Datasets:
Upload schema.json with huggingface_hub
Browse files- schema.json +21 -0
schema.json
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"description": "Schema for dsfasdfaserwer/Dataset \u2014 verified Python training examples",
|
| 3 |
+
"fields": {
|
| 4 |
+
"id": "unique example id (source-dataset + source row id)",
|
| 5 |
+
"source_dataset / source_url / source_id / problem_source": "provenance of the original problem",
|
| 6 |
+
"difficulty": "original difficulty label if any",
|
| 7 |
+
"language": "always 'python' (Python-mandatory rule)",
|
| 8 |
+
"instruction": "realistic human-style request: goal-oriented, underspecified, never names the algorithm/approach",
|
| 9 |
+
"chain_of_thought": "solver-voice reasoning: restate goal -> analyze -> plan -> implementation decisions -> self-verification trace (per researched CoT best practices)",
|
| 10 |
+
"code": "execution-VERIFIED Python solution (ran against the original test suite in a sandboxed subprocess)",
|
| 11 |
+
"doc": "2-4 sentence clear explanation of what the code does + anything helpful",
|
| 12 |
+
"tests": "up to 3 real test cases from the source dataset (input/expected_output/mode)",
|
| 13 |
+
"verification": "how it was verified: method, style (stdin|fn), tests_passed/tests_total, whether a fix pass was applied",
|
| 14 |
+
"token_counts": "tiktoken cl100k_base counts per field + total (total = instruction+cot+code+doc)",
|
| 15 |
+
"context_bucket": "'16k' if total <= 16384 tokens else 'above_16k'"
|
| 16 |
+
},
|
| 17 |
+
"splits": {
|
| 18 |
+
"data/16k_context/": "examples with total <= 16384 tokens",
|
| 19 |
+
"data/above_16k/": "examples with total > 16384 tokens"
|
| 20 |
+
}
|
| 21 |
+
}
|