File size: 2,122 Bytes
c361859
 
 
 
 
 
 
 
 
 
5b71900
c361859
 
 
 
 
 
 
 
 
5b71900
c361859
 
 
 
 
 
 
 
 
c52e8b3
c361859
 
 
 
 
9029769
c361859
 
 
 
 
 
 
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
---
license: mit
train: false
inference: false
pipeline_tag: text-generation
---
This is a version of the <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B">DeepSeek-R1-Distill-Qwen-1.5B</a> model re-distilled for better performance.

## Performance

| Models            | <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B">DeepSeek-R1-Distill-Qwen-1.5B</a> | <a href="https://huggingface.co/mobiuslabsgmbh/DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1">DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1</a> | 
|:-------------------:|:--------:|:----------------:|
| ARC (25-shot)      | 40.96 | <b>41.3</b>  | 
| HellaSwag (10-shot)| 44    | <b>45.22</b> |
| MMLU (5-shot)      | 39.27 | <b>42.01</b> | 
| TruthfulQA-MC2     | 45.17 | <b>46.64</b> | 
| Winogrande (5-shot)| 55.49 | <b>56.75</b> | 
| GSM8K (5-shot)     | 69.9  | <b>73.24</b> | 
| Average            | 49.13 | <b>50.86</b> | 

| Models            | <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B">DeepSeek-R1-Distill-Qwen-1.5B</a> | <a href="https://huggingface.co/mobiuslabsgmbh/DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1">DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1</a>  | 
|:-------------------:|:--------:|:----------------:|
| GPQA (0-shot)     | 26.96 | <b>27.8</b>  | 
| MMLU PRO (5-shot) | 16.74 | <b>19.44</b> | 
| MUSR (0-shot)     | 35.93 | <b>35.94</b> | 
| BBH (3-shot)      | 35.12 | 35.11 | 
| Average           |  |  | 

## Usage
```Python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
compute_dtype = torch.bfloat16
device   = 'cuda'
model_id = "mobiuslabsgmbh/DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1"

model     = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=compute_dtype, attn_implementation="sdpa", device_map=device)
tokenizer = AutoTokenizer.from_pretrained(model_id)

chat    = tokenizer.apply_chat_template([{"role":"user", "content":"What is 1.5+102.2?"}], tokenize=True, add_generation_prompt=True, return_tensors="pt")
outputs = model_student.generate(chat.to(device), max_new_tokens=1024, do_sample=True) 
print(tokenizer.decode(outputs[0]))
```