|
| 70 |
- [atom-in-the-universe/fanfics-10k-50k](https://huggingface.co/datasets/atom-in-the-universe/fanfics-10k-50k)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
|
| 72 |
The dataset [shahules786/orca-chat](https://huggingface.co/datasets/shahules786/orca-chat) combines similar examples of the GPT-4 subset of [ehartford/dolphin](https://huggingface.co/datasets/ehartford/dolphin) to form longer conversations
|
| 73 |
to improve long-context training.
|
|
@@ -105,15 +117,6 @@ llama2_13b_orca_8k:
|
|
| 105 |
type: linear
|
| 106 |
scale: 2
|
| 107 |
datasets:
|
| 108 |
-
# Dataset Composition:
|
| 109 |
-
# Tain (sampled):
|
| 110 |
-
# orca-chat: 100.00% (188842)
|
| 111 |
-
# fanfics: 100.00% (47760)
|
| 112 |
-
# red_pajama: 25.00% (188262)
|
| 113 |
-
# Valid:
|
| 114 |
-
# orca-chat: 5000 (71.43%)
|
| 115 |
-
# fanfics: 1000 (14.29%)
|
| 116 |
-
# red_pajama: 1000 (14.29%)
|
| 117 |
- orca-chat:
|
| 118 |
max_val_set: 5000
|
| 119 |
- fanfics:
|
|
|
|
| 68 |
- [shahules786/orca-chat](https://huggingface.co/datasets/shahules786/orca-chat)
|
| 69 |
- [togethercomputer/RedPajama-Data-1T-Sample](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)
|
| 70 |
- [atom-in-the-universe/fanfics-10k-50k](https://huggingface.co/datasets/atom-in-the-universe/fanfics-10k-50k)
|
| 71 |
+
|
| 72 |
+
```
|
| 73 |
+
Dataset Composition:
|
| 74 |
+
Tain (sampled):
|
| 75 |
+
orca-chat: 188842 (100%)
|
| 76 |
+
fanfics: 47760 (100%)
|
| 77 |
+
red_pajama: 188262 (25%)
|
| 78 |
+
Valid:
|
| 79 |
+
orca-chat: 5000
|
| 80 |
+
fanfics: 1000
|
| 81 |
+
red_pajama: 1000
|
| 82 |
+
```
|
| 83 |
|
| 84 |
The dataset [shahules786/orca-chat](https://huggingface.co/datasets/shahules786/orca-chat) combines similar examples of the GPT-4 subset of [ehartford/dolphin](https://huggingface.co/datasets/ehartford/dolphin) to form longer conversations
|
| 85 |
to improve long-context training.
|
|
|
|
| 117 |
type: linear
|
| 118 |
scale: 2
|
| 119 |
datasets:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 120 |
- orca-chat:
|
| 121 |
max_val_set: 5000
|
| 122 |
- fanfics:
|