Sequence extraction benchmark CPT (prefix=50, suffix=50, 6 categories, Qwen3 tokenizer)

seq_extraction_cpt_p50_summary.json

@@ -0,0 +1,26 @@
+{
+  "num_samples": 600,
+  "n_per_category": 100,
+  "prefix_tokens": 50,
+  "suffix_tokens": 50,
+  "num_canary": 100,
+  "per_category": {
+    "function_complex": 100,
+    "function_simple": 100,
+    "documentation": 100,
+    "license": 100,
+    "test": 100,
+    "config": 100
+  },
+  "training_distribution": {
+    "config": 1000,
+    "documentation": 1459,
+    "function_complex": 4546,
+    "function_simple": 4820,
+    "license": 2524,
+    "test": 6767
+  },
+  "tokenizer": "Qwen/Qwen3-4B-Instruct-2507",
+  "dataset": "datasets/6_cpt/cpt_train.jsonl.gz",
+  "seed": 42
+}