Add Fine-Tuned benchmark results (85.3% accuracy) from checkpoint-25

- Update system3_inference.py to load LoRA adapter directly from
checkpoint-25/ instead of requiring a merged finetuned_model/ dir
- Add fill_r3.py: standalone script to populate r3 metrics in the
existing benchmark cache without re-running API calls
- Update data/benchmark_cache.json with real Fine-Tuned metrics for
all 50 questions (accuracy 85.3%, ROUGE-L 0.283, HR 0.0%)
- Update README benchmark table with real Fine-Tuned numbers

README.md

@@ -61,16 +61,16 @@ Covers CS fundamentals, algorithms, web technologies, databases, and tooling:
 
 | Metric | Baseline LLM | RAG Chatbot | Fine-Tuned | Unit |
 |--------|-------------|-------------|------------|------|
-| Answer Accuracy | 81.5 | 79.9 | — | % |
-| ROUGE-L | 0.216 | 0.291 | — | 0–1 |
-| Groundedness | 0.815 | 0.575 | — | 0–1 |
-| Answer Relevance | 0.720 | 0.768 | — | 0–1 |
-| Faithfulness | 0.216 | 0.172 | — | 0–1 |
-| Hallucination Rate | 0.0 | 6.0 | — | % |
-| Avg Response Time | 1.46 | 4.19 | — | sec |
-| Cost per Query | $0.0020 | $0.0020 | — | USD |
-
-> Fine-Tuned column shows `—` until `finetuned_model/` is placed in the project root (see Setup §4).  
+| Answer Accuracy | 81.5 | 79.9 | **85.3** | % |
+| ROUGE-L | 0.216 | 0.291 | **0.283** | 0–1 |
+| Groundedness | 0.815 | 0.575 | **0.853** | 0–1 |
+| Answer Relevance | 0.720 | 0.768 | **0.758** | 0–1 |
+| Faithfulness | 0.216 | 0.172 | **0.283** | 0–1 |
+| Hallucination Rate | 0.0 | 6.0 | **0.0** | % |
+| Avg Response Time | 1.46 | 4.19 | 169.2 | sec |
+| Cost per Query | $0.0020 | $0.0020 | **$0.0002** | USD |
+
+> Fine-Tuned runs locally from `checkpoint-25/` (LoRA adapter on Qwen2.5-1.5B-Instruct). High latency is expected on CPU — GPU inference brings it to ~5s.  
 > Full per-question results are in `data/benchmark_cache.json`.
 
 ---

data/benchmark_cache.json

@@ -3,1200 +3,1200 @@
     "question": "what is binary search?",
     "r1_time": 2.76,
     "r2_time": 0.67,
-    "r3_time": 0,
+    "r3_time": 43.48,
     "r1_rouge": 0.148,
     "r2_rouge": 0.6,
-    "r3_rouge": 0,
+    "r3_rouge": 0.271,
     "r1_sim": 0.754,
     "r2_sim": 0.894,
-    "r3_sim": 0,
+    "r3_sim": 0.809,
     "r1_ground": 0.754,
     "r2_ground": 0.727,
-    "r3_ground": 0,
+    "r3_ground": 0.809,
     "r1_relev": 0.639,
     "r2_relev": 0.785,
-    "r3_relev": 0,
+    "r3_relev": 0.648,
     "r1_faith": 0.148,
     "r2_faith": 0.258,
-    "r3_faith": 0,
+    "r3_faith": 0.271,
     "r1_cost": 0.0023,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is the time complexity of binary search?",
     "r1_time": 0.51,
     "r2_time": 0.21,
-    "r3_time": 0,
+    "r3_time": 11.84,
     "r1_rouge": 0.184,
     "r2_rouge": 0.245,
-    "r3_rouge": 0,
+    "r3_rouge": 0.231,
     "r1_sim": 0.892,
     "r2_sim": 0.892,
-    "r3_sim": 0,
+    "r3_sim": 0.805,
     "r1_ground": 0.892,
     "r2_ground": 0.6,
-    "r3_ground": 0,
+    "r3_ground": 0.805,
     "r1_relev": 0.845,
     "r2_relev": 0.94,
-    "r3_relev": 0,
+    "r3_relev": 0.802,
     "r1_faith": 0.184,
     "r2_faith": 0.113,
-    "r3_faith": 0,
+    "r3_faith": 0.231,
     "r1_cost": 0.002,
     "r2_cost": 0.0016,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is the difference between stack and queue?",
     "r1_time": 0.59,
     "r2_time": 0.31,
-    "r3_time": 0,
+    "r3_time": 35.97,
     "r1_rouge": 0.297,
     "r2_rouge": 0.321,
-    "r3_rouge": 0,
+    "r3_rouge": 0.344,
     "r1_sim": 0.822,
     "r2_sim": 0.811,
-    "r3_sim": 0,
+    "r3_sim": 0.859,
     "r1_ground": 0.822,
     "r2_ground": 0.723,
-    "r3_ground": 0,
+    "r3_ground": 0.859,
     "r1_relev": 0.743,
     "r2_relev": 0.806,
-    "r3_relev": 0,
+    "r3_relev": 0.72,
     "r1_faith": 0.297,
     "r2_faith": 0.17,
-    "r3_faith": 0,
+    "r3_faith": 0.344,
     "r1_cost": 0.0022,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "stack vs queue?",
     "r1_time": 0.91,
     "r2_time": 0.92,
-    "r3_time": 0,
+    "r3_time": 54.39,
     "r1_rouge": 0.301,
     "r2_rouge": 0.161,
-    "r3_rouge": 0,
+    "r3_rouge": 0.38,
     "r1_sim": 0.827,
     "r2_sim": 0.716,
-    "r3_sim": 0,
+    "r3_sim": 0.843,
     "r1_ground": 0.827,
     "r2_ground": 0.883,
-    "r3_ground": 0,
+    "r3_ground": 0.843,
     "r1_relev": 0.701,
     "r2_relev": 0.757,
-    "r3_relev": 0,
+    "r3_relev": 0.73,
     "r1_faith": 0.301,
     "r2_faith": 0.292,
-    "r3_faith": 0,
+    "r3_faith": 0.38,
     "r1_cost": 0.0022,
     "r2_cost": 0.0039,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "explain merge sort.",
     "r1_time": 0.55,
     "r2_time": 0.72,
-    "r3_time": 0,
+    "r3_time": 5988.36,
     "r1_rouge": 0.211,
     "r2_rouge": 0.252,
-    "r3_rouge": 0,
+    "r3_rouge": 0.267,
     "r1_sim": 0.76,
     "r2_sim": 0.885,
-    "r3_sim": 0,
+    "r3_sim": 0.812,
     "r1_ground": 0.76,
     "r2_ground": 0.662,
-    "r3_ground": 0,
+    "r3_ground": 0.812,
     "r1_relev": 0.632,
     "r2_relev": 0.754,
-    "r3_relev": 0,
+    "r3_relev": 0.644,
     "r1_faith": 0.211,
     "r2_faith": 0.191,
-    "r3_faith": 0,
+    "r3_faith": 0.267,
     "r1_cost": 0.0021,
     "r2_cost": 0.0041,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is quicksort?",
     "r1_time": 0.46,
     "r2_time": 1.33,
-    "r3_time": 0,
+    "r3_time": 61.32,
     "r1_rouge": 0.252,
     "r2_rouge": 0.509,
-    "r3_rouge": 0,
+    "r3_rouge": 0.351,
     "r1_sim": 0.788,
     "r2_sim": 0.872,
-    "r3_sim": 0,
+    "r3_sim": 0.885,
     "r1_ground": 0.788,
     "r2_ground": 0.714,
-    "r3_ground": 0,
+    "r3_ground": 0.885,
     "r1_relev": 0.641,
     "r2_relev": 0.736,
-    "r3_relev": 0,
+    "r3_relev": 0.666,
     "r1_faith": 0.252,
     "r2_faith": 0.328,
-    "r3_faith": 0,
+    "r3_faith": 0.351,
     "r1_cost": 0.0022,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a linked list?",
     "r1_time": 0.72,
     "r2_time": 0.46,
-    "r3_time": 0,
+    "r3_time": 63.82,
     "r1_rouge": 0.213,
     "r2_rouge": 0.556,
-    "r3_rouge": 0,
+    "r3_rouge": 0.32,
     "r1_sim": 0.826,
     "r2_sim": 0.918,
-    "r3_sim": 0,
+    "r3_sim": 0.867,
     "r1_ground": 0.826,
     "r2_ground": 0.712,
-    "r3_ground": 0,
+    "r3_ground": 0.867,
     "r1_relev": 0.73,
     "r2_relev": 0.87,
-    "r3_relev": 0,
+    "r3_relev": 0.735,
     "r1_faith": 0.213,
     "r2_faith": 0.255,
-    "r3_faith": 0,
+    "r3_faith": 0.32,
     "r1_cost": 0.0023,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a binary search tree?",
     "r1_time": 0.6,
     "r2_time": 0.41,
-    "r3_time": 0,
+    "r3_time": 58.2,
     "r1_rouge": 0.281,
     "r2_rouge": 0.44,
-    "r3_rouge": 0,
+    "r3_rouge": 0.393,
     "r1_sim": 0.832,
     "r2_sim": 0.95,
-    "r3_sim": 0,
+    "r3_sim": 0.828,
     "r1_ground": 0.832,
     "r2_ground": 0.86,
-    "r3_ground": 0,
+    "r3_ground": 0.828,
     "r1_relev": 0.752,
     "r2_relev": 0.779,
-    "r3_relev": 0,
+    "r3_relev": 0.729,
     "r1_faith": 0.281,
     "r2_faith": 0.266,
-    "r3_faith": 0,
+    "r3_faith": 0.393,
     "r1_cost": 0.0022,
     "r2_cost": 0.0024,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is dynamic programming?",
     "r1_time": 0.65,
     "r2_time": 1.34,
-    "r3_time": 0,
+    "r3_time": 58.48,
     "r1_rouge": 0.171,
     "r2_rouge": 0.429,
-    "r3_rouge": 0,
+    "r3_rouge": 0.291,
     "r1_sim": 0.822,
     "r2_sim": 0.818,
-    "r3_sim": 0,
+    "r3_sim": 0.867,
     "r1_ground": 0.822,
     "r2_ground": 0.608,
-    "r3_ground": 0,
+    "r3_ground": 0.867,
     "r1_relev": 0.646,
     "r2_relev": 0.742,
-    "r3_relev": 0,
+    "r3_relev": 0.701,
     "r1_faith": 0.171,
     "r2_faith": 0.347,
-    "r3_faith": 0,
+    "r3_faith": 0.291,
     "r1_cost": 0.0023,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what are react hooks?",
     "r1_time": 1.0,
     "r2_time": 0.27,
-    "r3_time": 0,
+    "r3_time": 61.88,
     "r1_rouge": 0.225,
     "r2_rouge": 0.513,
-    "r3_rouge": 0,
+    "r3_rouge": 0.373,
     "r1_sim": 0.91,
     "r2_sim": 0.928,
-    "r3_sim": 0,
+    "r3_sim": 0.947,
     "r1_ground": 0.91,
     "r2_ground": 0.811,
-    "r3_ground": 0,
+    "r3_ground": 0.947,
     "r1_relev": 0.697,
     "r2_relev": 0.837,
-    "r3_relev": 0,
+    "r3_relev": 0.792,
     "r1_faith": 0.225,
     "r2_faith": 0.314,
-    "r3_faith": 0,
+    "r3_faith": 0.373,
     "r1_cost": 0.0022,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is usestate in react?",
     "r1_time": 0.93,
     "r2_time": 0.95,
-    "r3_time": 0,
+    "r3_time": 59.19,
     "r1_rouge": 0.221,
     "r2_rouge": 0.387,
-    "r3_rouge": 0,
+    "r3_rouge": 0.276,
     "r1_sim": 0.786,
     "r2_sim": 0.831,
-    "r3_sim": 0,
+    "r3_sim": 0.85,
     "r1_ground": 0.786,
     "r2_ground": 0.692,
-    "r3_ground": 0,
+    "r3_ground": 0.85,
     "r1_relev": 0.765,
     "r2_relev": 0.835,
-    "r3_relev": 0,
+    "r3_relev": 0.676,
     "r1_faith": 0.221,
     "r2_faith": 0.222,
-    "r3_faith": 0,
+    "r3_faith": 0.276,
     "r1_cost": 0.0022,
     "r2_cost": 0.0023,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is useeffect in react?",
     "r1_time": 0.84,
     "r2_time": 0.51,
-    "r3_time": 0,
+    "r3_time": 58.83,
     "r1_rouge": 0.207,
     "r2_rouge": 0.612,
-    "r3_rouge": 0,
+    "r3_rouge": 0.245,
     "r1_sim": 0.772,
     "r2_sim": 0.92,
-    "r3_sim": 0,
+    "r3_sim": 0.839,
     "r1_ground": 0.772,
     "r2_ground": 0.844,
-    "r3_ground": 0,
+    "r3_ground": 0.839,
     "r1_relev": 0.758,
     "r2_relev": 0.804,
-    "r3_relev": 0,
+    "r3_relev": 0.765,
     "r1_faith": 0.207,
     "r2_faith": 0.337,
-    "r3_faith": 0,
+    "r3_faith": 0.245,
     "r1_cost": 0.0021,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a rest api?",
     "r1_time": 0.76,
     "r2_time": 2.56,
-    "r3_time": 0,
+    "r3_time": 58.8,
     "r1_rouge": 0.21,
     "r2_rouge": 0.409,
-    "r3_rouge": 0,
+    "r3_rouge": 0.298,
     "r1_sim": 0.771,
     "r2_sim": 0.935,
-    "r3_sim": 0,
+    "r3_sim": 0.875,
     "r1_ground": 0.771,
     "r2_ground": 0.757,
-    "r3_ground": 0,
+    "r3_ground": 0.875,
     "r1_relev": 0.753,
     "r2_relev": 0.777,
-    "r3_relev": 0,
+    "r3_relev": 0.733,
     "r1_faith": 0.21,
     "r2_faith": 0.362,
-    "r3_faith": 0,
+    "r3_faith": 0.298,
     "r1_cost": 0.0022,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what http methods does rest use?",
     "r1_time": 0.67,
     "r2_time": 3.38,
-    "r3_time": 0,
+    "r3_time": 38.29,
     "r1_rouge": 0.271,
     "r2_rouge": 0.277,
-    "r3_rouge": 0,
+    "r3_rouge": 0.4,
     "r1_sim": 0.841,
     "r2_sim": 0.749,
-    "r3_sim": 0,
+    "r3_sim": 0.815,
     "r1_ground": 0.841,
     "r2_ground": 0.635,
-    "r3_ground": 0,
+    "r3_ground": 0.815,
     "r1_relev": 0.753,
     "r2_relev": 0.685,
-    "r3_relev": 0,
+    "r3_relev": 0.774,
     "r1_faith": 0.271,
     "r2_faith": 0.135,
-    "r3_faith": 0,
+    "r3_faith": 0.4,
     "r1_cost": 0.0019,
     "r2_cost": 0.0017,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what does http status code 404 mean?",
     "r1_time": 0.45,
     "r2_time": 4.3,
-    "r3_time": 0,
+    "r3_time": 25.47,
     "r1_rouge": 0.187,
     "r2_rouge": 0.286,
-    "r3_rouge": 0,
+    "r3_rouge": 0.286,
     "r1_sim": 0.902,
     "r2_sim": 0.919,
-    "r3_sim": 0,
+    "r3_sim": 0.899,
     "r1_ground": 0.902,
     "r2_ground": 0.646,
-    "r3_ground": 0,
+    "r3_ground": 0.899,
     "r1_relev": 0.796,
     "r2_relev": 0.86,
-    "r3_relev": 0,
+    "r3_relev": 0.906,
     "r1_faith": 0.187,
     "r2_faith": 0.124,
-    "r3_faith": 0,
+    "r3_faith": 0.286,
     "r1_cost": 0.0017,
     "r2_cost": 0.0016,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is css flexbox?",
     "r1_time": 0.57,
     "r2_time": 5.92,
-    "r3_time": 0,
+    "r3_time": 59.45,
     "r1_rouge": 0.225,
     "r2_rouge": 0.405,
-    "r3_rouge": 0,
+    "r3_rouge": 0.274,
     "r1_sim": 0.846,
     "r2_sim": 0.93,
-    "r3_sim": 0,
+    "r3_sim": 0.891,
     "r1_ground": 0.846,
     "r2_ground": 0.919,
-    "r3_ground": 0,
+    "r3_ground": 0.891,
     "r1_relev": 0.777,
     "r2_relev": 0.826,
-    "r3_relev": 0,
+    "r3_relev": 0.696,
     "r1_faith": 0.225,
     "r2_faith": 0.389,
-    "r3_faith": 0,
+    "r3_faith": 0.274,
     "r1_cost": 0.0022,
     "r2_cost": 0.0017,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a javascript promise?",
     "r1_time": 1.11,
     "r2_time": 4.46,
-    "r3_time": 0,
+    "r3_time": 59.11,
     "r1_rouge": 0.161,
     "r2_rouge": 0.259,
-    "r3_rouge": 0,
+    "r3_rouge": 0.196,
     "r1_sim": 0.774,
     "r2_sim": 0.829,
-    "r3_sim": 0,
+    "r3_sim": 0.75,
     "r1_ground": 0.774,
     "r2_ground": 0.7,
-    "r3_ground": 0,
+    "r3_ground": 0.75,
     "r1_relev": 0.683,
     "r2_relev": 0.87,
-    "r3_relev": 0,
+    "r3_relev": 0.76,
     "r1_faith": 0.161,
     "r2_faith": 0.28,
-    "r3_faith": 0,
+    "r3_faith": 0.196,
     "r1_cost": 0.0023,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is inorder traversal?",
     "r1_time": 2.38,
     "r2_time": 4.85,
-    "r3_time": 0,
+    "r3_time": 57.69,
     "r1_rouge": 0.216,
     "r2_rouge": 0.415,
-    "r3_rouge": 0,
+    "r3_rouge": 0.218,
     "r1_sim": 0.869,
     "r2_sim": 0.847,
-    "r3_sim": 0,
+    "r3_sim": 0.853,
     "r1_ground": 0.869,
     "r2_ground": 0.69,
-    "r3_ground": 0,
+    "r3_ground": 0.853,
     "r1_relev": 0.772,
     "r2_relev": 0.854,
-    "r3_relev": 0,
+    "r3_relev": 0.834,
     "r1_faith": 0.216,
     "r2_faith": 0.127,
-    "r3_faith": 0,
+    "r3_faith": 0.218,
     "r1_cost": 0.002,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is the difference between bfs and dfs?",
     "r1_time": 0.62,
     "r2_time": 6.27,
-    "r3_time": 0,
+    "r3_time": 59.32,
     "r1_rouge": 0.268,
     "r2_rouge": 0.273,
-    "r3_rouge": 0,
+    "r3_rouge": 0.202,
     "r1_sim": 0.855,
     "r2_sim": 0.913,
-    "r3_sim": 0,
+    "r3_sim": 0.866,
     "r1_ground": 0.855,
     "r2_ground": 0.757,
-    "r3_ground": 0,
+    "r3_ground": 0.866,
     "r1_relev": 0.479,
     "r2_relev": 0.706,
-    "r3_relev": 0,
+    "r3_relev": 0.675,
     "r1_faith": 0.268,
     "r2_faith": 0.227,
-    "r3_faith": 0,
+    "r3_faith": 0.202,
     "r1_cost": 0.0022,
     "r2_cost": 0.003,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is memoization?",
     "r1_time": 0.62,
     "r2_time": 5.53,
-    "r3_time": 0,
+    "r3_time": 59.06,
     "r1_rouge": 0.245,
     "r2_rouge": 0.307,
-    "r3_rouge": 0,
+    "r3_rouge": 0.267,
     "r1_sim": 0.793,
     "r2_sim": 0.884,
-    "r3_sim": 0,
+    "r3_sim": 0.898,
     "r1_ground": 0.793,
     "r2_ground": 0.523,
-    "r3_ground": 0,
+    "r3_ground": 0.898,
     "r1_relev": 0.544,
     "r2_relev": 0.762,
-    "r3_relev": 0,
+    "r3_relev": 0.729,
     "r1_faith": 0.245,
     "r2_faith": 0.117,
-    "r3_faith": 0,
+    "r3_faith": 0.267,
     "r1_cost": 0.0022,
     "r2_cost": 0.0023,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is flex-grow in css?",
     "r1_time": 1.71,
     "r2_time": 5.43,
-    "r3_time": 0,
+    "r3_time": 59.97,
     "r1_rouge": 0.242,
     "r2_rouge": 0.505,
-    "r3_rouge": 0,
+    "r3_rouge": 0.33,
     "r1_sim": 0.888,
     "r2_sim": 0.931,
-    "r3_sim": 0,
+    "r3_sim": 0.863,
     "r1_ground": 0.888,
     "r2_ground": 0.732,
-    "r3_ground": 0,
+    "r3_ground": 0.863,
     "r1_relev": 0.782,
     "r2_relev": 0.846,
-    "r3_relev": 0,
+    "r3_relev": 0.712,
     "r1_faith": 0.242,
     "r2_faith": 0.176,
-    "r3_faith": 0,
+    "r3_faith": 0.33,
     "r1_cost": 0.0021,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a hash table?",
     "r1_time": 1.85,
     "r2_time": 4.5,
-    "r3_time": 0,
+    "r3_time": 52.51,
     "r1_rouge": 0.19,
     "r2_rouge": 0.359,
-    "r3_rouge": 0,
+    "r3_rouge": 0.44,
     "r1_sim": 0.763,
     "r2_sim": 0.909,
-    "r3_sim": 0,
+    "r3_sim": 0.95,
     "r1_ground": 0.763,
     "r2_ground": 0.358,
-    "r3_ground": 0,
+    "r3_ground": 0.95,
     "r1_relev": 0.717,
     "r2_relev": 0.851,
-    "r3_relev": 0,
+    "r3_relev": 0.772,
     "r1_faith": 0.19,
     "r2_faith": 0.158,
-    "r3_faith": 0,
+    "r3_faith": 0.44,
     "r1_cost": 0.0023,
     "r2_cost": 0.0022,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is time complexity?",
     "r1_time": 1.84,
     "r2_time": 5.43,
-    "r3_time": 0,
+    "r3_time": 57.57,
     "r1_rouge": 0.273,
     "r2_rouge": 0.37,
-    "r3_rouge": 0,
+    "r3_rouge": 0.335,
     "r1_sim": 0.871,
     "r2_sim": 0.875,
-    "r3_sim": 0,
+    "r3_sim": 0.893,
     "r1_ground": 0.871,
     "r2_ground": 0.357,
-    "r3_ground": 0,
+    "r3_ground": 0.893,
     "r1_relev": 0.714,
     "r2_relev": 0.824,
-    "r3_relev": 0,
+    "r3_relev": 0.7,
     "r1_faith": 0.273,
     "r2_faith": 0.109,
-    "r3_faith": 0,
+    "r3_faith": 0.335,
     "r1_cost": 0.0024,
     "r2_cost": 0.002,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is space complexity?",
     "r1_time": 2.0,
     "r2_time": 4.34,
-    "r3_time": 0,
+    "r3_time": 58.56,
     "r1_rouge": 0.213,
     "r2_rouge": 0.33,
-    "r3_rouge": 0,
+    "r3_rouge": 0.245,
     "r1_sim": 0.792,
     "r2_sim": 0.84,
-    "r3_sim": 0,
+    "r3_sim": 0.819,
     "r1_ground": 0.792,
     "r2_ground": 0.354,
-    "r3_ground": 0,
+    "r3_ground": 0.819,
     "r1_relev": 0.727,
     "r2_relev": 0.842,
-    "r3_relev": 0,
+    "r3_relev": 0.76,
     "r1_faith": 0.213,
     "r2_faith": 0.079,
-    "r3_faith": 0,
+    "r3_faith": 0.245,
     "r1_cost": 0.0023,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is recursion?",
     "r1_time": 1.49,
     "r2_time": 5.38,
-    "r3_time": 0,
+    "r3_time": 57.38,
     "r1_rouge": 0.264,
     "r2_rouge": 0.321,
-    "r3_rouge": 0,
+    "r3_rouge": 0.265,
     "r1_sim": 0.806,
     "r2_sim": 0.926,
-    "r3_sim": 0,
+    "r3_sim": 0.819,
     "r1_ground": 0.806,
     "r2_ground": 0.455,
-    "r3_ground": 0,
+    "r3_ground": 0.819,
     "r1_relev": 0.7,
     "r2_relev": 0.875,
-    "r3_relev": 0,
+    "r3_relev": 0.74,
     "r1_faith": 0.264,
     "r2_faith": 0.078,
-    "r3_faith": 0,
+    "r3_faith": 0.265,
     "r1_cost": 0.0022,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a heap data structure?",
     "r1_time": 1.6,
     "r2_time": 5.53,
-    "r3_time": 0,
+    "r3_time": 57.95,
     "r1_rouge": 0.239,
     "r2_rouge": 0.393,
-    "r3_rouge": 0,
+    "r3_rouge": 0.325,
     "r1_sim": 0.85,
     "r2_sim": 0.844,
-    "r3_sim": 0,
+    "r3_sim": 0.872,
     "r1_ground": 0.85,
     "r2_ground": 0.591,
-    "r3_ground": 0,
+    "r3_ground": 0.872,
     "r1_relev": 0.727,
     "r2_relev": 0.847,
-    "r3_relev": 0,
+    "r3_relev": 0.731,
     "r1_faith": 0.239,
     "r2_faith": 0.148,
-    "r3_faith": 0,
+    "r3_faith": 0.325,
     "r1_cost": 0.0023,
     "r2_cost": 0.0026,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a graph data structure?",
     "r1_time": 1.84,
     "r2_time": 5.63,
-    "r3_time": 0,
+    "r3_time": 50.98,
     "r1_rouge": 0.211,
     "r2_rouge": 0.327,
-    "r3_rouge": 0,
+    "r3_rouge": 0.26,
     "r1_sim": 0.768,
     "r2_sim": 0.689,
-    "r3_sim": 0,
+    "r3_sim": 0.744,
     "r1_ground": 0.768,
     "r2_ground": 0.635,
-    "r3_ground": 0,
+    "r3_ground": 0.744,
     "r1_relev": 0.742,
     "r2_relev": 0.906,
-    "r3_relev": 0,
+    "r3_relev": 0.831,
     "r1_faith": 0.211,
     "r2_faith": 0.191,
-    "r3_faith": 0,
+    "r3_faith": 0.26,
     "r1_cost": 0.0021,
     "r2_cost": 0.002,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is depth first search?",
     "r1_time": 1.77,
     "r2_time": 6.19,
-    "r3_time": 0,
+    "r3_time": 52.07,
     "r1_rouge": 0.183,
     "r2_rouge": 0.541,
-    "r3_rouge": 0,
+    "r3_rouge": 0.282,
     "r1_sim": 0.855,
     "r2_sim": 0.957,
-    "r3_sim": 0,
+    "r3_sim": 0.938,
     "r1_ground": 0.855,
     "r2_ground": 0.685,
-    "r3_ground": 0,
+    "r3_ground": 0.938,
     "r1_relev": 0.562,
     "r2_relev": 0.673,
-    "r3_relev": 0,
+    "r3_relev": 0.662,
     "r1_faith": 0.183,
     "r2_faith": 0.24,
-    "r3_faith": 0,
+    "r3_faith": 0.282,
     "r1_cost": 0.0022,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is breadth first search?",
     "r1_time": 0.63,
     "r2_time": 5.43,
-    "r3_time": 0,
+    "r3_time": 49.98,
     "r1_rouge": 0.22,
     "r2_rouge": 0.429,
-    "r3_rouge": 0,
+    "r3_rouge": 0.33,
     "r1_sim": 0.852,
     "r2_sim": 0.978,
-    "r3_sim": 0,
+    "r3_sim": 0.957,
     "r1_ground": 0.852,
     "r2_ground": 0.728,
-    "r3_ground": 0,
+    "r3_ground": 0.957,
     "r1_relev": 0.713,
     "r2_relev": 0.8,
-    "r3_relev": 0,
+    "r3_relev": 0.816,
     "r1_faith": 0.22,
     "r2_faith": 0.272,
-    "r3_faith": 0,
+    "r3_faith": 0.33,
     "r1_cost": 0.0022,
     "r2_cost": 0.0021,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a hash map?",
     "r1_time": 1.71,
     "r2_time": 7.17,
-    "r3_time": 0,
+    "r3_time": 50.37,
     "r1_rouge": 0.229,
     "r2_rouge": 0.256,
-    "r3_rouge": 0,
+    "r3_rouge": 0.239,
     "r1_sim": 0.796,
     "r2_sim": 0.87,
-    "r3_sim": 0,
+    "r3_sim": 0.838,
     "r1_ground": 0.796,
     "r2_ground": 0.386,
-    "r3_ground": 0,
+    "r3_ground": 0.838,
     "r1_relev": 0.655,
     "r2_relev": 0.759,
-    "r3_relev": 0,
+    "r3_relev": 0.71,
     "r1_faith": 0.229,
     "r2_faith": 0.15,
-    "r3_faith": 0,
+    "r3_faith": 0.239,
     "r1_cost": 0.0023,
     "r2_cost": 0.0032,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is big o notation?",
     "r1_time": 1.71,
     "r2_time": 5.95,
-    "r3_time": 0,
+    "r3_time": 49.92,
     "r1_rouge": 0.239,
     "r2_rouge": 0.331,
-    "r3_rouge": 0,
+    "r3_rouge": 0.305,
     "r1_sim": 0.851,
     "r2_sim": 0.904,
-    "r3_sim": 0,
+    "r3_sim": 0.895,
     "r1_ground": 0.851,
     "r2_ground": 0.473,
-    "r3_ground": 0,
+    "r3_ground": 0.895,
     "r1_relev": 0.736,
     "r2_relev": 0.817,
-    "r3_relev": 0,
+    "r3_relev": 0.852,
     "r1_faith": 0.239,
     "r2_faith": 0.179,
-    "r3_faith": 0,
+    "r3_faith": 0.305,
     "r1_cost": 0.0023,
     "r2_cost": 0.0034,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a tree data structure?",
     "r1_time": 2.02,
     "r2_time": 5.66,
-    "r3_time": 0,
+    "r3_time": 49.82,
     "r1_rouge": 0.182,
     "r2_rouge": 0.273,
-    "r3_rouge": 0,
+    "r3_rouge": 0.253,
     "r1_sim": 0.798,
     "r2_sim": 0.835,
-    "r3_sim": 0,
+    "r3_sim": 0.804,
     "r1_ground": 0.798,
     "r2_ground": 0.59,
-    "r3_ground": 0,
+    "r3_ground": 0.804,
     "r1_relev": 0.74,
     "r2_relev": 0.899,
-    "r3_relev": 0,
+    "r3_relev": 0.865,
     "r1_faith": 0.182,
     "r2_faith": 0.215,
-    "r3_faith": 0,
+    "r3_faith": 0.253,
     "r1_cost": 0.0022,
     "r2_cost": 0.0022,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is object oriented programming?",
     "r1_time": 4.66,
     "r2_time": 2.78,
-    "r3_time": 0,
+    "r3_time": 50.02,
     "r1_rouge": 0.18,
     "r2_rouge": 0.164,
-    "r3_rouge": 0,
+    "r3_rouge": 0.374,
     "r1_sim": 0.881,
     "r2_sim": 0.897,
-    "r3_sim": 0,
+    "r3_sim": 0.892,
     "r1_ground": 0.881,
     "r2_ground": 0.345,
-    "r3_ground": 0,
+    "r3_ground": 0.892,
     "r1_relev": 0.754,
     "r2_relev": 0.752,
-    "r3_relev": 0,
+    "r3_relev": 0.78,
     "r1_faith": 0.18,
     "r2_faith": 0.089,
-    "r3_faith": 0,
+    "r3_faith": 0.374,
     "r1_cost": 0.0022,
     "r2_cost": 0.0038,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is inheritance in oop?",
     "r1_time": 3.5,
     "r2_time": 6.5,
-    "r3_time": 0,
+    "r3_time": 51.25,
     "r1_rouge": 0.187,
     "r2_rouge": 0.108,
-    "r3_rouge": 0,
+    "r3_rouge": 0.222,
     "r1_sim": 0.821,
     "r2_sim": 0.782,
-    "r3_sim": 0,
+    "r3_sim": 0.829,
     "r1_ground": 0.821,
     "r2_ground": 0.382,
-    "r3_ground": 0,
+    "r3_ground": 0.829,
     "r1_relev": 0.731,
     "r2_relev": 0.792,
-    "r3_relev": 0,
+    "r3_relev": 0.656,
     "r1_faith": 0.187,
     "r2_faith": 0.077,
-    "r3_faith": 0,
+    "r3_faith": 0.222,
     "r1_cost": 0.0021,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is polymorphism in oop?",
     "r1_time": 0.79,
     "r2_time": 6.04,
-    "r3_time": 0,
+    "r3_time": 51.84,
     "r1_rouge": 0.188,
     "r2_rouge": 0.145,
-    "r3_rouge": 0,
+    "r3_rouge": 0.205,
     "r1_sim": 0.759,
     "r2_sim": 0.65,
-    "r3_sim": 0,
+    "r3_sim": 0.797,
     "r1_ground": 0.759,
     "r2_ground": 0.473,
-    "r3_ground": 0,
+    "r3_ground": 0.797,
     "r1_relev": 0.662,
     "r2_relev": 0.598,
-    "r3_relev": 0,
+    "r3_relev": 0.741,
     "r1_faith": 0.188,
     "r2_faith": 0.088,
-    "r3_faith": 0,
+    "r3_faith": 0.205,
     "r1_cost": 0.0021,
     "r2_cost": 0.0034,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is encapsulation in oop?",
     "r1_time": 2.73,
     "r2_time": 6.14,
-    "r3_time": 0,
+    "r3_time": 49.75,
     "r1_rouge": 0.128,
     "r2_rouge": 0.078,
-    "r3_rouge": 0,
+    "r3_rouge": 0.193,
     "r1_sim": 0.661,
     "r2_sim": 0.569,
-    "r3_sim": 0,
+    "r3_sim": 0.72,
     "r1_ground": 0.661,
     "r2_ground": 0.511,
-    "r3_ground": 0,
+    "r3_ground": 0.72,
     "r1_relev": 0.801,
     "r2_relev": 0.823,
-    "r3_relev": 0,
+    "r3_relev": 0.819,
     "r1_faith": 0.128,
     "r2_faith": 0.066,
-    "r3_faith": 0,
+    "r3_faith": 0.193,
     "r1_cost": 0.0022,
     "r2_cost": 0.002,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is async await in javascript?",
     "r1_time": 0.63,
     "r2_time": 6.14,
-    "r3_time": 0,
+    "r3_time": 49.53,
     "r1_rouge": 0.186,
     "r2_rouge": 0.301,
-    "r3_rouge": 0,
+    "r3_rouge": 0.259,
     "r1_sim": 0.815,
     "r2_sim": 0.837,
-    "r3_sim": 0,
+    "r3_sim": 0.778,
     "r1_ground": 0.815,
     "r2_ground": 0.772,
-    "r3_ground": 0,
+    "r3_ground": 0.778,
     "r1_relev": 0.792,
     "r2_relev": 0.835,
-    "r3_relev": 0,
+    "r3_relev": 0.737,
     "r1_faith": 0.186,
     "r2_faith": 0.276,
-    "r3_faith": 0,
+    "r3_faith": 0.259,
     "r1_cost": 0.0023,
     "r2_cost": 0.0036,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a closure in javascript?",
     "r1_time": 3.34,
     "r2_time": 5.94,
-    "r3_time": 0,
+    "r3_time": 54.74,
     "r1_rouge": 0.252,
     "r2_rouge": 0.253,
-    "r3_rouge": 0,
+    "r3_rouge": 0.276,
     "r1_sim": 0.842,
     "r2_sim": 0.871,
-    "r3_sim": 0,
+    "r3_sim": 0.807,
     "r1_ground": 0.842,
     "r2_ground": 0.4,
-    "r3_ground": 0,
+    "r3_ground": 0.807,
     "r1_relev": 0.797,
     "r2_relev": 0.84,
-    "r3_relev": 0,
+    "r3_relev": 0.793,
     "r1_faith": 0.252,
     "r2_faith": 0.123,
-    "r3_faith": 0,
+    "r3_faith": 0.276,
     "r1_cost": 0.0022,
     "r2_cost": 0.0035,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is the event loop in javascript?",
     "r1_time": 3.18,
     "r2_time": 5.21,
-    "r3_time": 0,
+    "r3_time": 49.3,
     "r1_rouge": 0.169,
     "r2_rouge": 0.182,
-    "r3_rouge": 0,
+    "r3_rouge": 0.219,
     "r1_sim": 0.802,
     "r2_sim": 0.835,
-    "r3_sim": 0,
+    "r3_sim": 0.873,
     "r1_ground": 0.802,
     "r2_ground": 0.381,
-    "r3_ground": 0,
+    "r3_ground": 0.873,
     "r1_relev": 0.766,
     "r2_relev": 0.79,
-    "r3_relev": 0,
+    "r3_relev": 0.8,
     "r1_faith": 0.169,
     "r2_faith": 0.081,
-    "r3_faith": 0,
+    "r3_faith": 0.219,
     "r1_cost": 0.0023,
     "r2_cost": 0.0043,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is typescript?",
     "r1_time": 3.14,
     "r2_time": 4.37,
-    "r3_time": 0,
+    "r3_time": 48.43,
     "r1_rouge": 0.183,
     "r2_rouge": 0.095,
-    "r3_rouge": 0,
+    "r3_rouge": 0.21,
     "r1_sim": 0.847,
     "r2_sim": 0.687,
-    "r3_sim": 0,
+    "r3_sim": 0.87,
     "r1_ground": 0.847,
     "r2_ground": 0.559,
-    "r3_ground": 0,
+    "r3_ground": 0.87,
     "r1_relev": 0.738,
     "r2_relev": 0.739,
-    "r3_relev": 0,
+    "r3_relev": 0.787,
     "r1_faith": 0.183,
     "r2_faith": 0.145,
-    "r3_faith": 0,
+    "r3_faith": 0.21,
     "r1_cost": 0.0022,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is sql?",
     "r1_time": 1.56,
     "r2_time": 5.33,
-    "r3_time": 0,
+    "r3_time": 48.68,
     "r1_rouge": 0.23,
     "r2_rouge": 0.0,
-    "r3_rouge": 0,
+    "r3_rouge": 0.252,
     "r1_sim": 0.805,
     "r2_sim": 0.167,
-    "r3_sim": 0,
+    "r3_sim": 0.792,
     "r1_ground": 0.805,
     "r2_ground": 0.076,
-    "r3_ground": 0,
+    "r3_ground": 0.792,
     "r1_relev": 0.723,
     "r2_relev": 0.236,
-    "r3_relev": 0,
+    "r3_relev": 0.759,
     "r1_faith": 0.23,
     "r2_faith": 0.014,
-    "r3_faith": 0,
+    "r3_faith": 0.252,
     "r1_cost": 0.0022,
     "r2_cost": 0.0016,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a primary key in sql?",
     "r1_time": 1.95,
     "r2_time": 4.38,
-    "r3_time": 0,
+    "r3_time": 42.23,
     "r1_rouge": 0.261,
     "r2_rouge": 0.048,
-    "r3_rouge": 0,
+    "r3_rouge": 0.289,
     "r1_sim": 0.82,
     "r2_sim": 0.378,
-    "r3_sim": 0,
+    "r3_sim": 0.897,
     "r1_ground": 0.82,
     "r2_ground": 0.303,
-    "r3_ground": 0,
+    "r3_ground": 0.897,
     "r1_relev": 0.779,
     "r2_relev": 0.513,
-    "r3_relev": 0,
+    "r3_relev": 0.882,
     "r1_faith": 0.261,
     "r2_faith": 0.059,
-    "r3_faith": 0,
+    "r3_faith": 0.289,
     "r1_cost": 0.0023,
     "r2_cost": 0.0017,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is a foreign key in sql?",
     "r1_time": 2.09,
     "r2_time": 5.47,
-    "r3_time": 0,
+    "r3_time": 48.6,
     "r1_rouge": 0.237,
     "r2_rouge": 0.083,
-    "r3_rouge": 0,
+    "r3_rouge": 0.352,
     "r1_sim": 0.612,
     "r2_sim": 0.597,
-    "r3_sim": 0,
+    "r3_sim": 0.839,
     "r1_ground": 0.612,
     "r2_ground": 0.238,
-    "r3_ground": 0,
+    "r3_ground": 0.839,
     "r1_relev": 0.642,
     "r2_relev": 0.759,
-    "r3_relev": 0,
+    "r3_relev": 0.771,
     "r1_faith": 0.237,
     "r2_faith": 0.048,
-    "r3_faith": 0,
+    "r3_faith": 0.352,
     "r1_cost": 0.0023,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is an index in sql?",
     "r1_time": 0.95,
     "r2_time": 5.37,
-    "r3_time": 0,
+    "r3_time": 50.73,
     "r1_rouge": 0.18,
     "r2_rouge": 0.099,
-    "r3_rouge": 0,
+    "r3_rouge": 0.187,
     "r1_sim": 0.834,
     "r2_sim": 0.705,
-    "r3_sim": 0,
+    "r3_sim": 0.794,
     "r1_ground": 0.834,
     "r2_ground": 0.348,
-    "r3_ground": 0,
+    "r3_ground": 0.794,
     "r1_relev": 0.79,
     "r2_relev": 0.703,
-    "r3_relev": 0,
+    "r3_relev": 0.756,
     "r1_faith": 0.18,
     "r2_faith": 0.021,
-    "r3_faith": 0,
+    "r3_faith": 0.187,
     "r1_cost": 0.0025,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is normalization in databases?",
     "r1_time": 1.71,
     "r2_time": 5.94,
-    "r3_time": 0,
+    "r3_time": 28.71,
     "r1_rouge": 0.183,
     "r2_rouge": 0.099,
-    "r3_rouge": 0,
+    "r3_rouge": 0.311,
     "r1_sim": 0.778,
     "r2_sim": 0.564,
-    "r3_sim": 0,
+    "r3_sim": 0.895,
     "r1_ground": 0.778,
     "r2_ground": 0.451,
-    "r3_ground": 0,
+    "r3_ground": 0.895,
     "r1_relev": 0.759,
     "r2_relev": 0.645,
-    "r3_relev": 0,
+    "r3_relev": 0.858,
     "r1_faith": 0.183,
     "r2_faith": 0.033,
-    "r3_faith": 0,
+    "r3_faith": 0.311,
     "r1_cost": 0.0023,
     "r2_cost": 0.0018,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is git?",
     "r1_time": 0.8,
     "r2_time": 6.36,
-    "r3_time": 0,
+    "r3_time": 31.68,
     "r1_rouge": 0.141,
     "r2_rouge": 0.079,
-    "r3_rouge": 0,
+    "r3_rouge": 0.295,
     "r1_sim": 0.841,
     "r2_sim": 0.644,
-    "r3_sim": 0,
+    "r3_sim": 0.891,
     "r1_ground": 0.841,
     "r2_ground": 0.355,
-    "r3_ground": 0,
+    "r3_ground": 0.891,
     "r1_relev": 0.656,
     "r2_relev": 0.755,
-    "r3_relev": 0,
+    "r3_relev": 0.757,
     "r1_faith": 0.141,
     "r2_faith": 0.021,
-    "r3_faith": 0,
+    "r3_faith": 0.295,
     "r1_cost": 0.0024,
     "r2_cost": 0.0017,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is a merge conflict in git?",
     "r1_time": 1.29,
     "r2_time": 5.36,
-    "r3_time": 0,
+    "r3_time": 60.03,
     "r1_rouge": 0.257,
     "r2_rouge": 0.102,
-    "r3_rouge": 0,
+    "r3_rouge": 0.335,
     "r1_sim": 0.845,
     "r2_sim": 0.64,
-    "r3_sim": 0,
+    "r3_sim": 0.923,
     "r1_ground": 0.845,
     "r2_ground": 0.393,
-    "r3_ground": 0,
+    "r3_ground": 0.923,
     "r1_relev": 0.802,
     "r2_relev": 0.803,
-    "r3_relev": 0,
+    "r3_relev": 0.835,
     "r1_faith": 0.257,
     "r2_faith": 0.082,
-    "r3_faith": 0,
+    "r3_faith": 0.335,
     "r1_cost": 0.0024,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is docker?",
     "r1_time": 0.55,
     "r2_time": 5.55,
-    "r3_time": 0,
+    "r3_time": 53.62,
     "r1_rouge": 0.229,
     "r2_rouge": 0.083,
-    "r3_rouge": 0,
+    "r3_rouge": 0.184,
     "r1_sim": 0.851,
     "r2_sim": 0.43,
-    "r3_sim": 0,
+    "r3_sim": 0.874,
     "r1_ground": 0.851,
     "r2_ground": 0.683,
-    "r3_ground": 0,
+    "r3_ground": 0.874,
     "r1_relev": 0.682,
     "r2_relev": 0.517,
-    "r3_relev": 0,
+    "r3_relev": 0.743,
     "r1_faith": 0.229,
     "r2_faith": 0.072,
-    "r3_faith": 0,
+    "r3_faith": 0.184,
     "r1_cost": 0.0023,
     "r2_cost": 0.0019,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   },
   {
     "question": "what is json?",
     "r1_time": 0.95,
     "r2_time": 5.8,
-    "r3_time": 0,
+    "r3_time": 29.45,
     "r1_rouge": 0.23,
     "r2_rouge": 0.261,
-    "r3_rouge": 0,
+    "r3_rouge": 0.28,
     "r1_sim": 0.78,
     "r2_sim": 0.796,
-    "r3_sim": 0,
+    "r3_sim": 0.888,
     "r1_ground": 0.78,
     "r2_ground": 0.564,
-    "r3_ground": 0,
+    "r3_ground": 0.888,
     "r1_relev": 0.702,
     "r2_relev": 0.614,
-    "r3_relev": 0,
+    "r3_relev": 0.797,
     "r1_faith": 0.23,
     "r2_faith": 0.168,
-    "r3_faith": 0,
+    "r3_faith": 0.28,
     "r1_cost": 0.002,
     "r2_cost": 0.0016,
-    "r3_cost": 0
+    "r3_cost": 0.0001
   },
   {
     "question": "what is css grid?",
     "r1_time": 1.24,
     "r2_time": 4.71,
-    "r3_time": 0,
+    "r3_time": 42.78,
     "r1_rouge": 0.203,
     "r2_rouge": 0.292,
-    "r3_rouge": 0,
+    "r3_rouge": 0.231,
     "r1_sim": 0.833,
     "r2_sim": 0.92,
-    "r3_sim": 0,
+    "r3_sim": 0.849,
     "r1_ground": 0.833,
     "r2_ground": 0.692,
-    "r3_ground": 0,
+    "r3_ground": 0.849,
     "r1_relev": 0.823,
     "r2_relev": 0.764,
-    "r3_relev": 0,
+    "r3_relev": 0.779,
     "r1_faith": 0.203,
     "r2_faith": 0.268,
-    "r3_faith": 0,
+    "r3_faith": 0.231,
     "r1_cost": 0.0022,
     "r2_cost": 0.0032,
-    "r3_cost": 0
+    "r3_cost": 0.0002
   }
 ]

fill_r3.py

@@ -0,0 +1,106 @@
+"""
+Fills r3_* (Fine-Tuned) metrics into existing benchmark_cache.json.
+Run this ONCE after system3_inference.py is updated to use checkpoint-25/.
+
+python fill_r3.py
+"""
+import os, json, time
+import numpy as np
+
+# rouge_score MUST come before any heavy ML imports
+from rouge_score import rouge_scorer as rs
+from dotenv import load_dotenv
+
+load_dotenv()
+
+# ── ML imports AFTER rouge_score ──────────────────────────────────────────────
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_community.vectorstores import FAISS
+from system3_inference import ask_finetuned  # now points to checkpoint-25
+
+# ── Load reference answers ────────────────────────────────────────────────────
+with open("data/reference_answers.json", encoding="utf-8") as f:
+    ref_answers = json.load(f)
+
+# ── Load embeddings + vector store (for metric computation) ──────────────────
+INDEX_PATH = "data/faiss_index"
+print("Loading embeddings / vector store...")
+emb = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
+vs  = FAISS.load_local(INDEX_PATH, emb, allow_dangerous_deserialization=True)
+print("Ready.\n")
+
+scorer = rs.RougeScorer(["rougeL"], use_stemmer=True)
+
+def _cosine(a, b):
+    n = np.linalg.norm(a) * np.linalg.norm(b)
+    return float(np.dot(a, b) / (n + 1e-8))
+
+def compute_metrics(answer: str, question: str) -> dict:
+    if not answer or not answer.strip():
+        return {"accuracy": 0, "rouge_l": 0, "groundedness": 0,
+                "answer_relevance": 0, "faithfulness": 0}
+    try:
+        a_emb = np.array(vs.embeddings.embed_query(answer))
+        q_emb = np.array(vs.embeddings.embed_query(question))
+        answer_relevance = round(max(0.0, _cosine(a_emb, q_emb)), 3)
+
+        ref = ref_answers.get(question.strip().lower(), "")
+        accuracy, rouge_l = 0.0, 0.0
+        if ref:
+            rouge_l  = round(scorer.score(ref, answer)["rougeL"].fmeasure, 3)
+            r_emb    = np.array(vs.embeddings.embed_query(ref))
+            accuracy = round(max(0.0, _cosine(a_emb, r_emb)), 3)
+
+        # For fine-tuned: no retrieval context, so groundedness = accuracy, faithfulness = rouge_l
+        return {"accuracy": accuracy, "rouge_l": rouge_l,
+                "groundedness": accuracy, "answer_relevance": answer_relevance,
+                "faithfulness": rouge_l}
+    except Exception as e:
+        print(f"  [metric error] {e}")
+        return {"accuracy": 0, "rouge_l": 0, "groundedness": 0,
+                "answer_relevance": 0, "faithfulness": 0}
+
+def _cost(answer: str) -> float:
+    tokens = max(1, len(answer.split()))
+    return round(tokens * 0.0000015, 4)  # local model, near-zero cost
+
+# ── Load existing cache ───────────────────────────────────────────────────────
+OUT_PATH = "data/benchmark_cache.json"
+with open(OUT_PATH, encoding="utf-8") as f:
+    results = json.load(f)
+
+total = len(results)
+print(f"Filling r3 values for {total} cached questions...\n")
+print("(Loading fine-tuned model on first question — ~30s)\n")
+
+for i, row in enumerate(results):
+    q = row["question"]
+    print(f"[{i+1:02d}/{total}] {q[:65]}")
+
+    r3 = ask_finetuned(q)
+    answer = r3.get("answer", "")
+    elapsed = r3.get("response_time", 0)
+
+    m = compute_metrics(answer, q)
+
+    row["r3_time"]   = elapsed
+    row["r3_rouge"]  = m["rouge_l"]
+    row["r3_sim"]    = m["accuracy"]
+    row["r3_ground"] = m["groundedness"]
+    row["r3_relev"]  = m["answer_relevance"]
+    row["r3_faith"]  = m["faithfulness"]
+    row["r3_cost"]   = _cost(answer)
+
+    print(f"       r3_acc={m['accuracy']:.3f}  r3_rouge={m['rouge_l']:.3f}  time={elapsed}s")
+
+    # Save after every row so we can resume on crash
+    with open(OUT_PATH, "w", encoding="utf-8") as f:
+        json.dump(results, f, indent=2)
+
+# ── Summary ───────────────────────────────────────────────────────────────────
+n = len(results)
+r3_acc  = round(sum(r["r3_sim"]   for r in results) / n * 100, 1)
+r3_rl   = round(sum(r["r3_rouge"] for r in results) / n, 3)
+r3_t    = round(sum(r["r3_time"]  for r in results) / n, 2)
+print(f"\nDone! r3 values saved to {OUT_PATH}")
+print(f"  Fine-Tuned — accuracy {r3_acc}%  rouge_l {r3_rl}  avg_time {r3_t}s")

system3_inference.py

@@ -14,7 +14,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 
 BASE_MODEL = "Qwen/Qwen2.5-1.5B-Instruct"
-FINETUNED_PATH = "./finetuned_model"
+FINETUNED_PATH = "./checkpoint-25"
 
 _model = None
 _tokenizer = None