Update README.md
Browse files
README.md
CHANGED
|
@@ -1,73 +1,77 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
🔥 LLaMAX2-7B-MetaMath
|
| 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 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- Multilingual
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
### Model Sources
|
| 7 |
+
- **Paper**: LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages
|
| 8 |
+
- **Link**: https://arxiv.org/pdf/2407.05975
|
| 9 |
+
- **Repository**: https://github.com/CONE-MT/LLaMAX/
|
| 10 |
+
|
| 11 |
+
### Model Description
|
| 12 |
+
|
| 13 |
+
🔥 LLaMAX2-7B-MetaMath is fully fine-tuned on the MetaMathQA dataset based on the powerful multilingual model LLaMAX2-7B.
|
| 14 |
+
|
| 15 |
+
🔥 Compared with the [MetaMath-7B](https://huggingface.co/meta-math/MetaMath-7B-V1.0), LLaMAX2-7B-MetaMath performs significantly better in mathematical reasoning in low-resource languages, improving the average accuracy of low-resource languages on MGSM dataset by up to 18.8%.
|
| 16 |
+
|
| 17 |
+
🔥 LLaMAX2-7B-MetaMath demonstrates good multilingual math reasoning capability in all languages, improving the average accuracy by 6.2% across all languages in MGSM dataset.
|
| 18 |
+
|
| 19 |
+
### Experiments
|
| 20 |
+
We evaluated LLaMAX2-7B-MetaMath on the MGSM dataset. Compared with MetaMath-7B, LLaMAX-7B-MetaMath achieves a leading on both high-resource languages (Hrl.) and low-resource languages (Lrl.).
|
| 21 |
+
|
| 22 |
+
| MGSM | Avg. | Lrl. | Hrl. | Bn | Th | Sw | Ja | Zh | De | Fr | Ru | Es | En |
|
| 23 |
+
|---------------------------|---------|------|--------|--------|------|----|----|------|----|----|------|------|--------|
|
| 24 |
+
| MetaMath-7B (official) | 38.32 | 6.9 | 51.8 | 6.8 | 7.2 |6.8| 36.4 | 38.4 | 55.2|54.4| 52.0 |57.2|68.8|
|
| 25 |
+
| MetaMath-7B (Reproduced) | 38.08 | 6.8 | 51.5 | 6.0 | 10.0 |4.4| 36.4 |42.8|52.8|56.0|48.8|58.8|64.8|
|
| 26 |
+
| LLaMAX2-7B-MetaMath | 44.28 | 25.6 | 52.3 | 26.8 | 24.0 |26.0| 35.6 |42.4|56.8|55.2|53.6|56.8|65.6|
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
### Model Usage
|
| 31 |
+
|
| 32 |
+
Prompt template:
|
| 33 |
+
```angular2html
|
| 34 |
+
def Prompt_template(query):
|
| 35 |
+
prompt = (
|
| 36 |
+
"Below is an instruction that describes a task. "
|
| 37 |
+
"Write a response that appropriately completes the request.\n\n"
|
| 38 |
+
f"### Instruction:\n{query}\n\n### Response: Let's think step by step."
|
| 39 |
+
)
|
| 40 |
+
return prompt
|
| 41 |
+
```
|
| 42 |
+
|
| 43 |
+
Code Example:
|
| 44 |
+
```angular2html
|
| 45 |
+
from transformers import AutoTokenizer, LlamaForCausalLM
|
| 46 |
+
|
| 47 |
+
model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
|
| 48 |
+
tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
|
| 49 |
+
|
| 50 |
+
query = "Bert fills out the daily crossword puzzle in the newspaper every day. He uses a pencil to fill out the puzzles every two weeks. On average, it takes him 1050 words to use up a pencil. How many words are in each crossword puzzle on average?"
|
| 51 |
+
prompt = Prompt_template(query)
|
| 52 |
+
inputs = tokenizer(prompt, return_tensors="pt")
|
| 53 |
+
|
| 54 |
+
generate_ids = model.generate(inputs.input_ids, max_length=30)
|
| 55 |
+
tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
| 56 |
+
|
| 57 |
+
# => "If Bert uses up a pencil to fill out the puzzles every two weeks and it takes him 1050
|
| 58 |
+
words to use up a pencil, then he must be filling out 1050 words of crossword puzzles every
|
| 59 |
+
two weeks. To find out how many words are in each daily crossword puzzle, we need to divide
|
| 60 |
+
the total number of words (1050) by the number of days in two weeks (14). So, there are
|
| 61 |
+
1050/14 = 75 words in each daily crossword puzzle on average. #### The answer is: 75“
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
### Citation
|
| 65 |
+
if our model helps your work, please cite this paper:
|
| 66 |
+
|
| 67 |
+
```
|
| 68 |
+
@misc{lu2024llamaxscalinglinguistichorizons,
|
| 69 |
+
title={LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages},
|
| 70 |
+
author={Yinquan Lu and Wenhao Zhu and Lei Li and Yu Qiao and Fei Yuan},
|
| 71 |
+
year={2024},
|
| 72 |
+
eprint={2407.05975},
|
| 73 |
+
archivePrefix={arXiv},
|
| 74 |
+
primaryClass={cs.CL},
|
| 75 |
+
url={https://arxiv.org/abs/2407.05975},
|
| 76 |
+
}
|
| 77 |
+
```
|