File size: 2,948 Bytes
94002a1
 
8dbe171
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
94002a1
 
8dbe171
94002a1
8dbe171
 
 
 
 
94002a1
8dbe171
94002a1
8dbe171
 
 
 
 
 
 
94002a1
8dbe171
94002a1
8dbe171
94002a1
 
 
8dbe171
 
 
94002a1
8dbe171
 
 
 
94002a1
8dbe171
94002a1
 
 
8dbe171
 
 
 
 
 
 
94002a1
8dbe171
94002a1
8dbe171
 
 
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
---
library_name: transformers
tags:
- small-lm
- math
- code
- reasoning
- slm
- french
- merge
license: apache-2.0
base_model:
- SamsungSAILMontreal/Qwen3-1.7B
- SamsungSAILMontreal/Qwen3-1.7B-Math
- SamsungSAILMontreal/Qwen3-1.7B-Code
- SamsungSAILMontreal/Qwen3-1.7B-Fr
language:
- fr
- en
---

# Qwen3-1.7B-Math-Fr

This model is obtained by merging [SamsungSAILMontreal/Qwen3-1.7B-Math](https://huggingface.co/SamsungSAILMontreal/Qwen3-1.7B-Math), 
[SamsungSAILMontreal/Qwen3-1.7B-Code](https://huggingface.co/SamsungSAILMontreal/Qwen3-1.7B-Code) and 
[SamsungSAILMontreal/Qwen3-1.7B-Fr](https://huggingface.co/SamsungSAILMontreal/Qwen3-1.7B-Fr).
The model is used in the experiments described in https://bknyaz.github.io/blog/2026/meta-merge/. 
Single A100 was used for merging and evaluation.

The following versions were used for merge/eval:

- python >= 3.10
- torch               : 2.9.0+cu128
- lm_eval             : 0.4.9.1
- vllm                : 0.11.1
- transformers        : 4.57.6
- datasets            : 3.2.0
- numpy               : 2.2.6

## Merging

Merging was done using parameter averaging implemented in [merge_qwen.py](https://github.com/SamsungSAILMontreal/nino/blob/main/merge_qwen.py).

## Evaluation

Evaluation was done with lm_eval on the test split of [gsm8k](https://huggingface.co/datasets/openai/gsm8k), 
[french_bench (avg score)](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/french_bench), 
[gsm8k-fr](https://huggingface.co/datasets/cmh/gsm8k_fr) and humaneval (instruct):

```bash
python -m lm_eval --model vllm --model_args pretrained=${model},tensor_parallel_size=1,dtype=auto,gpu_memory_utilization=0.9,data_parallel_size=1 \
 --tasks gsm8k,french_bench,gsm8k-fr,humaneval_instruct --batch_size 1 --apply_chat_template=True --confirm_run_unsafe_code --trust_remote_code
```

To evaluate on gsm8k-fr you can use our fork https://github.com/bknyaz/lm-evaluation-harness/tree/main/lm_eval/tasks/gsm8k. 

### Results

| Model                     | gsm8k | french | gsm8k-fr | humaneval_instruct | avg  |
|---------------------------|-------|--------|----------|--------------------|------|
| Qwen3-1.7B                | 20.6  | 26.2   | 20.2     | 67.1               | 33.5 |
| Qwen3-1.7B-Math           | 62.1  | 28.3   | 41.5     | 48.2               | 45.0 |
| Qwen3-1.7B-Fr             | 60.9  | 32.8   | 43.9     | 56.1               | 48.4 |
| Qwen3-1.7B-Code           | 56.7  | 28.2   | 36.5     | 69.5               | 47.7 |
| Qwen3-1.7B-Math-Code-Fr   | 64.6  | 29.2   | 48.4     | 65.2               | 51.8 |

## License

Please refer to the license of the base models [SamsungSAILMontreal/Qwen3-1.7B-Math](https://huggingface.co/SamsungSAILMontreal/Qwen3-1.7B-Math), 
[SamsungSAILMontreal/Qwen3-1.7B-Code](https://huggingface.co/SamsungSAILMontreal/Qwen3-1.7B-Code) and 
[SamsungSAILMontreal/Qwen3-1.7B-Fr](https://huggingface.co/SamsungSAILMontreal/Qwen3-1.7B-Fr).