File size: 6,597 Bytes
a9bd396 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 | <!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
โ ๏ธ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
<div style="float: right;">
<div class="flex flex-wrap space-x-1">
<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
<img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
</div>
</div>
# GPT-2[[gpt-2]]
[GPT-2](https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf)๋ GPT์ ํ์ฅ ๋ฒ์ ์ผ๋ก, ์ธ๊ณผ์ ํธ๋์คํฌ๋จธ ์ธ์ด ๋ชจ๋ธ์ด๋ฉฐ, 10๋ฐฐ ๋ ๋ง์ ๋งค๊ฐ๋ณ์์ ํ์ต ๋ฐ์ดํฐ๋ฅผ ๊ฐ์ง๊ณ ์์ต๋๋ค. ์ด ๋ชจ๋ธ์ ์ด์ ์ ๋ชจ๋ ๋จ์ด๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ๋ค์ ๋จ์ด๋ฅผ ์์ธกํ๋๋ก 40GB ๋ฐ์ดํฐ ์ธํธ์์ ์ฌ์ ํ์ต๋์์ต๋๋ค. ์ด๋ฌํ ์ ๊ทผ ๋ฐฉ์์ ํตํด ์ด ๋ชจ๋ธ์ ์ ๋ก์ท ์ค์ ์์ ๋ง์ ๋ค์ด์คํธ๋ฆผ ์์
์ ์ํํ ์ ์๊ฒ ๋์์ต๋๋ค.
๋ชจ๋ธ ์ํคํ
์ฒ๋ ๊ฐ ํ ํฐ์ด ์ด์ ํ ํฐ์๋ง ์ฃผ์๋ฅผ ๊ธฐ์ธ์ผ ์ ์๋ ๋จ๋ฐฉํฅ(์ธ๊ณผ์ ) ์ดํ
์
๋ฉ์ปค๋์ฆ์ ์ฌ์ฉํ๋ฏ๋ก, ํ
์คํธ ์์ฑ ์์
์ ํนํ ํจ๊ณผ์ ์
๋๋ค.
๋ชจ๋ ์๋ณธ GPT-2 ์ฒดํฌํฌ์ธํธ๋ [OpenAI community](https://huggingface.co/openai-community?search_models=gpt) ์กฐ์ง์์ ์ฐพ์ ์ ์์ต๋๋ค.
> [!TIP]
> ์ค๋ฅธ์ชฝ ์ฌ์ด๋๋ฐ์ GPT-2 ๋ชจ๋ธ์ ํด๋ฆญํ์ฌ GPT-2๋ฅผ ๋ค์ํ ์ธ์ด ์์
์ ์ ์ฉํ๋ ๋ ๋ง์ ์์๋ฅผ ํ์ธํ์ธ์.
์๋ ์์๋ [`Pipeline`] ๋๋ [`AutoModel`], ๊ทธ๋ฆฌ๊ณ ๋ช
๋ น์ค์์ GPT-2๋ก ํ
์คํธ๋ฅผ ์์ฑํ๋ ๋ฐฉ๋ฒ์ ๋ณด์ฌ์ค๋๋ค.
<hfoptions id="usage">
<hfoption id="Pipeline">
```py
import torch
from transformers import pipeline
# ํ
์คํธ ์์ฑ์ ์ํ ํ์ดํ๋ผ์ธ ์์ฑ
pipeline = pipeline(task="text-generation", model="openai-community/gpt2", dtype=torch.float16, device=0)
pipeline("Hello, I'm a language model")
```
</hfoption>
<hfoption id="AutoModel">
```py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
# ์ฌ์ ํ์ต๋ ๋ชจ๋ธ๊ณผ ํ ํฌ๋์ด์ ๋ก๋
model = AutoModelForCausalLM.from_pretrained("openai-community/gpt2", dtype=torch.float16, device_map="auto", attn_implementation="sdpa")
tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")
# ์
๋ ฅ ํ
์คํธ๋ฅผ ํ ํฐํํ๊ณ GPU๋ก ์ด๋
input_ids = tokenizer("Hello, I'm a language model", return_tensors="pt").to("cuda")
# ํ
์คํธ ์์ฑ
output = model.generate(**input_ids, cache_implementation="static")
print(tokenizer.decode(output[0], skip_special_tokens=True))
```
</hfoption>
<hfoption id="transformers CLI">
```bash
echo -e "Hello, I'm a language model" | transformers run --task text-generation --model openai-community/gpt2 --device 0
```
</hfoption>
</hfoptions>
`transformers backend`๋ฅผ ์ฌ์ฉํ์ฌ vLLM์ผ๋ก ๋ชจ๋ธ์ ์๋นํ ์๋ ์์ต๋๋ค.
```
vllm serve openai-community/gpt2 --model-imp transformers
```
์์ํ๋ ๊ฐ์ค์น๋ฅผ ๋ ๋ฎ์ ์ ๋ฐ๋๋ก ํํํ์ฌ ๋ํ ๋ชจ๋ธ์ ๋ฉ๋ชจ๋ฆฌ ๋ถ๋ด์ ์ค์
๋๋ค. ์ฌ์ฉํ ์ ์๋ ๋ ๋ง์ ์์ํ ๋ฐฑ์๋์ ๋ํด์๋ [Quantization](../quantization/overview) ๊ฐ์๋ฅผ ์ฐธ์กฐํ์ธ์.
์๋ ์์๋ [bitsandbytes](../quantization/bitsandbytes)๋ฅผ ์ฌ์ฉํ์ฌ ๊ฐ์ค์น๋ง 4๋นํธ๋ก ์์ํํฉ๋๋ค.
```py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
# ์์ํ ์ค์ ๊ตฌ์ฑ
quantization_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype="float16",
bnb_4bit_use_double_quant=True
)
# ์์ํ๋ ๋ชจ๋ธ ๋ก๋
model = AutoModelForCausalLM.from_pretrained(
"openai-community/gpt2-xl",
quantization_config=quantization_config,
device_map="auto"
)
# ํ ํฌ๋์ด์ ๋ก๋ ๋ฐ ํ
์คํธ ์์ฑ
tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2-xl")
inputs = tokenizer("Once upon a time, there was a magical forest", return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
## ์ฐธ๊ณ ์ฌํญ[[notes]]
- GPT-2๋ ์ ๋ ์์น ์๋ฒ ๋ฉ์ ์ฌ์ฉํ๋ฏ๋ก ์
๋ ฅ์ ์ค๋ฅธ์ชฝ์ ํจ๋ฉํ์ธ์.
- GPT-2๋ ์ด์ ์ ๊ณ์ฐ๋ ํค-๊ฐ ์ดํ
์
์์ ์ฌ์ฌ์ฉํ ์ ์์ต๋๋ค. [`GPT2Model.forward`]์ [past_key_values](https://huggingface.co/docs/transformers//en/model_doc/gpt2#transformers.GPT2Model.forward.past_key_values) ๋งค๊ฐ๋ณ์๋ก ์ด ๊ธฐ๋ฅ์ ์ ๊ทผํ์ธ์.
- [Mistral](./mistral)์ ํ์ต ์์ ์ฑ ๊ฐ์ ์ฌํญ์ ์ ์ฉํ๋ ค๋ฉด [scale_attn_by_inverse_layer_idx](https://huggingface.co/docs/transformers/en/model_doc/gpt2#transformers.GPT2Config.scale_attn_by_inverse_layer_idx)์ [reorder_and_upcast_attn](https://huggingface.co/docs/transformers/en/model_doc/gpt2#transformers.GPT2Config.reorder_and_upcast_attn) ๋งค๊ฐ๋ณ์๋ฅผ ํ์ฑํํ์ธ์.
## GPT2Config
[[autodoc]] GPT2Config
## GPT2Tokenizer
[[autodoc]] GPT2Tokenizer
- save_vocabulary
## GPT2TokenizerFast
[[autodoc]] GPT2TokenizerFast
## GPT2 ํน์ ์ถ๋ ฅ[[gpt2-specific-outputs]]
[[autodoc]] models.gpt2.modeling_gpt2.GPT2DoubleHeadsModelOutput
## GPT2Model
[[autodoc]] GPT2Model
- forward
## GPT2LMHeadModel
[[autodoc]] GPT2LMHeadModel
- forward
## GPT2DoubleHeadsModel
[[autodoc]] GPT2DoubleHeadsModel
- forward
## GPT2ForQuestionAnswering
[[autodoc]] GPT2ForQuestionAnswering
- forward
## GPT2ForSequenceClassification
[[autodoc]] GPT2ForSequenceClassification
- forward
## GPT2ForTokenClassification
[[autodoc]] GPT2ForTokenClassification
- forward
|