phmartins commited on
Commit
4a6927d
1 Parent(s): c3d0551

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +147 -40
README.md CHANGED
@@ -1,59 +1,166 @@
1
  ---
2
- library_name: transformers
3
- base_model: Unbabel/EuroLLM-1B-annealed-fw
4
- tags:
5
- - generated_from_trainer
6
- model-index:
7
- - name: experiments/EuroLLM/EuroLLM-1B-annealed-fw-euroblocks-v0.1
8
- results: []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9
  ---
10
 
11
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
12
- should probably proofread and complete it, then remove this comment. -->
13
 
14
- [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
15
- # experiments/EuroLLM/EuroLLM-1B-annealed-fw-euroblocks-v0.1
16
 
17
- This model is a fine-tuned version of [Unbabel/EuroLLM-1B-annealed-fw](https://huggingface.co/Unbabel/EuroLLM-1B-annealed-fw) on the None dataset.
18
 
19
- ## Model description
20
 
21
- More information needed
22
 
23
- ## Intended uses & limitations
24
 
25
- More information needed
26
 
27
- ## Training and evaluation data
28
 
29
- More information needed
30
 
31
- ## Training procedure
 
 
 
 
32
 
33
- ### Training hyperparameters
34
 
35
- The following hyperparameters were used during training:
36
- - learning_rate: 7e-06
37
- - train_batch_size: 2
38
- - eval_batch_size: 1
39
- - seed: 42
40
- - distributed_type: multi-GPU
41
- - num_devices: 2
42
- - gradient_accumulation_steps: 24
43
- - total_train_batch_size: 96
44
- - total_eval_batch_size: 2
45
- - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
46
- - lr_scheduler_type: cosine
47
- - lr_scheduler_warmup_steps: 300
48
- - num_epochs: 4
49
 
50
- ### Training results
51
 
52
 
 
53
 
54
- ### Framework versions
 
 
 
 
55
 
56
- - Transformers 4.44.2
57
- - Pytorch 2.2.2+cu121
58
- - Datasets 2.19.1
59
- - Tokenizers 0.19.1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: apache-2.0
3
+ language:
4
+ - en
5
+ - de
6
+ - es
7
+ - fr
8
+ - it
9
+ - pt
10
+ - pl
11
+ - nl
12
+ - tr
13
+ - sv
14
+ - cs
15
+ - el
16
+ - hu
17
+ - ro
18
+ - fi
19
+ - uk
20
+ - sl
21
+ - sk
22
+ - da
23
+ - lt
24
+ - lv
25
+ - et
26
+ - bg
27
+ - no
28
+ - ca
29
+ - hr
30
+ - ga
31
+ - mt
32
+ - gl
33
+ - zh
34
+ - ru
35
+ - ko
36
+ - ja
37
+ - ar
38
+ - hi
39
  ---
40
 
41
+ ## *Model updated on September 24*
 
42
 
 
 
43
 
 
44
 
 
45
 
 
46
 
47
+ # Model Card for EuroLLM-1.7B-Instruct
48
 
 
49
 
 
50
 
51
+ This is the model card for the first instruction tuned model of the EuroLLM series: EuroLLM-1.7B-Instruct. You can also check the pre-trained version: [EuroLLM-1.7B](https://huggingface.co/utter-project/EuroLLM-1.7B).
52
 
53
+ - **Developed by:** Unbabel, Instituto Superior Técnico, University of Edinburgh, Aveni, University of Paris-Saclay, University of Amsterdam, Naver Labs, Sorbonne Université.
54
+ - **Funded by:** European Union.
55
+ - **Model type:** A 1.7B parameter instruction tuned multilingual transfomer LLM.
56
+ - **Language(s) (NLP):** Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Irish, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Arabic, Catalan, Chinese, Galician, Hindi, Japanese, Korean, Norwegian, Russian, Turkish, and Ukrainian.
57
+ - **License:** Apache License 2.0.
58
 
59
+ ## Model Details
60
 
61
+ The EuroLLM project has the goal of creating a suite of LLMs capable of understanding and generating text in all European Union languages as well as some additional relevant languages.
62
+ EuroLLM-1.7B is a 1.7B parameter model trained on 4 trillion tokens divided across the considered languages and several data sources: Web data, parallel data (en-xx and xx-en), and high-quality datasets.
63
+ EuroLLM-1.7B-Instruct was further instruction tuned on EuroBlocks, an instruction tuning dataset with focus on general instruction-following and machine translation.
 
 
 
 
 
 
 
 
 
 
 
64
 
 
65
 
66
 
67
+ ### Model Description
68
 
69
+ EuroLLM uses a standard, dense Transformer architecture:
70
+ - We use grouped query attention (GQA) with 8 key-value heads, since it has been shown to increase speed at inference time while maintaining downstream performance.
71
+ - We perform pre-layer normalization, since it improves the training stability, and use the RMSNorm, which is faster.
72
+ - We use the SwiGLU activation function, since it has been shown to lead to good results on downstream tasks.
73
+ - We use rotary positional embeddings (RoPE) in every layer, since these have been shown to lead to good performances while allowing the extension of the context length.
74
 
75
+ For pre-training, we use 256 Nvidia H100 GPUs of the Marenostrum 5 supercomputer, training the model with a constant batch size of 3,072 sequences, which corresponds to approximately 12 million tokens, using the Adam optimizer, and BF16 precision.
76
+ Here is a summary of the model hyper-parameters:
77
+ | | |
78
+ |--------------------------------------|----------------------|
79
+ | Sequence Length | 4,096 |
80
+ | Number of Layers | 24 |
81
+ | Embedding Size | 2,048 |
82
+ | FFN Hidden Size | 5,632 |
83
+ | Number of Heads | 16 |
84
+ | Number of KV Heads (GQA) | 8 |
85
+ | Activation Function | SwiGLU |
86
+ | Position Encodings | RoPE (\Theta=10,000) |
87
+ | Layer Norm | RMSNorm |
88
+ | Tied Embeddings | No |
89
+ | Embedding Parameters | 0.262B |
90
+ | LM Head Parameters | 0.262B |
91
+ | Non-embedding Parameters | 1.133B |
92
+ | Total Parameters | 1.657B |
93
+
94
+ ## Run the model
95
+
96
+ from transformers import AutoModelForCausalLM, AutoTokenizer
97
+
98
+ model_id = "utter-project/EuroLLM-1.7B-Instruct"
99
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
100
+ model = AutoModelForCausalLM.from_pretrained(model_id)
101
+
102
+ text = '<|im_start|>system\n<|im_end|>\n<|im_start|>user\nTranslate the following English source text to Portuguese:\nEnglish: I am a language model for european languages. \nPortuguese: <|im_end|>\n<|im_start|>assistant\n'
103
+
104
+ inputs = tokenizer(text, return_tensors="pt")
105
+ outputs = model.generate(**inputs, max_new_tokens=20)
106
+ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
107
+
108
+
109
+
110
+ ## Results
111
+
112
+ ### Machine Translation
113
+
114
+ We evaluate EuroLLM-1.7B-Instruct on several machine translation benchmarks: FLORES-200, WMT-23, and WMT-24 comparing it with [Gemma-2B](https://huggingface.co/google/gemma-2b) and [Gemma-7B](https://huggingface.co/google/gemma-7b) (also instruction tuned on EuroBlocks).
115
+ The results show that EuroLLM-1.7B is substantially better than Gemma-2B in Machine Translation and competitive with Gemma-7B.
116
+
117
+
118
+ #### Flores-200
119
+ | Model | AVG | AVG en-xx | AVG xx-en | en-ar | en-bg | en-ca | en-cs | en-da | en-de | en-el | en-es-latam | en-et | en-fi | en-fr | en-ga | en-gl | en-hi | en-hr | en-hu | en-it | en-ja | en-ko | en-lt | en-lv | en-mt | en-nl | en-no | en-pl | en-pt-br | en-ro | en-ru | en-sk | en-sl | en-sv | en-tr | en-uk | en-zh-cn | ar-en | bg-en | ca-en | cs-en | da-en | de-en | el-en | es-latam-en | et-en | fi-en | fr-en | ga-en | gl-en | hi-en | hr-en | hu-en | it-en | ja-en | ko-en | lt-en | lv-en | mt-en | nl-en | no-en | pl-en | pt-br-en | ro-en | ru-en | sk-en | sl-en | sv-en | tr-en | uk-en | zh-cn-en |
120
+ |--------------------------------|------|-----------|-----------|-------|-------|-------|-------|-------|-------|-------|--------------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|----------|-------|-------|-------|-------|-------|-------|-------|----------|-------|-------|-------|-------|-------|-------|-------|--------------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|----------|-------|-------|-------|-------|-------|-------|-------|----------|
121
+ | EuroLLM-1.7B-Instruct |86.89 | 86.53 | 87.25 | 85.17 | 89.42 | 84.72 | 89.13 | 89.47 | 86.90 | 87.60 | 86.29 | 88.95 | 89.40 | 87.69 | 74.89 | 86.41 | 76.92 | 84.79 | 86.78 | 88.17 | 89.76 | 87.70 | 87.27 | 87.62 | 67.84 | 87.10 | 90.00 | 88.18 | 89.29 | 89.49 | 88.32 | 88.18 | 86.85 | 90.00 | 87.31 | 87.89 | 86.60 | 86.34 | 87.45 | 87.57 | 87.95 | 89.72 | 88.80 | 87.00 | 86.77 | 88.34 | 89.09 | 88.95 | 82.69 | 87.80 | 88.37 | 86.71 | 87.20 | 87.81 | 86.79 | 86.79 | 85.62 | 86.48 | 81.10 | 86.97 | 90.25 | 85.75 | 89.20 | 88.88 | 86.00 | 87.38 | 86.76 | 89.61 | 87.94 |
122
+ | Gemma-2B-EuroBlocks | 81.59 | 78.97 | 84.21 | 76.68 | 82.73 | 83.14 | 81.63 | 84.63 | 83.15 | 79.42 | 84.05 | 72.58 | 79.73 | 84.97 | 40.50 | 82.13 | 67.79 | 80.53 | 78.36 | 84.90 | 87.43 | 82.98 | 72.29 | 68.68 | 58.55 | 83.13 | 86.15 | 82.78 | 86.79 | 83.14 | 84.61 | 78.18 | 75.37 | 80.89 | 78.38 | 84.38 | 84.35 | 83.88 | 85.77 | 86.85 | 86.31 | 88.24 | 88.12 | 84.79 | 84.90 | 82.51 | 86.32 | 88.29 | 54.78 | 86.53 | 85.83 | 85.41 | 85.18 | 86.77 | 85.78 | 84.99 | 81.65 | 81.78 | 67.27 | 85.92 | 89.07 | 84.14 | 88.07 | 87.17 | 85.23 | 85.09 | 83.95 | 87.57 | 84.77 |
123
+ | Gemma-7B-EuroBlocks |85.27 | 83.90 | 86.64 | 86.38 | 87.87 | 85.74 | 84.25 | 85.69 | 81.49 | 85.52 | 86.93 | 62.83 | 84.96 | 75.34 | 84.93 | 83.91 | 86.92 | 88.19 | 86.11 | 81.73 | 80.55 | 66.85 | 85.31 | 89.36 | 85.87 | 88.62 | 88.06 | 86.67 | 84.79 | 82.71 | 86.45 | 85.19 | 86.67 | 85.77 | 86.36 | 87.21 | 88.09 | 87.17 | 89.40 | 88.26 | 86.74 | 86.73 | 87.25 | 88.87 | 88.81 | 72.45 | 87.62 | 87.86 | 87.08 | 87.01 | 87.58 | 86.92 | 86.70 | 85.10 | 85.74 | 77.81 | 86.83 | 90.40 | 85.41 | 89.04 | 88.77 | 86.13 | 86.67 | 86.32 | 89.27 | 87.92 |
124
+
125
+
126
+ #### WMT-23
127
+ | Model | AVG | AVG en-xx | AVG xx-en | AVG xx-xx | en-de | en-cs | en-uk | en-ru | en-zh-cn | de-en | uk-en | ru-en | zh-cn-en | cs-uk |
128
+ |--------------------------------|------|-----------|-----------|-----------|-------|-------|-------|-------|----------|-------|-------|-------|----------|-------|
129
+ | EuroLLM-1.7B-Instruct | 82.91 | 83.20 | 81.77 | 86.82 | 81.56 | 85.23 | 81.30 | 82.47 | 83.61 | 85.03 | 84.06 | 85.25 | 81.31 | 78.83 | 79.42 | 86.82 |
130
+ | Gemma-2B-EuroBlocks | 79.96 | 79.01 | 80.86 | 81.15 | 76.82 | 76.05 | 77.92 | 78.98 | 81.58 | 82.73 | 82.71 | 83.99 | 80.35 | 78.27 | 78.99 | 81.15 |
131
+ | Gemma-7B-EuroBlocks | 82.76 | 82.26 | 82.70 | 85.98 | 81.37 | 82.42 | 81.54 | 82.18 | 82.90 | 83.17 | 84.29 | 85.70 | 82.46 | 79.73 | 81.33 | 85.98 |
132
+
133
+
134
+
135
+ #### WMT-24
136
+ | Model | AVG | AVG en-xx | AVG xx-xx | en-de | en-es-latam | en-cs | en-ru | en-uk | en-ja | en-zh-cn | en-hi | cs-uk | ja-zh-cn |
137
+ |---------|------|------|-------|----|---|-------|-------|--------|--------|-------|-------|-------|-----|
138
+ | EuroLLM-1.7B-Instruct|79.32 | 79.32 | 79.34 | 79.42 | 80.67 | 80.55 | 78.65 | 80.12 | 82.96 | 80.60 | 71.59 | 83.48 | 75.20 |
139
+ |Gemma-2B-EuroBlocks| 74.72 | 74.41 | 75.97 | 74.93 | 78.81 | 70.54 | 74.90 | 75.84 | 79.48 | 78.06 | 62.70 | 79.87 | 72.07 |
140
+ |Gemma-7B-EuroBlocks| 78.67 | 78.34 | 80.00 | 78.88 | 80.47 | 78.55 | 78.55 | 80.12 | 80.55 | 78.90 | 70.71 | 84.33 | 75.66 |
141
+
142
+
143
+ ### General Benchmarks
144
+ We also compare EuroLLM-1.7B with [TinyLlama-v1.1](https://huggingface.co/TinyLlama/TinyLlama_v1.1) and [Gemma-2B](https://huggingface.co/google/gemma-2b) on 3 general benchmarks: Arc Challenge and Hellaswag.
145
+ For the non-english languages we use the [Okapi](https://aclanthology.org/2023.emnlp-demo.28.pdf) datasets.
146
+ Results show that EuroLLM-1.7B is superior to TinyLlama-v1.1 and similar to Gemma-2B on Hellaswag but worse on Arc Challenge. This can be due to the lower number of parameters of EuroLLM-1.7B (1.133B non-embedding parameters against 1.981B).
147
+
148
+ #### Arc Challenge
149
+ | Model | Average | English | German | Spanish | French | Italian | Portuguese | Chinese | Russian | Dutch | Arabic | Swedish | Hindi | Hungarian | Romanian | Ukrainian | Danish | Catalan |
150
+ |--------------------|---------|---------|--------|---------|--------|---------|------------|---------|---------|-------|--------|---------|--------|-----------|----------|-----------|--------|---------|
151
+ | EuroLLM-1.7B | 0.3496 | 0.4061 | 0.3464 | 0.3684 | 0.3627 | 0.3738 | 0.3855 | 0.3521 | 0.3208 | 0.3507 | 0.3045 | 0.3605 | 0.2928 | 0.3271 | 0.3488 | 0.3516 | 0.3513 | 0.3396 |
152
+ | TinyLlama-v1.1 | 0.2650 | 0.3712 | 0.2524 | 0.2795 | 0.2883 | 0.2652 | 0.2906 | 0.2410 | 0.2669 | 0.2404 | 0.2310 | 0.2687 | 0.2354 | 0.2449 | 0.2476 | 0.2524 | 0.2494 | 0.2796 |
153
+ | Gemma-2B | 0.3617 | 0.4846 | 0.3755 | 0.3940 | 0.4080 | 0.3687 | 0.3872 | 0.3726 | 0.3456 | 0.3328 | 0.3122 | 0.3519 | 0.2851 | 0.3039 | 0.3590 | 0.3601 | 0.3565 | 0.3516 |
154
+
155
+
156
+ #### Hellaswag
157
+ | Model | Average | English | German | Spanish | French | Italian | Portuguese | Russian | Dutch | Arabic | Swedish | Hindi | Hungarian | Romanian | Ukrainian | Danish | Catalan |
158
+ |--------------------|---------|---------|--------|---------|--------|---------|------------|---------|--------|--------|---------|--------|-----------|----------|-----------|--------|---------|
159
+ | EuroLLM-1.7B | 0.4744 | 0.4760 | 0.6057 | 0.4793 | 0.5337 | 0.5298 | 0.5085 | 0.5224 | 0.4654 | 0.4949 | 0.4104 | 0.4800 | 0.3655 | 0.4097 | 0.4606 | 0.436 | 0.4702 | 0.4445 |
160
+ | TinyLlama-v1.1 |0.3674 | 0.6248 | 0.3650 | 0.4137 | 0.4010 | 0.3780 | 0.3892 | 0.3494 | 0.3588 | 0.2880 | 0.3561 | 0.2841 | 0.3073 | 0.3267 | 0.3349 | 0.3408 | 0.3613 |
161
+ | Gemma-2B |0.4666 | 0.7165 | 0.4756 | 0.5414 | 0.5180 | 0.4841 | 0.5081 | 0.4664 | 0.4655 | 0.3868 | 0.4383 | 0.3413 | 0.3710 | 0.4316 | 0.4291 | 0.4471 | 0.4448 |
162
+
163
+
164
+ ## Bias, Risks, and Limitations
165
+
166
+ EuroLLM-1.7B-Instruct has not been aligned to human preferences, so the model may generate problematic outputs (e.g., hallucinations, harmful content, or false statements).