harithoppil commited on
Commit
cdfbb49
·
verified ·
1 Parent(s): 774e52e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +7 -9
README.md CHANGED
@@ -13,7 +13,7 @@ task_categories:
13
  language:
14
  - en
15
  size_categories:
16
- - 10K<n<100K
17
  configs:
18
  - config_name: all
19
  data_files:
@@ -61,21 +61,23 @@ dataset_info:
61
  dtype: string
62
  splits:
63
  - name: train
64
- num_examples: 12980
65
  ---
66
 
67
  # ML SWE Prompts
68
 
69
  Unified collection of ML/training-related software engineering prompts for OPD distillation training. All prompts are in English.
70
 
 
 
71
  ## Splits
72
 
73
  | Config | Source | Rows | Description |
74
  |--------|--------|------|-------------|
75
- | `all` | Combined | 12,980 | All prompts combined |
76
- | `swe_bench_ml` | SWE-bench train | 9,472 | Problem statements from ML repos (PyTorch, HF, Lightning, Ray, numpy, scipy, TF, etc.) |
77
  | `swe_dev_sft` | SWE-Dev-train | 3,054 | ML-related agent conversations (SFT format) |
78
- | `swe_dev_rft` | SWE-Dev-train | 347 | ML-related agent conversations (RFT format, with rewards) |
79
  | `terminal_bench_verified` | Terminal-Bench 2 Verified | 89 | Task instructions from TB2 verified tasks |
80
  | `terminal_bench_trajectories` | TB2 Leaderboard | 18 | Unique ML task prompts from agent trajectories (deduplicated) |
81
 
@@ -88,10 +90,6 @@ All rows have at minimum `source` and `prompt`. Additional fields vary by source
88
  - **terminal_bench_verified**: `task_name`
89
  - **terminal_bench_trajectories**: `task_name`, `model`, `agent`, `reward`
90
 
91
- ## ML Repos in SWE-bench
92
-
93
- huggingface (1,058), Lightning-AI (377), ray-project (342), numpy (937), scipy (101), tensorflow (50), open-mmlab (111), explosion/spaCy (41), pandas-dev (5,049), Qiskit (1,406)
94
-
95
  ## Citation
96
 
97
  ```bibtex
 
13
  language:
14
  - en
15
  size_categories:
16
+ - 1K<n<10K
17
  configs:
18
  - config_name: all
19
  data_files:
 
61
  dtype: string
62
  splits:
63
  - name: train
64
+ num_examples: 6220
65
  ---
66
 
67
  # ML SWE Prompts
68
 
69
  Unified collection of ML/training-related software engineering prompts for OPD distillation training. All prompts are in English.
70
 
71
+ Filtered to core ML repos: huggingface (1,058), numpy (937), Lightning-AI (377), ray-project (342). Excludes pandas-dev, qiskit, open-mmlab, scipy, tensorflow, spaCy.
72
+
73
  ## Splits
74
 
75
  | Config | Source | Rows | Description |
76
  |--------|--------|------|-------------|
77
+ | `all` | Combined | 6,220 | All prompts combined |
78
+ | `swe_bench_ml` | SWE-bench train | 2,714 | Problem statements from core ML repos (HF, numpy, Lightning, Ray) + keyword-matched from SWE-Dev |
79
  | `swe_dev_sft` | SWE-Dev-train | 3,054 | ML-related agent conversations (SFT format) |
80
+ | `swe_dev_rft` | SWE-Dev-train | 345 | ML-related agent conversations (RFT format, with rewards) |
81
  | `terminal_bench_verified` | Terminal-Bench 2 Verified | 89 | Task instructions from TB2 verified tasks |
82
  | `terminal_bench_trajectories` | TB2 Leaderboard | 18 | Unique ML task prompts from agent trajectories (deduplicated) |
83
 
 
90
  - **terminal_bench_verified**: `task_name`
91
  - **terminal_bench_trajectories**: `task_name`, `model`, `agent`, `reward`
92
 
 
 
 
 
93
  ## Citation
94
 
95
  ```bibtex