leran1995 commited on
Commit
ee9e9cd
·
verified ·
1 Parent(s): 59c140c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -21
README.md CHANGED
@@ -1,16 +1,16 @@
1
- ---
2
- license: apache-2.0
3
- language:
4
- - en
5
- - zh
6
- library_name: transformers
7
- pipeline_tag: text-generation
8
- tags:
9
- - llm
10
- - nanbeige
11
- base_model:
12
- - Nanbeige/Nanbeige4-3B-Base
13
- ---
14
  <div align="center">
15
 
16
  <img src="figures/nbg.png" width="220" alt="Nanbeige Logo">
@@ -24,22 +24,17 @@ base_model:
24
 
25
  # Introduction
26
  Nanbeige4-3B-Thinking-2511 is an enhanced iteration over our previous Nanbeige4-3B-Thinking-2510.
27
- Through advanced distillation techniques and reinforcement learning (RL) optimization, we have effectively scaled the model’s reasoning capacity, resulting in superior performance across a broad range of benchmarks.
28
- On math and science reasoning benchmarks, Nanbeige4-3B-Thinking-2511 outperforms Qwen3-4B-Thinking-2507, Qwen3-8B-Thinking-2504, and Qwen3-14B-Thinking-2504 with a significant margin.
29
- Besides, Nanbeige4-3B-Thinking-2511 achieves state-of-the-art (SOTA) results among models smaller than 32B parameters on general tasks like Arena-Hard-V2 and BFCL-V4.
30
- This marks a major milestone in delivering powerful, efficient reasoning performance at a compact scale.
31
 
32
  * Technical Report: https://arxiv.org/pdf/2512.06266
33
 
34
  <div align="center">
35
 
36
- <img src="figures/performance_reasoning.png">
37
  </div>
38
 
39
- <div align="center">
40
 
41
- <img src="figures/performance_2511.png">
42
- </div>
43
 
44
 
45
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - en
5
+ - zh
6
+ library_name: transformers
7
+ pipeline_tag: text-generation
8
+ tags:
9
+ - llm
10
+ - nanbeige
11
+ base_model:
12
+ - Nanbeige/Nanbeige4-3B-Base
13
+ ---
14
  <div align="center">
15
 
16
  <img src="figures/nbg.png" width="220" alt="Nanbeige Logo">
 
24
 
25
  # Introduction
26
  Nanbeige4-3B-Thinking-2511 is an enhanced iteration over our previous Nanbeige4-3B-Thinking-2510.
27
+ Through advanced knowledge distillation techniques and targeted reinforcement learning (RL) optimization, we have significantly scaled the model’s reasoning capabilities, delivering stronger and more reliable performance on diverse challenging benchmarks.
28
+ This version establishes new state-of-the-art (SOTA) results among open models under 32B parameters on AIME, GPQA-Diamond, Arena-Hard-V2, and BFCL-V4, which marks a major milestone in delivering powerful yet efficient reasoning capabilities at a compact scale.
 
 
29
 
30
  * Technical Report: https://arxiv.org/pdf/2512.06266
31
 
32
  <div align="center">
33
 
34
+ <img src="figures/nbg_performance.png">
35
  </div>
36
 
 
37
 
 
 
38
 
39
 
40