Update README.md
Browse files
README.md
CHANGED
@@ -2,6 +2,9 @@
|
|
2 |
library_name: peft
|
3 |
datasets:
|
4 |
- HachiML/databricks-dolly-15k-ja-for-peft
|
|
|
|
|
|
|
5 |
---
|
6 |
## JGLUE Score
|
7 |
We evaluated our model using the following JGLUE tasks. Here are the scores:
|
@@ -16,8 +19,34 @@ We evaluated our model using the following JGLUE tasks. Here are the scores:
|
|
16 |
- The JGLUE scores were measured using the following script:
|
17 |
[Stability-AI/lm-evaluation-harness](https://github.com/Stability-AI/lm-evaluation-harness/tree/jp-stable)
|
18 |
|
19 |
-
##
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
The following `bitsandbytes` quantization config was used during training:
|
23 |
- load_in_8bit: False
|
|
|
2 |
library_name: peft
|
3 |
datasets:
|
4 |
- HachiML/databricks-dolly-15k-ja-for-peft
|
5 |
+
language:
|
6 |
+
- en
|
7 |
+
- ja
|
8 |
---
|
9 |
## JGLUE Score
|
10 |
We evaluated our model using the following JGLUE tasks. Here are the scores:
|
|
|
19 |
- The JGLUE scores were measured using the following script:
|
20 |
[Stability-AI/lm-evaluation-harness](https://github.com/Stability-AI/lm-evaluation-harness/tree/jp-stable)
|
21 |
|
22 |
+
## How to use
|
23 |
+
|
24 |
+
```python
|
25 |
+
import torch
|
26 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoTokenizer
|
27 |
+
from peft import PeftModel
|
28 |
+
|
29 |
+
model_name = "meta-llama/Llama-2-13b-hf"
|
30 |
+
bnb_config = BitsAndBytesConfig(
|
31 |
+
load_in_4bit=True,
|
32 |
+
bnb_4bit_use_double_quant=True,
|
33 |
+
bnb_4bit_quant_type="nf4",
|
34 |
+
bnb_4bit_compute_dtype=torch.float16,
|
35 |
+
)
|
36 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
37 |
+
pt_model = AutoModelForCausalLM.from_pretrained(
|
38 |
+
model_name,
|
39 |
+
quantization_config=bnb_config,
|
40 |
+
)
|
41 |
|
42 |
+
peft_name = "HachiML/Llama-2-13b-hf-qlora-dolly-ja-2ep"
|
43 |
+
model = PeftModel.from_pretrained(
|
44 |
+
pt_model,
|
45 |
+
peft_name,
|
46 |
+
)
|
47 |
+
```
|
48 |
+
|
49 |
+
## Training procedure
|
50 |
|
51 |
The following `bitsandbytes` quantization config was used during training:
|
52 |
- load_in_8bit: False
|