Upload runs/exp_t2_grid_separatedQA_20260426_h100_vllm_full/zs_raw/metrics.json with huggingface_hub

runs/exp_t2_grid_separatedQA_20260426_h100_vllm_full/zs_raw/metrics.json

@@ -0,0 +1,86 @@
+{
+  "name": "pair_prediction-zeroshot-raw",
+  "task_type": "pair_prediction",
+  "model_tag": "Qwen3.5-2B-zeroshot-raw",
+  "metrics": {
+    "accuracy": 0.5000080599661482,
+    "f1": 6.984928673132204e-05,
+    "precision": 0.65,
+    "recall": 3.492651997528277e-05,
+    "parse_rate": 1.0,
+    "n": 744420.0
+  },
+  "per_cell_type": {
+    "Ast": {
+      "accuracy": 0.4999794336013821,
+      "f1": 0.0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "parse_rate": 1.0,
+      "n": 97246.0
+    },
+    "End": {
+      "accuracy": 0.5,
+      "f1": 0.0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "parse_rate": 1.0,
+      "n": 16718.0
+    },
+    "Ex": {
+      "accuracy": 0.5000116160208159,
+      "f1": 9.292169024554559e-05,
+      "precision": 0.6666666666666666,
+      "recall": 4.646408326363721e-05,
+      "parse_rate": 1.0,
+      "n": 172176.0
+    },
+    "In": {
+      "accuracy": 0.5000196571788017,
+      "f1": 0.00011793379982703042,
+      "precision": 0.75,
+      "recall": 5.897153640509514e-05,
+      "parse_rate": 1.0,
+      "n": 101744.0
+    },
+    "Mic": {
+      "accuracy": 0.5,
+      "f1": 2.672724842977416e-05,
+      "precision": 0.5,
+      "recall": 1.3363981397337895e-05,
+      "parse_rate": 1.0,
+      "n": 149656.0
+    },
+    "OPC": {
+      "accuracy": 0.5000124495792042,
+      "f1": 4.979707691158529e-05,
+      "precision": 1.0,
+      "recall": 2.4899158408445795e-05,
+      "parse_rate": 1.0,
+      "n": 80324.0
+    },
+    "Oli": {
+      "accuracy": 0.5000237049211417,
+      "f1": 0.000126416257130667,
+      "precision": 0.8,
+      "recall": 6.321312304434401e-05,
+      "parse_rate": 1.0,
+      "n": 126556.0
+    }
+  },
+  "wall_seconds": 6943.698348045349,
+  "config": {
+    "task": "pair_prediction",
+    "mode": "zeroshot",
+    "prompt": "raw",
+    "model": "Qwen/Qwen3.5-2B",
+    "eval_jsonl": "/dev/shm/dnathinker/data/prod_full_test/jsonl/test.pair_prediction.jsonl",
+    "output_dir": "/dev/shm/dnathinker/runs/exp_t2_grid_separatedQA_20260426_h100_vllm_full/zs_raw",
+    "n": 744420,
+    "max_new_tokens": 128,
+    "max_model_len": 4096,
+    "gpu_mem_frac": 0.85,
+    "temperature": 0.0,
+    "engine": "vllm"
+  }
+}