Commit
•
faac4b1
1
Parent(s):
6d03d90
Update README.md (#1)
Browse files- Update README.md (b34c57a73130affcefa7d878ce45122a4fa7ad92)
Co-authored-by: Souvik Datta <souvik0306@users.noreply.huggingface.co>
README.md
CHANGED
|
@@ -19,4 +19,56 @@ The following `bitsandbytes` quantization config was used during training:
|
|
| 19 |
### Framework versions
|
| 20 |
|
| 21 |
|
| 22 |
-
- PEFT 0.5.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
### Framework versions
|
| 20 |
|
| 21 |
|
| 22 |
+
- PEFT 0.5.0
|
| 23 |
+
|
| 24 |
+
- ---
|
| 25 |
+
library_name: peft
|
| 26 |
+
tags:
|
| 27 |
+
- code
|
| 28 |
+
- instruct
|
| 29 |
+
- gpt2
|
| 30 |
+
datasets:
|
| 31 |
+
- HuggingFaceH4/no_robots
|
| 32 |
+
base_model: gpt2
|
| 33 |
+
license: apache-2.0
|
| 34 |
+
---
|
| 35 |
+
|
| 36 |
+
### Finetuning Overview:
|
| 37 |
+
|
| 38 |
+
**Model Used:** gpt2
|
| 39 |
+
|
| 40 |
+
**Dataset:** HuggingFaceH4/no_robots
|
| 41 |
+
|
| 42 |
+
#### Dataset Insights:
|
| 43 |
+
|
| 44 |
+
[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.
|
| 45 |
+
|
| 46 |
+
#### Finetuning Details:
|
| 47 |
+
|
| 48 |
+
With the utilization of [MonsterAPI](https://monsterapi.ai)'s [LLM finetuner](https://docs.monsterapi.ai/fine-tune-a-large-language-model-llm), this finetuning:
|
| 49 |
+
|
| 50 |
+
- Was achieved with great cost-effectiveness.
|
| 51 |
+
- Completed in a total duration of 3mins 40s for 1 epoch using an A6000 48GB GPU.
|
| 52 |
+
- Costed `$0.101` for the entire epoch.
|
| 53 |
+
|
| 54 |
+
#### Hyperparameters & Additional Details:
|
| 55 |
+
|
| 56 |
+
- **Epochs:** 1
|
| 57 |
+
- **Cost Per Epoch:** $0.101
|
| 58 |
+
- **Total Finetuning Cost:** $0.101
|
| 59 |
+
- **Model Path:** gpt2
|
| 60 |
+
- **Learning Rate:** 0.0002
|
| 61 |
+
- **Data Split:** 100% train
|
| 62 |
+
- **Gradient Accumulation Steps:** 4
|
| 63 |
+
- **lora r:** 32
|
| 64 |
+
- **lora alpha:** 64
|
| 65 |
+
|
| 66 |
+
#### Prompt Structure
|
| 67 |
+
```
|
| 68 |
+
<|system|> <|endoftext|> <|user|> [USER PROMPT]<|endoftext|> <|assistant|> [ASSISTANT ANSWER] <|endoftext|>
|
| 69 |
+
```
|
| 70 |
+
#### Training loss :
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
license: apache-2.0
|