File size: 3,623 Bytes
28e3194
27375e8
 
900bfe6
 
27375e8
 
900bfe6
 
 
 
28e3194
82e56a5
27375e8
82e56a5
27375e8
82e56a5
 
27375e8
82e56a5
27375e8
82e56a5
 
 
 
 
 
 
27375e8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
305796b
 
 
 
 
 
27375e8
 
 
 
 
 
 
 
305796b
27375e8
 
 
 
 
 
 
 
 
 
 
82e56a5
 
 
 
27375e8
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
---
language:
- en
license: apache-2.0
library_name: peft
tags:
- text-generation-inference
datasets:
- Abirate/english_quotes
pipeline_tag: text-generation
base_model: EleutherAI/gpt-neox-20b
---

# hipnologo/GPT-Neox-20b-QLoRA-FineTune-english_quotes_dataset

## Training procedure

The following `bitsandbytes` quantization config was used during training:

- load_in_8bit: False
- load_in-4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16

## Model description

This model is a fine-tuned version of the `EleutherAI/gpt-neox-20b` model using the QLoRa library and the PEFT library.

#### How to use

The code below performs the following steps:

1. Imports the necessary libraries: `torch` and classes from the `transformers` library.
2. Specifies the `model_id` as "hipnologo/GPT-Neox-20b-QLoRA-FineTune-english_quotes_dataset".
3. Defines a `BitsAndBytesConfig` object named `bnb_config` with the following configuration:
   - `load_in_4bit` set to `True`
   - `bnb_4bit_use_double_quant` set to `True`
   - `bnb_4bit_quant_type` set to "nf4"
   - `bnb_4bit_compute_dtype` set to `torch.bfloat16`
4. Initializes an `AutoTokenizer` object named `tokenizer` by loading the tokenizer for the specified `model_id`.
5. Initializes an `AutoModelForCausalLM` object named `model` by loading the pre-trained model for the specified `model_id` and providing the `quantization_config` as `bnb_config`. The model is loaded on device `cuda:0`.
6. Defines a variable `text` with the value "Twenty years from now".
7. Defines a variable `device` with the value "cuda:0", representing the device on which the model will be executed.
8. Encodes the `text` using the `tokenizer` and converts it to a PyTorch tensor, assigning it to the `inputs` variable. The tensor is moved to the specified `device`.
9. Generates text using the `model.generate` method by passing the `inputs` tensor and setting the `max_new_tokens` parameter to 20. The generated output is assigned to the `outputs` variable.
10. Decodes the `outputs` tensor using the `tokenizer` to obtain the generated text without special tokens, and assigns it to the `generated_text` variable.
11. Prints the `generated_text`.

```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

# Load the base pre-trained model
base_model_id = "EleutherAI/gpt-neox-20b"
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
model = AutoModelForCausalLM.from_pretrained(base_model_id)

# Fine-tuning model
model_id = "hipnologo/GPT-Neox-20b-QLoRA-FineTune-english_quotes_dataset"
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

# Load the fine-tuned model
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map={"":0})

text = "Twenty years from now"
device = "cuda:0"
inputs = tokenizer(text, return_tensors="pt").to(device)

outputs = model.generate(**inputs, max_new_tokens=20)
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_text)
```

### Framework versions

- PEFT 0.4.0.dev0

## Training procedure

- Trainable params: 8650752 
- all params: 10597552128 
- trainable%: 0.08162971878329976


## License

This model is licensed under Apache 2.0. Please see the [LICENSE](https://www.apache.org/licenses/LICENSE-2.0) for more information.