Commit
•
0e4fa9a
1
Parent(s):
abe74bf
Update README.md
Browse files
README.md
CHANGED
@@ -1,10 +1,51 @@
|
|
1 |
---
|
2 |
library_name: peft
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
license: apache-2.0
|
4 |
---
|
5 |
-
## Training procedure
|
6 |
|
7 |
-
###
|
8 |
|
|
|
9 |
|
10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
library_name: peft
|
3 |
+
tags:
|
4 |
+
- code
|
5 |
+
- instruct
|
6 |
+
- gpt2
|
7 |
+
datasets:
|
8 |
+
- HuggingFaceH4/no_robots
|
9 |
+
base_model: gpt2
|
10 |
license: apache-2.0
|
11 |
---
|
|
|
12 |
|
13 |
+
### Finetuning Overview:
|
14 |
|
15 |
+
**Model Used:** gpt2
|
16 |
|
17 |
+
**Dataset:** HuggingFaceH4/no_robots
|
18 |
+
|
19 |
+
#### Dataset Insights:
|
20 |
+
|
21 |
+
[No Robots](https://huggingface.co/datasets/HuggingFaceH4/no_robots) is a high-quality dataset of 10,000 instructions and demonstrations created by skilled human annotators. This data can be used for supervised fine-tuning (SFT) to make language models follow instructions better.
|
22 |
+
|
23 |
+
#### Finetuning Details:
|
24 |
+
|
25 |
+
With the utilization of [MonsterAPI](https://monsterapi.ai)'s [LLM finetuner](https://docs.monsterapi.ai/fine-tune-a-large-language-model-llm), this finetuning:
|
26 |
+
|
27 |
+
- Was achieved with great cost-effectiveness.
|
28 |
+
- Completed in a total duration of 3mins 40s for 1 epoch using an A6000 48GB GPU.
|
29 |
+
- Costed `$0.101` for the entire epoch.
|
30 |
+
|
31 |
+
#### Hyperparameters & Additional Details:
|
32 |
+
|
33 |
+
- **Epochs:** 1
|
34 |
+
- **Cost Per Epoch:** $0.101
|
35 |
+
- **Total Finetuning Cost:** $0.101
|
36 |
+
- **Model Path:** gpt2
|
37 |
+
- **Learning Rate:** 0.0002
|
38 |
+
- **Data Split:** 100% train
|
39 |
+
- **Gradient Accumulation Steps:** 4
|
40 |
+
- **lora r:** 32
|
41 |
+
- **lora alpha:** 64
|
42 |
+
|
43 |
+
#### Prompt Structure
|
44 |
+
```
|
45 |
+
<|system|> <|endoftext|> <|user|> [USER PROMPT]<|endoftext|> <|assistant|> [ASSISTANT ANSWER] <|endoftext|>
|
46 |
+
```
|
47 |
+
#### Training loss :
|
48 |
+
|
49 |
+
![training loss](https://cdn-uploads.huggingface.co/production/uploads/63ba46aa0a9866b28cb19a14/9bgb518kFwtDsFtrHzmTu.png)
|
50 |
+
|
51 |
+
license: apache-2.0
|