File size: 8,293 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
<!--Copyright 2024 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.

โš ๏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="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>

# Jamba[[jamba]]

[Jamba](https://huggingface.co/papers/2403.19887)๋Š” Transformer์™€ Mamba ๊ธฐ๋ฐ˜์˜ ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ์ „๋ฌธ๊ฐ€ ํ˜ผํ•ฉ(MoE) ์–ธ์–ด ๋ชจ๋ธ๋กœ, ์ด ๋งค๊ฐœ๋ณ€์ˆ˜ ์ˆ˜๋Š” 52B์—์„œ 398B๊นŒ์ง€ ๋‹ค์–‘ํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ชจ๋ธ์€ Transformer ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ๊ณผ Mamba์™€ ๊ฐ™์€ ์ƒํƒœ ๊ณต๊ฐ„ ๋ชจ๋ธ์˜ ํšจ์œจ์„ฑ ๋ฐ ๊ธด ์ปจํ…์ŠคํŠธ ์ฒ˜๋ฆฌ ๋Šฅ๋ ฅ(256K ํ† ํฐ)์„ ๋ชจ๋‘ ํ™œ์šฉํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค.

Jamba์˜ ์•„ํ‚คํ…์ฒ˜๋Š” ๋ธ”๋ก๊ณผ ๋ ˆ์ด์–ด ๊ธฐ๋ฐ˜ ๊ตฌ์กฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Transformer์™€ Mamba ์•„ํ‚คํ…์ฒ˜๋ฅผ ํ†ตํ•ฉํ•  ์ˆ˜ ์žˆ๋„๋ก ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ Jamba ๋ธ”๋ก์€ ์–ดํ…์…˜ ๋ ˆ์ด์–ด ๋˜๋Š” Mamba ๋ ˆ์ด์–ด ์ค‘ ํ•˜๋‚˜์™€ ๊ทธ ๋’ค๋ฅผ ์ž‡๋Š” ๋‹ค์ธต ํผ์…‰ํŠธ๋ก (MLP)์œผ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. Transformer ๋ ˆ์ด์–ด๋Š” 8๊ฐœ์˜ ๋ ˆ์ด์–ด ์ค‘ ํ•˜๋‚˜์˜ ๋น„์œจ๋กœ ์ฃผ๊ธฐ์ ์œผ๋กœ ๋ฐฐ์น˜๋ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ชจ๋ธ ์šฉ๋Ÿ‰์„ ํ™•์žฅํ•˜๊ธฐ ์œ„ํ•ด MoE ๋ ˆ์ด์–ด๊ฐ€ ํ˜ผํ•ฉ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.

๋ชจ๋“  ์›๋ณธ Jamba ์ฒดํฌํฌ์ธํŠธ๋Š” [AI21](https://huggingface.co/ai21labs) ์กฐ์ง์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

> [!TIP]
> ์˜ค๋ฅธ์ชฝ ์‚ฌ์ด๋“œ๋ฐ”์— ์žˆ๋Š” Jamba ๋ชจ๋ธ์„ ๋ˆ„๋ฅด๋ฉด ๋‹ค์–‘ํ•œ ์–ธ์–ด ์ž‘์—…์— Jamba๋ฅผ ์ ์šฉํ•˜๋Š” ์˜ˆ์ œ๋ฅผ ๋” ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

์•„๋ž˜ ์˜ˆ์ œ๋Š” [`Pipeline`]๊ณผ [`AutoModel`], ๊ทธ๋ฆฌ๊ณ  ์ปค๋งจ๋“œ๋ผ์ธ์„ ํ†ตํ•ด ํ…์ŠคํŠธ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

<hfoptions id="usage">
<hfoption id="Pipeline">

```py
# ์ตœ์ ํ™”๋œ Mamba ๊ตฌํ˜„ ์„ค์น˜
# !pip install mamba-ssm causal-conv1d>=1.2.0
import torch
from transformers import pipeline

pipeline = pipeline(
    task="text-generation",
    model="ai21labs/AI21-Jamba-Mini-1.6",
    dtype=torch.float16,
    device=0
)
pipeline("Plants create energy through a process known as")
```

</hfoption>
<hfoption id="AutoModel">

```py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained(
    "ai21labs/AI21-Jamba-Large-1.6",
)
model = AutoModelForCausalLM.from_pretrained(
    "ai21labs/AI21-Jamba-Large-1.6",
    dtype=torch.float16,
    device_map="auto",
    attn_implementation="sdpa"
)
input_ids = tokenizer("Plants create energy through a process known as", return_tensors="pt").to("cuda")

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 "Plants create energy through a process known as" | transformers run --task text-generation --model ai21labs/AI21-Jamba-Mini-1.6 --device 0
```

</hfoption>
</hfoptions>

์–‘์žํ™”๋Š” ๊ฐ€์ค‘์น˜๋ฅผ ๋” ๋‚ฎ์€ ์ •๋ฐ€๋„๋กœ ํ‘œํ˜„ํ•˜์—ฌ ๋Œ€๊ทœ๋ชจ ๋ชจ๋ธ์˜ ๋ฉ”๋ชจ๋ฆฌ ๋ถ€๋‹ด์„ ์ค„์—ฌ์ค๋‹ˆ๋‹ค. ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๋‹ค์–‘ํ•œ ์–‘์žํ™” ๋ฐฑ์—”๋“œ์— ๋Œ€ํ•ด์„œ๋Š” [Quantization](../quantization/overview)๋ฅผ ์ฐธ๊ณ ํ•˜์„ธ์š”.

์•„๋ž˜ ์˜ˆ์ œ๋Š” [bitsandbytes](../quantization/bitsandbytes)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ€์ค‘์น˜๋งŒ 8๋น„ํŠธ๋กœ ์–‘์žํ™”ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

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

quantization_config = BitsAndBytesConfig(load_in_8bit=True,
                                         llm_int8_skip_modules=["mamba"])

# ๋ชจ๋ธ์„ 8๊ฐœ์˜ GPU์— ๊ณ ๋ฅด๊ฒŒ ๋ถ„์‚ฐ์‹œํ‚ค๊ธฐ ์œ„ํ•œ ๋””๋ฐ”์ด์Šค ๋งต
device_map = {'model.embed_tokens': 0, 'model.layers.0': 0, 'model.layers.1': 0, 'model.layers.2': 0, 'model.layers.3': 0, 'model.layers.4': 0, 'model.layers.5': 0, 'model.layers.6': 0, 'model.layers.7': 0, 'model.layers.8': 0, 'model.layers.9': 1, 'model.layers.10': 1, 'model.layers.11': 1, 'model.layers.12': 1, 'model.layers.13': 1, 'model.layers.14': 1, 'model.layers.15': 1, 'model.layers.16': 1, 'model.layers.17': 1, 'model.layers.18': 2, 'model.layers.19': 2, 'model.layers.20': 2, 'model.layers.21': 2, 'model.layers.22': 2, 'model.layers.23': 2, 'model.layers.24': 2, 'model.layers.25': 2, 'model.layers.26': 2, 'model.layers.27': 3, 'model.layers.28': 3, 'model.layers.29': 3, 'model.layers.30': 3, 'model.layers.31': 3, 'model.layers.32': 3, 'model.layers.33': 3, 'model.layers.34': 3, 'model.layers.35': 3, 'model.layers.36': 4, 'model.layers.37': 4, 'model.layers.38': 4, 'model.layers.39': 4, 'model.layers.40': 4, 'model.layers.41': 4, 'model.layers.42': 4, 'model.layers.43': 4, 'model.layers.44': 4, 'model.layers.45': 5, 'model.layers.46': 5, 'model.layers.47': 5, 'model.layers.48': 5, 'model.layers.49': 5, 'model.layers.50': 5, 'model.layers.51': 5, 'model.layers.52': 5, 'model.layers.53': 5, 'model.layers.54': 6, 'model.layers.55': 6, 'model.layers.56': 6, 'model.layers.57': 6, 'model.layers.58': 6, 'model.layers.59': 6, 'model.layers.60': 6, 'model.layers.61': 6, 'model.layers.62': 6, 'model.layers.63': 7, 'model.layers.64': 7, 'model.layers.65': 7, 'model.layers.66': 7, 'model.layers.67': 7, 'model.layers.68': 7, 'model.layers.69': 7, 'model.layers.70': 7, 'model.layers.71': 7, 'model.final_layernorm': 7, 'lm_head': 7}
model = AutoModelForCausalLM.from_pretrained("ai21labs/AI21-Jamba-Large-1.6",
                                             dtype=torch.bfloat16,
                    attn_implementation="flash_attention_2",
                                             quantization_config=quantization_config,
                                             device_map=device_map)

tokenizer = AutoTokenizer.from_pretrained("ai21labs/AI21-Jamba-Large-1.6")

messages = [
   {"role": "system", "content": "You are an ancient oracle who speaks in cryptic but wise phrases, always hinting at deeper meanings."},
   {"role": "user", "content": "Hello!"},
]

input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors='pt').to(model.device)

outputs = model.generate(input_ids, max_new_tokens=216)

# ์ถœ๋ ฅ ๋””์ฝ”๋”ฉ
conversation = tokenizer.decode(outputs[0], skip_special_tokens=True)

# ์–ด์‹œ์Šคํ„ดํŠธ์˜ ์‘๋‹ต๋งŒ ์ถ”์ถœ
assistant_response = conversation.split(messages[-1]['content'])[1].strip()
print(assistant_response)
# ์ถœ๋ ฅ: Seek and you shall find. The path is winding, but the journey is enlightening. What wisdom do you seek from the ancient echoes?
```

## ์ฐธ๊ณ [[notes]]

- ๋ชจ๋ธ ์„ฑ๋Šฅ ์ €ํ•˜๋ฅผ ๋ฐฉ์ง€ํ•˜๊ธฐ ์œ„ํ•ด Mamba ๋ธ”๋ก์€ ์–‘์žํ™”ํ•˜์ง€ ๋งˆ์„ธ์š”.
- ์ตœ์ ํ™”๋œ Mamba ์ปค๋„ ์—†์ด Mamba๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ์ง€์—ฐ ์‹œ๊ฐ„์ด ํฌ๊ฒŒ ์ฆ๊ฐ€ํ•˜๋ฏ€๋กœ ๊ถŒ์žฅ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๊ทธ๋ž˜๋„ ์ปค๋„ ์—†์ด Mamba๋ฅผ ์‚ฌ์šฉํ•˜๊ณ ์ž ํ•œ๋‹ค๋ฉด [`~AutoModel.from_pretrained`]์—์„œ `use_mamba_kernels=False`๋กœ ์„ค์ •ํ•˜์„ธ์š”.

  ```py
  import torch
  from transformers import AutoModelForCausalLM
  model = AutoModelForCausalLM.from_pretrained("ai21labs/AI21-Jamba-1.5-Large",
                                               use_mamba_kernels=False)
  ```

## JambaConfig[[transformers.JambaConfig]]

[[autodoc]] JambaConfig


## JambaModel[[transformers.JambaModel]]

[[autodoc]] JambaModel
    - forward


## JambaForCausalLM[[transformers.JambaForCausalLM]]

[[autodoc]] JambaForCausalLM
    - forward


## JambaForSequenceClassification[[transformers.JambaForSequenceClassification]]

[[autodoc]] transformers.JambaForSequenceClassification
    - forward