Hugging Face's logo

Skywork
/
Skywork-SWE-32B

Text Generation
Transformers
Safetensors
qwen2
swe-bench
conversational
text-generation-inference
Model card Files
xet
Community
2
ChangshiLi commited on
Commit
6a2c5a8
·
verified ·
1 Parent(s): 7d8496a

update assets path

Browse files
Files changed (1) hide show
  1. README.md +3 -3
README.md CHANGED
@@ -16,7 +16,7 @@ pipeline_tag: text-generation
16
 
17
  # Skywork-SWE
18
 
19
- ![image/png](https://huggingface.co/SWE-bench/SWE-agent-LM-32B/resolve/refs%2Fpr%2F6/assets/combined_swe.png)
20
 
21
  📖 [Report]() 📰 [Blog](https://quixotic-sting-239.notion.site/eb17f379610040ceb54da5d5d24065bd)
22
 
@@ -35,11 +35,11 @@ We also introduce an efficient and automated pipeline for SWE data collection, c
35
 
36
  ## Evaluation
37
 
38
- ![image/png](https://huggingface.co/SWE-bench/SWE-agent-LM-32B/resolve/refs%2Fpr%2F6/assets/data_scaling.png)
39
 
40
  Data Scaling Law for Pass@1 Accuracy on Qwen2.5-Coder-32B-Based LLMs Using the OpenHands v0.32.0 Code Agent Framework. Skywork-SWE-32B establishes a new state-of-the-art (SoTA) among the Qwen2.5-Coder-32B-based LLM, achieving the highest pass@1 accuracy without using verifiers or multiple rollouts.
41
 
42
- ![image/png](https://huggingface.co/SWE-bench/SWE-agent-LM-32B/resolve/refs%2Fpr%2F6/assets/accuracy.png)
43
 
44
  With the incorporation of test-time scaling techniques, Skywork-SWE-32B further improves to 47.0% pass@1 accuracy, surpassing the previous SoTA results for sub-32B parameter models.
45
 
 
16
 
17
  # Skywork-SWE
18
 
19
+ ![image/png](https://huggingface.co/Skywork/Skywork-SWE-32B/resolve/main/assets/combined_swe.png)
20
 
21
  📖 [Report]() 📰 [Blog](https://quixotic-sting-239.notion.site/eb17f379610040ceb54da5d5d24065bd)
22
 
 
35
 
36
  ## Evaluation
37
 
38
+ ![image/png](https://huggingface.co/Skywork/Skywork-SWE-32B/resolve/main/assets/data_scaling.png)
39
 
40
  Data Scaling Law for Pass@1 Accuracy on Qwen2.5-Coder-32B-Based LLMs Using the OpenHands v0.32.0 Code Agent Framework. Skywork-SWE-32B establishes a new state-of-the-art (SoTA) among the Qwen2.5-Coder-32B-based LLM, achieving the highest pass@1 accuracy without using verifiers or multiple rollouts.
41
 
42
+ ![image/png](https://huggingface.co/Skywork/Skywork-SWE-32B/resolve/main/assets/accuracy.png)
43
 
44
  With the incorporation of test-time scaling techniques, Skywork-SWE-32B further improves to 47.0% pass@1 accuracy, surpassing the previous SoTA results for sub-32B parameter models.
45