Hugging Face's logo

JetLM
/
SDAR-1.7B-Chat

Text Generation
Transformers
Safetensors
sdar
conversational
custom_code
Model card Files
xet
Community
1
dav1dliu commited on
Commit
3588ef6
·
verified ·
1 Parent(s): 5f8f25c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -4
README.md CHANGED
@@ -6,7 +6,8 @@ library_name: transformers
6
  # SDAR
7
 
8
  <div align="center">
9
- <img src="https://raw.githubusercontent.com/JetAstra/SDAR/refs/heads/main/assets/SDAR_doc_head.png?token=GHSAT0AAAAAADFKOPNMXTQ6UR7GA34QYJJ22FDIWHQ" />
 
10
 
11
  <div>&nbsp;</div>
12
 
@@ -34,7 +35,7 @@ For **SDAR** models, inference hyperparameters are set to: `block_length = 4`, `
34
  For **Qwen3-1.7B-AR-SFT** and **Qwen3-30B-AR-SFT**, we use *greedy decoding*, and the base models **Qwen3-1.7B-Base** and **Qwen3-30B-Base** are derived from the [Qwen3 Technical Report](https://arxiv.org/abs/2505.09388).
35
 
36
  <p align="center">
37
- <img src="https://raw.githubusercontent.com/JetAstra/SDAR/refs/heads/main/assets/table1.png?token=GHSAT0AAAAAADFKOPNMBLKTAWINPIVDQHKC2FDI34A" style="max-width:95%; height:auto;">
38
  <p align="center">
39
 
40
  ### SDAR-Sci v.s. AR Baseline
@@ -43,7 +44,7 @@ This table presents a **controlled comparison** between AR and SDAR under the sa
43
  The results are averaged over 8 runs for GPQA, and over 32 runs each for AIME 2024, AIME 2025, and LiveMathBench.
44
 
45
  <p align="center">
46
- <img src="https://raw.githubusercontent.com/JetAstra/SDAR/refs/heads/main/assets/table2.png?token=GHSAT0AAAAAADFKOPNMDZXQX3RWFLUXAZQU2FDI4KA" style="max-width:95%; height:auto;">
47
  <p align="center">
48
 
49
  #### SDAR-Sci v.s. Other Models
@@ -52,5 +53,5 @@ This table positions **SDAR-30B-A3B-Sci(sample)** against leading open-source an
52
  Scores for external models are sourced from the [InternLM/Intern-S1](https://github.com/InternLM/Intern-S1) repository.
53
 
54
  <p align="center">
55
- <img src="https://raw.githubusercontent.com/JetAstra/SDAR/refs/heads/main/assets/table3.png?token=GHSAT0AAAAAADFKOPNNMHMGMDWZ37WFK2MW2FDI4UQ" style="max-width:95%; height:auto;">
56
  <p align="center">
 
6
  # SDAR
7
 
8
  <div align="center">
9
+ <img src="https://raw.githubusercontent.com/JetAstra/SDAR/main/assets/SDAR_doc_head.png">
10
+
11
 
12
  <div>&nbsp;</div>
13
 
 
35
  For **Qwen3-1.7B-AR-SFT** and **Qwen3-30B-AR-SFT**, we use *greedy decoding*, and the base models **Qwen3-1.7B-Base** and **Qwen3-30B-Base** are derived from the [Qwen3 Technical Report](https://arxiv.org/abs/2505.09388).
36
 
37
  <p align="center">
38
+ <img src="https://raw.githubusercontent.com/JetAstra/SDAR/main/assets/table1.png" style="max-width:80%; height:auto;">
39
  <p align="center">
40
 
41
  ### SDAR-Sci v.s. AR Baseline
 
44
  The results are averaged over 8 runs for GPQA, and over 32 runs each for AIME 2024, AIME 2025, and LiveMathBench.
45
 
46
  <p align="center">
47
+ <img src="https://raw.githubusercontent.com/JetAstra/SDAR/main/assets/table2.png" style="max-width:80%; height:auto;">
48
  <p align="center">
49
 
50
  #### SDAR-Sci v.s. Other Models
 
53
  Scores for external models are sourced from the [InternLM/Intern-S1](https://github.com/InternLM/Intern-S1) repository.
54
 
55
  <p align="center">
56
+ <img src="https://raw.githubusercontent.com/JetAstra/SDAR/main/assets/table3.png" style="max-width:80%; height:auto;">
57
  <p align="center">