Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,36 @@
|
|
| 1 |
---
|
| 2 |
license: other
|
| 3 |
license_name: mixed-upstream-licenses
|
| 4 |
-
license_link: LICENSE
|
| 5 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: other
|
| 3 |
license_name: mixed-upstream-licenses
|
|
|
|
| 4 |
---
|
| 5 |
+
|
| 6 |
+
## Dataset Description
|
| 7 |
+
|
| 8 |
+
[SWE-Critix/test_dataset](https://huggingface.co/datasets/SWE-Critix/test_dataset) is the testing set of the models [SWE-Critix/SWE-Critix-Qwen3-30B-A3B-SFT-Epoch3](https://huggingface.co/SWE-Critix/SWE-Critix-Qwen3-30B-A3B-SFT-Epoch3) and [SWE-Critix/SWE-Critix-Qwen3-30B-A3B-RL-Step-3396](https://huggingface.co/SWE-Critix/SWE-Critix-Qwen3-30B-A3B-RL-Step-3396). It contains 43445 samples (28741 positives vs. 14704 negatives). Each sample takes the following format:
|
| 9 |
+
|
| 10 |
+
<pre><code>{
|
| 11 |
+
"prompt": a prompt using <a href="https://huggingface.co/SWE-Critix/SWE-Critix-Qwen3-30B-A3B-SFT-Epoch3/blob/main/codeforge_preview_patch_correctness_judge.txt">codeforge_preview_patch_correctness_judge.txt</a> as the template,
|
| 12 |
+
"ground_truth": "1" if the trajectory passed unit testing, and "0" otherwise,
|
| 13 |
+
"base_model_output": the response generated by the base model (i.e., Qwen3-30B-A3B-Thinking-2507),
|
| 14 |
+
"sft_epoch_1_output": the response generated by the SFT checkpoint at iteration 325 (i.e., the first epoch),
|
| 15 |
+
"sft_epoch_2_output": the response generated by the SFT checkpoint at iteration 650 (i.e., the second epoch),
|
| 16 |
+
"sft_epoch_3_output": the response generated by the SFT checkpoint at iteration 975 (i.e., the third epoch),
|
| 17 |
+
"rl_step_1100_output": the response generated by the RL checkpoint at iteration 1100,
|
| 18 |
+
"rl_step_2300_output": the response generated by the RL checkpoint at iteration 2300,
|
| 19 |
+
"rl_step_3396_output": the response generated by the RL checkpoint at iteration 3396,
|
| 20 |
+
"
|
| 21 |
+
}</code></pre>
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
## Data Sources and Licenses
|
| 25 |
+
|
| 26 |
+
This dataset contains data derived from multiple upstream sources: [togethercomputer/CoderForge-Preview](https://huggingface.co/datasets/togethercomputer/CoderForge-Preview), [R2E-Gym](https://huggingface.co/R2E-Gym), [nebius/SWE-rebench](https://huggingface.co/datasets/nebius/SWE-rebench), and [SWE-bench/SWE-smith](https://huggingface.co/datasets/SWE-bench/SWE-smith). The original data remain subject to the licenses and terms of their respective sources. Users are responsible for complying with the applicable licenses and terms of the respective upstream datasets.
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
## Attribution
|
| 30 |
+
|
| 31 |
+
Please refer to the corresponding upstream datasets for their required attribution and licensing terms.
|
| 32 |
+
|
| 33 |
+
## License
|
| 34 |
+
|
| 35 |
+
Because this dataset incorporates data from multiple upstream sources with potentially different licensing terms, no single license is intended to override the licenses of the individual source datasets. The licensing terms of the original source data remain applicable to those portions of the dataset.
|
| 36 |
+
|