File size: 8,806 Bytes
197f827
 
53ba670
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
197f827
07aa290
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dc8d2fb
c1435ac
 
 
 
 
dc8d2fb
 
 
 
e45b2e8
07aa290
 
69262d0
07aa290
 
 
43ae814
 
 
 
 
 
c1435ac
43ae814
 
c1435ac
43ae814
69262d0
 
 
 
 
 
 
43ae814
07aa290
43ae814
 
07aa290
43ae814
07aa290
fed06c7
43ae814
07aa290
 
 
43ae814
07aa290
43ae814
 
 
 
 
 
 
 
07aa290
 
53ba670
 
 
 
 
 
 
 
 
 
 
 
 
 
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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
---
license: mit
model-index:
- name: ALMA-7B
  results:
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: AI2 Reasoning Challenge (25-Shot)
      type: ai2_arc
      config: ARC-Challenge
      split: test
      args:
        num_few_shot: 25
    metrics:
    - type: acc_norm
      value: 50.34
      name: normalized accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=haoranxu/ALMA-7B
      name: Open LLM Leaderboard
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: HellaSwag (10-Shot)
      type: hellaswag
      split: validation
      args:
        num_few_shot: 10
    metrics:
    - type: acc_norm
      value: 75.5
      name: normalized accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=haoranxu/ALMA-7B
      name: Open LLM Leaderboard
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: MMLU (5-Shot)
      type: cais/mmlu
      config: all
      split: test
      args:
        num_few_shot: 5
    metrics:
    - type: acc
      value: 38.04
      name: accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=haoranxu/ALMA-7B
      name: Open LLM Leaderboard
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: TruthfulQA (0-shot)
      type: truthful_qa
      config: multiple_choice
      split: validation
      args:
        num_few_shot: 0
    metrics:
    - type: mc2
      value: 35.64
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=haoranxu/ALMA-7B
      name: Open LLM Leaderboard
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: Winogrande (5-shot)
      type: winogrande
      config: winogrande_xl
      split: validation
      args:
        num_few_shot: 5
    metrics:
    - type: acc
      value: 72.38
      name: accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=haoranxu/ALMA-7B
      name: Open LLM Leaderboard
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: GSM8k (5-shot)
      type: gsm8k
      config: main
      split: test
      args:
        num_few_shot: 5
    metrics:
    - type: acc
      value: 0.0
      name: accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=haoranxu/ALMA-7B
      name: Open LLM Leaderboard
---
**ALMA** (**A**dvanced **L**anguage **M**odel-based tr**A**nslator) is an LLM-based translation model, which adopts a new translation model paradigm: it begins with fine-tuning on monolingual data and is further optimized using high-quality parallel data. This two-step fine-tuning process ensures strong translation performance. 
Please find more details in our [paper](https://arxiv.org/abs/2309.11674).
```
@misc{xu2023paradigm,
      title={A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models}, 
      author={Haoran Xu and Young Jin Kim and Amr Sharaf and Hany Hassan Awadalla},
      year={2023},
      eprint={2309.11674},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```
**[ALMA-R](https://arxiv.org/abs/2401.08417) (NEW!) is released now!** ALMA-R builds upon ALMA models, with further LoRA fine-tuning with our proposed **Contrastive Preference Optimization (CPO)** as opposed to the Supervised Fine-tuning used in ALMA. CPO fine-tuning requires our [triplet preference data](https://huggingface.co/datasets/haoranxu/ALMA-R-Preference) for preference learning. ALMA-R now can matches or even exceeds GPT-4 or WMT winners!


```
@misc{xu2024contrastive,
      title={Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translation}, 
      author={Haoran Xu and Amr Sharaf and Yunmo Chen and Weiting Tan and Lingfeng Shen and Benjamin Van Durme and Kenton Murray and Young Jin Kim},
      year={2024},
      eprint={2401.08417},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```
We release six translation models presented in the paper:
- **ALMA-7B**: Full-weight Fine-tune LLaMA-2-7B on 20B monolingual tokens and then **Full-weight** fine-tune on human-written parallel data
- **ALMA-7B-LoRA**: Full-weight Fine-tune LLaMA-2-7B on 20B monolingual tokens and then **LoRA** fine-tune on human-written parallel data
- **ALMA-7B-R (NEW!)**: Further LoRA fine-tuning upon ALMA-7B-LoRA with contrastive preference optimization.
- **ALMA-13B**: Full-weight Fine-tune LLaMA-2-7B on 12B monolingual tokens and then **Full-weight** fine-tune on human-written parallel data
- **ALMA-13B-LoRA** (Our best system): Full-weight Fine-tune LLaMA-2-7B on 12B monolingual tokens and then **LoRA** fine-tune on human-written parallel data
- **ALMA-13B-R (NEW!)**: Further LoRA fine-tuning upon ALMA-13B-LoRA with contrastive preference optimization. 
  
Model checkpoints are released at huggingface:
|     Models    | Base Model Link | LoRA Link |
|:-------------:|:---------------:|:---------:|
|    ALMA-7B    |        [haoranxu/ALMA-7B](https://huggingface.co/haoranxu/ALMA-7B)        |     -     |
|  ALMA-7B-LoRA |        [haoranxu/ALMA-7B-Pretrain](https://huggingface.co/haoranxu/ALMA-7B-Pretrain)        |     [haoranxu/ALMA-7B-Pretrain-LoRA](https://huggingface.co/haoranxu/ALMA-7B-Pretrain-LoRA)     |
|  **ALMA-7B-R (NEW!)** |        [haoranxu/ALMA-7B-R (LoRA merged)](https://huggingface.co/haoranxu/ALMA-7B-R)        |     -    |
|    ALMA-13B   |        [haoranxu/ALMA-13B](https://huggingface.co/haoranxu/ALMA-13B)        |     -     |
| ALMA-13B-LoRA |        [haoranxu/ALMA-13B-Pretrain](https://huggingface.co/haoranxu/ALMA-13B-Pretrain)        |     [haoranxu/ALMA-13B-Pretrain-LoRA](https://huggingface.co/haoranxu/ALMA-13B-Pretrain-LoRA)     |
| **ALMA-13B-R (NEW!)** |        [haoranxu/ALMA-13B-R (LoRA merged)](https://huggingface.co/haoranxu/ALMA-13B-R)        |    -   |

**Note that `ALMA-7B-Pretrain` and `ALMA-13B-Pretrain` are NOT translation models. They only experience stage 1 monolingual fine-tuning (20B tokens for the 7B model and 12B tokens for the 13B model), and should be utilized in conjunction with their LoRA models.** 

Datasets used by ALMA and ALMA-R are also released at huggingface now (NEW!)
|     Datasets    | Train / Validation| Test |
|:-------------:|:---------------:|:---------:|
|    Human-Written Parallel Data (ALMA)    |        [train and validation](https://huggingface.co/datasets/haoranxu/ALMA-Human-Parallel)        |     [WMT'22](https://huggingface.co/datasets/haoranxu/WMT22-Test)    |
|  Triplet Preference Data |        [train](https://huggingface.co/datasets/haoranxu/ALMA-R-Preference)        |   [WMT'22](https://huggingface.co/datasets/haoranxu/WMT22-Test) and [WMT'23](https://huggingface.co/datasets/haoranxu/WMT23-Test)   |

A quick start to use system ALMA-13B-LoRA for translation. An example of translating "我爱机器翻译。" into English:
```
import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM
from transformers import LlamaTokenizer

# Load base model and LoRA weights
model = AutoModelForCausalLM.from_pretrained("haoranxu/ALMA-13B-Pretrain", torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(model, "haoranxu/ALMA-13B-Pretrain-LoRA")
tokenizer = LlamaTokenizer.from_pretrained("haoranxu/ALMA-13B-Pretrain", padding_side='left')

# Add the source setence into the prompt template
prompt="Translate this from Chinese to English:\nChinese: 我爱机器翻译。\nEnglish:"
input_ids = tokenizer(prompt, return_tensors="pt", padding=True, max_length=40, truncation=True).input_ids.cuda()

# Translation
with torch.no_grad():
    generated_ids = model.generate(input_ids=input_ids, num_beams=5, max_new_tokens=20, do_sample=True, temperature=0.6, top_p=0.9)
outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
print(outputs)
```

Please find more details in our [GitHub repository](https://github.com/fe1ixxu/ALMA)
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_haoranxu__ALMA-7B)

|             Metric              |Value|
|---------------------------------|----:|
|Avg.                             |45.32|
|AI2 Reasoning Challenge (25-Shot)|50.34|
|HellaSwag (10-Shot)              |75.50|
|MMLU (5-Shot)                    |38.04|
|TruthfulQA (0-shot)              |35.64|
|Winogrande (5-shot)              |72.38|
|GSM8k (5-shot)                   | 0.00|