Commit
•
a119e81
1
Parent(s):
79cc219
add info
Browse files
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
datasets:
|
| 3 |
+
- alpindale/light-novels
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
I am training on this dataset as a test. Well a 6 day test RIP my A6000.
|
| 7 |
+
|
| 8 |
+
```
|
| 9 |
+
{"lora_name": "LightNovels",
|
| 10 |
+
"always_override": false,
|
| 11 |
+
"save_steps": 400.0,
|
| 12 |
+
"micro_batch_size": 4,
|
| 13 |
+
"batch_size": 128,
|
| 14 |
+
"epochs": 1.0,
|
| 15 |
+
"learning_rate": "3e-4",
|
| 16 |
+
"lr_scheduler_type": "constant",
|
| 17 |
+
"lora_rank": 256,
|
| 18 |
+
"lora_alpha": 512,
|
| 19 |
+
"lora_dropout": 0.05,
|
| 20 |
+
"cutoff_len": 2048,
|
| 21 |
+
"dataset": "None",
|
| 22 |
+
"eval_dataset": "None",
|
| 23 |
+
"format": "None",
|
| 24 |
+
"eval_steps": 100.0,
|
| 25 |
+
"raw_text_file": "LightNovels_clean",
|
| 26 |
+
"overlap_len": 512,
|
| 27 |
+
"newline_favor_len": 512,
|
| 28 |
+
"do_shuffle": true,
|
| 29 |
+
"higher_rank_limit": false,
|
| 30 |
+
"warmup_steps": 100.0,
|
| 31 |
+
"optimizer": "adamw_torch"}
|
| 32 |
+
```
|