File size: 8,100 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 | <!--Copyright 2023 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 contains specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
*์ด ๋ชจ๋ธ์ 2023๋
8์ 24์ผ์ ๊ณต๊ฐ๋์์ผ๋ฉฐ, 2023๋
8์ 25์ผ์ Hugging Face Transformers์ ์ถ๊ฐ๋์์ต๋๋ค.*
<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">
">
</div>
</div>
# CodeLlama[[codellama]]
[Code Llama](https://huggingface.co/papers/2308.12950)๋ ์ฝ๋ฉ ์์
์ ํนํ๋ ๋๊ท๋ชจ ์ธ์ด ๋ชจ๋ธ ๊ณ์ด๋ก, [Llama 2](./llama2)๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ๊ฐ๋ฐ๋์์ต๋๋ค. ์ผ๋ฐ์ ์ธ ์ฝ๋, Python ํนํ, ๋ช
๋ น์ด(์ง์) ๊ธฐ๋ฐ ๋ณํ ๋ฑ ๋ค์ํ ๋ฒ์ ์ผ๋ก ์ ๊ณต๋๋ฉฐ, ๋ชจ๋ 7B, 13B, 34B, 70B ๋งค๊ฐ๋ณ์ ํฌ๊ธฐ๋ก ์ฌ์ฉํ ์ ์์ต๋๋ค. Code Llama ๋ชจ๋ธ์ ์ฝ๋๋ฅผ ์์ฑํ๊ณ ์ค๋ช
ํ๋ฉฐ, ์ฝ๋์ ๋๋ฝ๋ ๋ถ๋ถ์ ์ฑ์ธ ์๋ ์์ต๋๋ค. ์ด๋ฅผ ์ธํ๋ง(infilling)์ด๋ผ๊ณ ํฉ๋๋ค. 16K ํ ํฐ ๊ธธ์ด๋ก ํ๋ จ๋์์ง๋ง, ์ต๋ 100K ํ ํฐ๊น์ง ์์ ์ ์ผ๋ก ์์ฑํ๋ฉฐ ๊ธด ์ปจํ
์คํธ๋ ์ฒ๋ฆฌํ ์ ์์ต๋๋ค.
[Code Llama](https://huggingface.co/collections/meta-llama/code-llama-family-661da32d0a9d678b6f55b933) ์ปฌ๋ ์
์์ ๋ชจ๋ ์๋ณธ Code Llama ์ฒดํฌํฌ์ธํธ๋ฅผ ์ฐพ์ ์ ์์ต๋๋ค.
> [!TIP]
> ๋ค์ํ ์ฝ๋ฉ ์์
์ Code Llama๋ฅผ ์ ์ฉํ๋ ๋ ๋ง์ ์์๋ฅผ ๋ณด๋ ค๋ฉด ์ค๋ฅธ์ชฝ ์ฌ์ด๋๋ฐ์ Code Llama ๋ชจ๋ธ์ ํด๋ฆญํ์ธ์.
์๋ ์์๋ [`Pipeline`], [`AutoModel`], ๊ทธ๋ฆฌ๊ณ ๋ช
๋ น์ค์์ ์ฝ๋๋ฅผ ์์ฑํ๋ ๋ฐฉ๋ฒ์ ๋ณด์ฌ์ค๋๋ค.
<hfoptions id="usage">
<hfoption id="Pipeline">
```py
import torch
from transformers import pipeline
pipe = pipeline(
"text-generation",
model="meta-llama/CodeLlama-7b-hf",
torch_dtype=torch.float16,
device_map=0
)
# ๊ธฐ๋ณธ ์ฝ๋ ์์ฑ
result = pipe("# Function to calculate the factorial of a number\ndef factorial(n):", max_new_tokens=256)
print(result[0]['generated_text'])
# ์ธํ๋ง
infill_result = pipe("def remove_non_ascii(s: str) -> str:\n \"\"\" <FILL_ME>\n return result", max_new_tokens=200)
print(infill_result[0]['generated_text'])
```
</hfoption>
<hfoption id="AutoModel">
```py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("meta-llama/CodeLlama-7b-hf")
model = AutoModelForCausalLM.from_pretrained(
"meta-llama/CodeLlama-7b-hf",
torch_dtype=torch.float16,
device_map="auto",
attn_implementation="sdpa"
)
# ๊ธฐ๋ณธ ์ฝ๋ ์์ฑ
prompt = "# Function to calculate the factorial of a number\ndef factorial(n):"
input_ids = tokenizer(prompt, return_tensors="pt").to("cuda")
output = model.generate(
**input_ids,
max_new_tokens=256,
cache_implementation="static"
)
print(tokenizer.decode(output[0], skip_special_tokens=True))
# ์ธํ๋ง
infill_prompt = "def remove_non_ascii(s: str) -> str:\n \"\"\" <FILL_ME>\n return result"
input_ids = tokenizer(infill_prompt, return_tensors="pt").to(model.device)
filled_output = model.generate(**input_ids, max_new_tokens=200)
filled_text = tokenizer.decode(filled_output[0], skip_special_tokens=True)
print(filled_text)
```
</hfoption>
<hfoption id="transformers CLI">
```bash
echo -e "# Function to calculate the factorial of a number\ndef factorial(n):" | transformers run --task text-generation --model meta-llama/CodeLlama-7b-hf --device 0
```
</hfoption>
</hfoptions>
์์ํ๋ ๊ฐ์ค์น๋ฅผ ๋ ๋ฎ์ ์ ๋ฐ๋๋ก ํํํ์ฌ ๋๊ท๋ชจ ๋ชจ๋ธ์ ๋ฉ๋ชจ๋ฆฌ ๋ถ๋ด์ ์ค์
๋๋ค. ๋ ๋ง์ ์ฌ์ฉ ๊ฐ๋ฅํ ์์ํ ๋ฐฑ์๋๋ [์์ํ](../quantization/overview) ๊ฐ์๋ฅผ ์ฐธ์กฐํ์ธ์.
์๋ ์์๋ [bitsandbytes](../quantization/bitsandbytes)๋ฅผ ์ฌ์ฉํ์ฌ ๊ฐ์ค์น๋ฅผ 4๋นํธ๋ก๋ง ์์ํํฉ๋๋ค.
```py
# bitsandbytes๋ฅผ ์ค์นํฉ๋๋ค.
import torch
from transformers import AutoModelForCausalLM, CodeLlamaTokenizer, BitsAndBytesConfig
bnb_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.bfloat16, bnb_4bit_quant_type="nf4", bnb_4bit_use_double_quant=True)
tokenizer = CodeLlamaTokenizer.from_pretrained("meta-llama/CodeLlama-34b-hf")
model = AutoModelForCausalLM.from_pretrained(
"meta-llama/CodeLlama-34b-hf",
torch_dtype=torch.bfloat16,
device_map="auto",
quantization_config=bnb_config
)
prompt = "# Write a Python function to check if a string is a palindrome\ndef is_palindrome(s):"
input_ids = tokenizer(prompt, return_tensors="pt").to("cuda")
output = model.generate(**input_ids, max_new_tokens=200, cache_implementation="static")
print(tokenizer.decode(output[0], skip_special_tokens=True))
```
[AttentionMaskVisualizer](https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139)๋ฅผ ์ฌ์ฉํ๋ฉด ๋ชจ๋ธ์ด ์ด๋ค ํ ํฐ์ ์ฃผ์๋ฅผ ๊ธฐ์ธ์ผ ์ ์๊ณ ๊ธฐ์ธ์ผ ์ ์๋์ง๋ฅผ ๋ ์ ์ดํดํ ์ ์์ต๋๋ค.
```py
from transformers.utils.attention_visualizer import AttentionMaskVisualizer
visualizer = AttentionMaskVisualizer("meta-llama/CodeLlama-7b-hf")
visualizer("""def func(a, b):
return a + b""")
```
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/codellama-attn-mask.png"/>
</div>
## ์ฐธ๊ณ ์ฌํญ[[notes]]
- ์ธํ๋ง ๊ธฐ๋ฅ์ 7B ๋ฐ 13B ๊ธฐ๋ฐ ๋ชจ๋ธ์์๋ง ์ฌ์ฉํ ์ ์์ผ๋ฉฐ, Python, Instruct, 34B ๋๋ 70B ๋ชจ๋ธ์์๋ ์ฌ์ฉํ ์ ์์ต๋๋ค.
- ์ฝ๋๋ฅผ ์ฑ์ ๋ฃ๊ณ ์ถ์ ๋ถ๋ถ์ `<FILL_ME>` ํ ํฐ์ ์ฌ์ฉํ์ธ์. ํ ํฌ๋์ด์ ๋ ์ด ํ ํฐ์ ๋ถํ ํ์ฌ [์๋ณธ ํ๋ จ ํจํด](https://github.com/facebookresearch/codellama/blob/cb51c14ec761370ba2e2bc351374a79265d0465e/llama/generation.py#L402) ์ ๋ฐ๋ฅด๋ ์
๋ ฅ ๋ฌธ์์ด๋ก ๋ณํํฉ๋๋ค. ์ด๋ ์ง์ ํจํด์ ์ค๋นํ๋ ๊ฒ๋ณด๋ค ๋ ์์ ์ ์
๋๋ค.
```py
from transformers import LlamaForCausalLM, CodeLlamaTokenizer
tokenizer = CodeLlamaTokenizer.from_pretrained("meta-llama/CodeLlama-7b-hf")
model = LlamaForCausalLM.from_pretrained("meta-llama/CodeLlama-7b-hf")
PROMPT = '''def remove_non_ascii(s: str) -> str:
""" <FILL_ME>
return result
'''
input_ids = tokenizer(PROMPT, return_tensors="pt")["input_ids"]
generated_ids = model.generate(input_ids, max_new_tokens=128)
filling = tokenizer.batch_decode(generated_ids[:, input_ids.shape[1]:], skip_special_tokens = True)[0]
print(PROMPT.replace("<FILL_ME>", filling))
```
- ์ถ๊ฐ ํ๋ จ์ด๋ ๋ฏธ์ธ ์กฐ์ ์๋ `bfloat16`์ ์ฌ์ฉํ๊ณ ์ถ๋ก ์๋ `float16`์ ์ฌ์ฉํ์ธ์.
- `BOS` ๋ฌธ์๋ ์ ๋์ฌ๋ ์ ๋ฏธ์ฌ๋ฅผ ์ธ์ฝ๋ฉํ ๋ ์ธํ๋ง ์์
์ ์ฌ์ฉ๋์ง ์์ผ๋ฉฐ, ๊ฐ ํ๋กฌํํธ์ ๋งจ ์์์๋ง ์ฌ์ฉ๋ฉ๋๋ค.
- ํ ํฌ๋์ด์ ๋ [SentencePiece](https://github.com/google/sentencepiece)๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ํ๋ byte-pair ์ธ์ฝ๋ฉ ๋ชจ๋ธ์
๋๋ค. ๋์ฝ๋ฉ ๊ณผ์ ์์ ์ฒซ ๋ฒ์งธ ํ ํฐ์ด ๋จ์ด์ ์์์ธ ๊ฒฝ์ฐ(์๋ฅผ ๋ค์ด "Banana"), ํ ํฌ๋์ด์ ๋ ๋ฌธ์์ด์ ์ ๋์ฌ ๊ณต๋ฐฑ์ ์ถ๊ฐํ์ง ์์ต๋๋ค.
## CodeLlamaTokenizer
[[autodoc]] CodeLlamaTokenizer
- get_special_tokens_mask
- save_vocabulary
## CodeLlamaTokenizerFast
[[autodoc]] CodeLlamaTokenizerFast
- get_special_tokens_mask
- update_post_processor
- save_vocabulary
|