haiqian-mit commited on
Commit
c5cd341
路
verified 路
1 Parent(s): 8265281

Prospector v0.2.3 release (part 21)

Browse files
This view is limited to 50 files because it contains too many changes. 聽 See raw diff
Files changed (50) hide show
  1. data/runs/q2bpool32-s108--tsmc_core_agentic/completions.jsonl +0 -0
  2. data/runs/q2bpool32-s108--tsmc_core_agentic/results.jsonl +0 -0
  3. data/runs/q2bpool32-s108--tsmc_core_agentic/summary.json +192 -0
  4. data/runs/q2bpool32-s108--tsmc_core_agentic/trajectories.jsonl +0 -0
  5. data/runs/q2bpool32-s109--tsmc_core_agentic/completions.jsonl +0 -0
  6. data/runs/q2bpool32-s109--tsmc_core_agentic/results.jsonl +0 -0
  7. data/runs/q2bpool32-s109--tsmc_core_agentic/summary.json +192 -0
  8. data/runs/q2bpool32-s109--tsmc_core_agentic/trajectories.jsonl +0 -0
  9. data/runs/q2bpool32-s110--tsmc_core_agentic/completions.jsonl +0 -0
  10. data/runs/q2bpool32-s110--tsmc_core_agentic/results.jsonl +0 -0
  11. data/runs/q2bpool32-s110--tsmc_core_agentic/summary.json +192 -0
  12. data/runs/q2bpool32-s110--tsmc_core_agentic/trajectories.jsonl +0 -0
  13. data/runs/q2bpool32-s111--tsmc_core_agentic/completions.jsonl +0 -0
  14. data/runs/q2bpool32-s111--tsmc_core_agentic/results.jsonl +0 -0
  15. data/runs/q2bpool32-s111--tsmc_core_agentic/summary.json +192 -0
  16. data/runs/q2bpool32-s111--tsmc_core_agentic/trajectories.jsonl +0 -0
  17. data/runs/q2bpool32-s112--tsmc_core_agentic/completions.jsonl +0 -0
  18. data/runs/q2bpool32-s112--tsmc_core_agentic/results.jsonl +0 -0
  19. data/runs/q2bpool32-s112--tsmc_core_agentic/summary.json +192 -0
  20. data/runs/q2bpool32-s112--tsmc_core_agentic/trajectories.jsonl +0 -0
  21. data/runs/q2bpool32-s113--tsmc_core_agentic/completions.jsonl +0 -0
  22. data/runs/q2bpool32-s113--tsmc_core_agentic/results.jsonl +0 -0
  23. data/runs/q2bpool32-s113--tsmc_core_agentic/summary.json +192 -0
  24. data/runs/q2bpool32-s113--tsmc_core_agentic/trajectories.jsonl +0 -0
  25. data/runs/q2bpool32-s114--tsmc_core_agentic/completions.jsonl +0 -0
  26. data/runs/q2bpool32-s114--tsmc_core_agentic/results.jsonl +0 -0
  27. data/runs/q2bpool32-s114--tsmc_core_agentic/summary.json +192 -0
  28. data/runs/q2bpool32-s114--tsmc_core_agentic/trajectories.jsonl +0 -0
  29. data/runs/q2bpool32-s115--tsmc_core_agentic/completions.jsonl +0 -0
  30. data/runs/q2bpool32-s115--tsmc_core_agentic/results.jsonl +0 -0
  31. data/runs/q2bpool32-s115--tsmc_core_agentic/summary.json +192 -0
  32. data/runs/q2bpool32-s115--tsmc_core_agentic/trajectories.jsonl +0 -0
  33. data/runs/q2bpool32-s116--tsmc_core_agentic/completions.jsonl +0 -0
  34. data/runs/q2bpool32-s116--tsmc_core_agentic/results.jsonl +0 -0
  35. data/runs/q2bpool32-s116--tsmc_core_agentic/summary.json +192 -0
  36. data/runs/q2bpool32-s116--tsmc_core_agentic/trajectories.jsonl +0 -0
  37. data/runs/q2bpool32-s117--tsmc_core_agentic/completions.jsonl +0 -0
  38. data/runs/q2bpool32-s117--tsmc_core_agentic/results.jsonl +0 -0
  39. data/runs/q2bpool32-s117--tsmc_core_agentic/summary.json +192 -0
  40. data/runs/q2bpool32-s117--tsmc_core_agentic/trajectories.jsonl +0 -0
  41. data/runs/q2bpool32-s118--tsmc_core_agentic/completions.jsonl +0 -0
  42. data/runs/q2bpool32-s118--tsmc_core_agentic/results.jsonl +0 -0
  43. data/runs/q2bpool32-s118--tsmc_core_agentic/summary.json +192 -0
  44. data/runs/q2bpool32-s118--tsmc_core_agentic/trajectories.jsonl +0 -0
  45. data/runs/q2bpool32-s119--tsmc_core_agentic/completions.jsonl +0 -0
  46. data/runs/q2bpool32-s119--tsmc_core_agentic/results.jsonl +0 -0
  47. data/runs/q2bpool32-s119--tsmc_core_agentic/summary.json +192 -0
  48. data/runs/q2bpool32-s119--tsmc_core_agentic/trajectories.jsonl +0 -0
  49. data/runs/q2bpool32-s120--tsmc_core_agentic/completions.jsonl +0 -0
  50. data/runs/q2bpool32-s120--tsmc_core_agentic/results.jsonl +0 -0
data/runs/q2bpool32-s108--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s108--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s108--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 148340,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 108,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.011904761904761906,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.7928571428571428,
71
+ "path_precision": 0.6228571428571429,
72
+ "em_strict": 0.8285714285714286,
73
+ "em_claim": 0.8285714285714286,
74
+ "success": 0.8285714285714286,
75
+ "evidence_recall": 0.9071428571428571,
76
+ "reached_answer": 0.9142857142857143,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.15714285714285714,
79
+ "efficiency": 0.03883034285766838,
80
+ "gate_pass": 0.9714285714285714,
81
+ "gated_acc": 0.8285714285714286,
82
+ "gated_abstained": 0.02857142857142857,
83
+ "n_calls": 12.228571428571428,
84
+ "n_edges_seen": 107.25714285714285,
85
+ "n_repeats": 0.0,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.02857142857142857,
88
+ "median_n_edges_seen": 98.5,
89
+ "median_edges_to_answer": 21.5,
90
+ "median_n_calls": 13.0,
91
+ "reached_rate": 0.9142857142857143,
92
+ "censored": 0.0
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 0.9824561403508771,
97
+ "halluc_strict": 0.0071428571428571435,
98
+ "halluc_soft": 0.0,
99
+ "path_recall": 0.37719298245614036,
100
+ "path_precision": 0.37234977324263036,
101
+ "em_strict": 0.631578947368421,
102
+ "em_claim": 0.631578947368421,
103
+ "success": 0.6140350877192983,
104
+ "evidence_recall": 0.6842105263157895,
105
+ "reached_answer": 0.6666666666666666,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.14035087719298245,
108
+ "efficiency": 0.04676203562190346,
109
+ "gate_pass": 0.9649122807017544,
110
+ "gated_acc": 0.6140350877192983,
111
+ "gated_abstained": 0.03508771929824561,
112
+ "n_calls": 15.473684210526315,
113
+ "n_edges_seen": 117.78947368421052,
114
+ "n_repeats": 0.017543859649122806,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.017543859649122806,
117
+ "median_n_edges_seen": 112.0,
118
+ "median_edges_to_answer": 31.5,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.6666666666666666,
121
+ "censored": 0.17543859649122806
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.004201680672268907,
127
+ "halluc_soft": 0.0021008403361344546,
128
+ "path_recall": 0.6617647058823529,
129
+ "path_precision": 0.6017156862745098,
130
+ "em_strict": 0.8382352941176471,
131
+ "em_claim": 0.8382352941176471,
132
+ "success": 0.8235294117647058,
133
+ "evidence_recall": 0.8480392156862745,
134
+ "reached_answer": 0.9117647058823529,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.10294117647058823,
137
+ "efficiency": 0.041853396670698415,
138
+ "gate_pass": 0.9852941176470589,
139
+ "gated_acc": 0.8235294117647058,
140
+ "gated_abstained": 0.014705882352941176,
141
+ "n_calls": 13.397058823529411,
142
+ "n_edges_seen": 127.0,
143
+ "n_repeats": 0.0,
144
+ "n_malformed": 0.014705882352941176,
145
+ "n_tool_errors": 0.029411764705882353,
146
+ "median_n_edges_seen": 111.0,
147
+ "median_edges_to_answer": 21.0,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.9117647058823529,
150
+ "censored": 0.08823529411764706
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 0.9948717948717949,
155
+ "halluc_strict": 0.00783014236622484,
156
+ "halluc_soft": 0.000736377025036819,
157
+ "path_recall": 0.6256410256410256,
158
+ "path_precision": 0.5431353297332678,
159
+ "em_strict": 0.7743589743589744,
160
+ "em_claim": 0.7743589743589744,
161
+ "success": 0.764102564102564,
162
+ "evidence_recall": 0.8213675213675213,
163
+ "reached_answer": 0.841025641025641,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.13333333333333333,
166
+ "efficiency": 0.04220303079021937,
167
+ "gate_pass": 0.9743589743589743,
168
+ "gated_acc": 0.764102564102564,
169
+ "gated_abstained": 0.02564102564102564,
170
+ "n_calls": 13.584615384615384,
171
+ "n_edges_seen": 117.22051282051282,
172
+ "n_repeats": 0.005128205128205128,
173
+ "n_malformed": 0.005128205128205128,
174
+ "n_tool_errors": 0.02564102564102564,
175
+ "median_n_edges_seen": 104.0,
176
+ "median_edges_to_answer": 23.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.841025641025641,
179
+ "censored": 0.08205128205128205
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.5,
186
+ "p50_s": 4.1,
187
+ "p75_s": 4.9,
188
+ "questions_per_minute": 14.55,
189
+ "turns_per_episode": 15.0,
190
+ "s_per_turn": 0.27
191
+ }
192
+ }
data/runs/q2bpool32-s108--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s109--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s109--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s109--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 146119,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 109,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.004081632653061224,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.8,
71
+ "path_precision": 0.6188095238095238,
72
+ "em_strict": 0.9142857142857143,
73
+ "em_claim": 0.9142857142857143,
74
+ "success": 0.9,
75
+ "evidence_recall": 0.95,
76
+ "reached_answer": 0.9714285714285714,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.12857142857142856,
79
+ "efficiency": 0.03567801917123511,
80
+ "gate_pass": 0.9857142857142858,
81
+ "gated_acc": 0.9,
82
+ "gated_abstained": 0.014285714285714285,
83
+ "n_calls": 12.157142857142857,
84
+ "n_edges_seen": 104.68571428571428,
85
+ "n_repeats": 0.0,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.02857142857142857,
88
+ "median_n_edges_seen": 96.0,
89
+ "median_edges_to_answer": 28.5,
90
+ "median_n_calls": 13.5,
91
+ "reached_rate": 0.9714285714285714,
92
+ "censored": 0.014285714285714285
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.0043859649122807015,
98
+ "halluc_soft": 0.0,
99
+ "path_recall": 0.4517543859649123,
100
+ "path_precision": 0.4288011695906433,
101
+ "em_strict": 0.6666666666666666,
102
+ "em_claim": 0.6666666666666666,
103
+ "success": 0.6491228070175439,
104
+ "evidence_recall": 0.7149122807017544,
105
+ "reached_answer": 0.7368421052631579,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.10526315789473684,
108
+ "efficiency": 0.04681528260663048,
109
+ "gate_pass": 0.9824561403508771,
110
+ "gated_acc": 0.6491228070175439,
111
+ "gated_abstained": 0.017543859649122806,
112
+ "n_calls": 14.982456140350877,
113
+ "n_edges_seen": 115.84210526315789,
114
+ "n_repeats": 0.0,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.0,
117
+ "median_n_edges_seen": 115.0,
118
+ "median_edges_to_answer": 37.5,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.7368421052631579,
121
+ "censored": 0.10526315789473684
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.015686274509803918,
127
+ "halluc_soft": 0.0,
128
+ "path_recall": 0.6323529411764706,
129
+ "path_precision": 0.5623774509803922,
130
+ "em_strict": 0.7794117647058824,
131
+ "em_claim": 0.7794117647058824,
132
+ "success": 0.7352941176470589,
133
+ "evidence_recall": 0.8186274509803921,
134
+ "reached_answer": 0.8529411764705882,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.07352941176470588,
137
+ "efficiency": 0.03908246736282574,
138
+ "gate_pass": 0.9264705882352942,
139
+ "gated_acc": 0.7352941176470589,
140
+ "gated_abstained": 0.07352941176470588,
141
+ "n_calls": 13.676470588235293,
142
+ "n_edges_seen": 129.77941176470588,
143
+ "n_repeats": 0.0,
144
+ "n_malformed": 0.014705882352941176,
145
+ "n_tool_errors": 0.10294117647058823,
146
+ "median_n_edges_seen": 115.0,
147
+ "median_edges_to_answer": 25.0,
148
+ "median_n_calls": 15.0,
149
+ "reached_rate": 0.8529411764705882,
150
+ "censored": 0.10294117647058823
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.008217338217338216,
156
+ "halluc_soft": 0.0,
157
+ "path_recall": 0.6397435897435897,
158
+ "path_precision": 0.5435897435897435,
159
+ "em_strict": 0.7948717948717948,
160
+ "em_claim": 0.7948717948717948,
161
+ "success": 0.7692307692307693,
162
+ "evidence_recall": 0.8354700854700854,
163
+ "reached_answer": 0.8615384615384616,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.10256410256410256,
166
+ "efficiency": 0.04012071913454639,
167
+ "gate_pass": 0.9641025641025641,
168
+ "gated_acc": 0.7692307692307693,
169
+ "gated_abstained": 0.035897435897435895,
170
+ "n_calls": 13.512820512820513,
171
+ "n_edges_seen": 116.6974358974359,
172
+ "n_repeats": 0.0,
173
+ "n_malformed": 0.005128205128205128,
174
+ "n_tool_errors": 0.046153846153846156,
175
+ "median_n_edges_seen": 108.0,
176
+ "median_edges_to_answer": 32.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8615384615384616,
179
+ "censored": 0.07179487179487179
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 4.1,
186
+ "p50_s": 4.6,
187
+ "p75_s": 5.3,
188
+ "questions_per_minute": 12.94,
189
+ "turns_per_episode": 15.6,
190
+ "s_per_turn": 0.3
191
+ }
192
+ }
data/runs/q2bpool32-s109--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s110--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s110--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s110--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 145375,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 110,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.7714285714285715,
71
+ "path_precision": 0.6086394557823129,
72
+ "em_strict": 0.8571428571428571,
73
+ "em_claim": 0.8571428571428571,
74
+ "success": 0.8571428571428571,
75
+ "evidence_recall": 0.8785714285714286,
76
+ "reached_answer": 0.9,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.15714285714285714,
79
+ "efficiency": 0.030334458876691624,
80
+ "gate_pass": 0.9857142857142858,
81
+ "gated_acc": 0.8571428571428571,
82
+ "gated_abstained": 0.014285714285714285,
83
+ "n_calls": 12.314285714285715,
84
+ "n_edges_seen": 105.65714285714286,
85
+ "n_repeats": 0.0,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.02857142857142857,
88
+ "median_n_edges_seen": 101.5,
89
+ "median_edges_to_answer": 21.5,
90
+ "median_n_calls": 13.0,
91
+ "reached_rate": 0.9,
92
+ "censored": 0.02857142857142857
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.005847953216374268,
98
+ "halluc_soft": 0.002923976608187134,
99
+ "path_recall": 0.40789473684210525,
100
+ "path_precision": 0.40822890559732666,
101
+ "em_strict": 0.6491228070175439,
102
+ "em_claim": 0.6491228070175439,
103
+ "success": 0.631578947368421,
104
+ "evidence_recall": 0.7412280701754386,
105
+ "reached_answer": 0.7368421052631579,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.10526315789473684,
108
+ "efficiency": 0.045004974761995054,
109
+ "gate_pass": 0.9649122807017544,
110
+ "gated_acc": 0.631578947368421,
111
+ "gated_abstained": 0.03508771929824561,
112
+ "n_calls": 15.035087719298245,
113
+ "n_edges_seen": 120.84210526315789,
114
+ "n_repeats": 0.017543859649122806,
115
+ "n_malformed": 0.03508771929824561,
116
+ "n_tool_errors": 0.03508771929824561,
117
+ "median_n_edges_seen": 109.0,
118
+ "median_edges_to_answer": 37.5,
119
+ "median_n_calls": 15.0,
120
+ "reached_rate": 0.7368421052631579,
121
+ "censored": 0.15789473684210525
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.006127450980392156,
127
+ "halluc_soft": 0.0,
128
+ "path_recall": 0.6176470588235294,
129
+ "path_precision": 0.5884803921568628,
130
+ "em_strict": 0.75,
131
+ "em_claim": 0.75,
132
+ "success": 0.7205882352941176,
133
+ "evidence_recall": 0.8137254901960784,
134
+ "reached_answer": 0.8235294117647058,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.08823529411764706,
137
+ "efficiency": 0.04121165509171038,
138
+ "gate_pass": 0.9558823529411765,
139
+ "gated_acc": 0.7205882352941176,
140
+ "gated_abstained": 0.04411764705882353,
141
+ "n_calls": 13.058823529411764,
142
+ "n_edges_seen": 118.23529411764706,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.0,
145
+ "n_tool_errors": 0.014705882352941176,
146
+ "median_n_edges_seen": 109.5,
147
+ "median_edges_to_answer": 28.0,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.8235294117647058,
150
+ "censored": 0.08823529411764706
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.0038461538461538455,
156
+ "halluc_soft": 0.0008547008547008545,
157
+ "path_recall": 0.6115384615384616,
158
+ "path_precision": 0.543028083028083,
159
+ "em_strict": 0.7589743589743589,
160
+ "em_claim": 0.7589743589743589,
161
+ "success": 0.7435897435897436,
162
+ "evidence_recall": 0.8158119658119659,
163
+ "reached_answer": 0.8256410256410256,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.11794871794871795,
166
+ "efficiency": 0.03841583707199199,
167
+ "gate_pass": 0.9692307692307692,
168
+ "gated_acc": 0.7435897435897436,
169
+ "gated_abstained": 0.03076923076923077,
170
+ "n_calls": 13.36923076923077,
171
+ "n_edges_seen": 114.48205128205129,
172
+ "n_repeats": 0.010256410256410256,
173
+ "n_malformed": 0.010256410256410256,
174
+ "n_tool_errors": 0.02564102564102564,
175
+ "median_n_edges_seen": 105.0,
176
+ "median_edges_to_answer": 29.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8256410256410256,
179
+ "censored": 0.08717948717948718
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.6,
186
+ "p50_s": 4.8,
187
+ "p75_s": 5.5,
188
+ "questions_per_minute": 12.43,
189
+ "turns_per_episode": 15.6,
190
+ "s_per_turn": 0.28
191
+ }
192
+ }
data/runs/q2bpool32-s110--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s111--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s111--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s111--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 143260,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 111,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0038265306122448987,
69
+ "halluc_soft": 0.0017857142857142857,
70
+ "path_recall": 0.7428571428571429,
71
+ "path_precision": 0.5896428571428571,
72
+ "em_strict": 0.8285714285714286,
73
+ "em_claim": 0.8285714285714286,
74
+ "success": 0.8142857142857143,
75
+ "evidence_recall": 0.8571428571428571,
76
+ "reached_answer": 0.8857142857142857,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.14285714285714285,
79
+ "efficiency": 0.03646429100931058,
80
+ "gate_pass": 0.9714285714285714,
81
+ "gated_acc": 0.8142857142857143,
82
+ "gated_abstained": 0.02857142857142857,
83
+ "n_calls": 12.285714285714286,
84
+ "n_edges_seen": 104.71428571428571,
85
+ "n_repeats": 0.014285714285714285,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.014285714285714285,
88
+ "median_n_edges_seen": 97.5,
89
+ "median_edges_to_answer": 20.0,
90
+ "median_n_calls": 13.5,
91
+ "reached_rate": 0.8857142857142857,
92
+ "censored": 0.05714285714285714
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.0058479532163742695,
98
+ "halluc_soft": 0.0,
99
+ "path_recall": 0.4298245614035088,
100
+ "path_precision": 0.4695071010860485,
101
+ "em_strict": 0.6491228070175439,
102
+ "em_claim": 0.6491228070175439,
103
+ "success": 0.631578947368421,
104
+ "evidence_recall": 0.7587719298245614,
105
+ "reached_answer": 0.7719298245614035,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.12280701754385964,
108
+ "efficiency": 0.04408509872960013,
109
+ "gate_pass": 0.9824561403508771,
110
+ "gated_acc": 0.631578947368421,
111
+ "gated_abstained": 0.017543859649122806,
112
+ "n_calls": 14.631578947368421,
113
+ "n_edges_seen": 121.87719298245614,
114
+ "n_repeats": 0.0,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.03508771929824561,
117
+ "median_n_edges_seen": 120.0,
118
+ "median_edges_to_answer": 40.0,
119
+ "median_n_calls": 15.0,
120
+ "reached_rate": 0.7719298245614035,
121
+ "censored": 0.05263157894736842
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.015196078431372547,
127
+ "halluc_soft": 0.004901960784313726,
128
+ "path_recall": 0.6225490196078431,
129
+ "path_precision": 0.5625,
130
+ "em_strict": 0.7352941176470589,
131
+ "em_claim": 0.7352941176470589,
132
+ "success": 0.7058823529411765,
133
+ "evidence_recall": 0.7941176470588235,
134
+ "reached_answer": 0.8235294117647058,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.08823529411764706,
137
+ "efficiency": 0.04428700255278451,
138
+ "gate_pass": 0.9411764705882353,
139
+ "gated_acc": 0.7058823529411765,
140
+ "gated_abstained": 0.058823529411764705,
141
+ "n_calls": 13.544117647058824,
142
+ "n_edges_seen": 124.05882352941177,
143
+ "n_repeats": 0.029411764705882353,
144
+ "n_malformed": 0.0,
145
+ "n_tool_errors": 0.029411764705882353,
146
+ "median_n_edges_seen": 114.5,
147
+ "median_edges_to_answer": 28.0,
148
+ "median_n_calls": 14.5,
149
+ "reached_rate": 0.8235294117647058,
150
+ "censored": 0.10294117647058823
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.008382173382173381,
156
+ "halluc_soft": 0.0023504273504273508,
157
+ "path_recall": 0.6094017094017093,
158
+ "path_precision": 0.5450610500610501,
159
+ "em_strict": 0.7435897435897436,
160
+ "em_claim": 0.7435897435897436,
161
+ "success": 0.7230769230769231,
162
+ "evidence_recall": 0.8064102564102564,
163
+ "reached_answer": 0.8307692307692308,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.11794871794871795,
166
+ "efficiency": 0.041419831650401515,
167
+ "gate_pass": 0.9641025641025641,
168
+ "gated_acc": 0.7230769230769231,
169
+ "gated_abstained": 0.035897435897435895,
170
+ "n_calls": 13.41025641025641,
171
+ "n_edges_seen": 116.47692307692307,
172
+ "n_repeats": 0.015384615384615385,
173
+ "n_malformed": 0.0,
174
+ "n_tool_errors": 0.02564102564102564,
175
+ "median_n_edges_seen": 112.0,
176
+ "median_edges_to_answer": 29.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8307692307692308,
179
+ "censored": 0.07179487179487179
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.4,
186
+ "p50_s": 4.3,
187
+ "p75_s": 5.0,
188
+ "questions_per_minute": 14.1,
189
+ "turns_per_episode": 15.1,
190
+ "s_per_turn": 0.29
191
+ }
192
+ }
data/runs/q2bpool32-s111--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s112--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s112--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s112--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 149817,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 112,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.01,
69
+ "halluc_soft": 0.007142857142857143,
70
+ "path_recall": 0.8214285714285714,
71
+ "path_precision": 0.6192857142857143,
72
+ "em_strict": 0.8285714285714286,
73
+ "em_claim": 0.8285714285714286,
74
+ "success": 0.8142857142857143,
75
+ "evidence_recall": 0.9071428571428571,
76
+ "reached_answer": 0.9285714285714286,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.11428571428571428,
79
+ "efficiency": 0.037559916026703985,
80
+ "gate_pass": 0.9571428571428572,
81
+ "gated_acc": 0.8142857142857143,
82
+ "gated_abstained": 0.04285714285714286,
83
+ "n_calls": 12.457142857142857,
84
+ "n_edges_seen": 107.3,
85
+ "n_repeats": 0.014285714285714285,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.0,
88
+ "median_n_edges_seen": 113.0,
89
+ "median_edges_to_answer": 23.0,
90
+ "median_n_calls": 13.0,
91
+ "reached_rate": 0.9285714285714286,
92
+ "censored": 0.07142857142857142
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.012740183792815373,
98
+ "halluc_soft": 0.0,
99
+ "path_recall": 0.5175438596491229,
100
+ "path_precision": 0.49490392648287385,
101
+ "em_strict": 0.7192982456140351,
102
+ "em_claim": 0.7192982456140351,
103
+ "success": 0.6666666666666666,
104
+ "evidence_recall": 0.7631578947368421,
105
+ "reached_answer": 0.8070175438596491,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.05263157894736842,
108
+ "efficiency": 0.039974207485221,
109
+ "gate_pass": 0.9473684210526315,
110
+ "gated_acc": 0.6666666666666666,
111
+ "gated_abstained": 0.05263157894736842,
112
+ "n_calls": 14.964912280701755,
113
+ "n_edges_seen": 134.7719298245614,
114
+ "n_repeats": 0.03508771929824561,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.017543859649122806,
117
+ "median_n_edges_seen": 135.0,
118
+ "median_edges_to_answer": 38.0,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.8070175438596491,
121
+ "censored": 0.17543859649122806
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.009558823529411765,
127
+ "halluc_soft": 0.003676470588235294,
128
+ "path_recall": 0.642156862745098,
129
+ "path_precision": 0.5727240896358543,
130
+ "em_strict": 0.7205882352941176,
131
+ "em_claim": 0.7205882352941176,
132
+ "success": 0.6911764705882353,
133
+ "evidence_recall": 0.8039215686274509,
134
+ "reached_answer": 0.8529411764705882,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.10294117647058823,
137
+ "efficiency": 0.03941118897497814,
138
+ "gate_pass": 0.9705882352941176,
139
+ "gated_acc": 0.6911764705882353,
140
+ "gated_abstained": 0.029411764705882353,
141
+ "n_calls": 13.470588235294118,
142
+ "n_edges_seen": 122.0,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.014705882352941176,
145
+ "n_tool_errors": 0.0,
146
+ "median_n_edges_seen": 116.0,
147
+ "median_edges_to_answer": 29.5,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.8529411764705882,
150
+ "censored": 0.04411764705882353
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.010647130647130649,
156
+ "halluc_soft": 0.0038461538461538464,
157
+ "path_recall": 0.67008547008547,
158
+ "path_precision": 0.5666910866910867,
159
+ "em_strict": 0.7589743589743589,
160
+ "em_claim": 0.7589743589743589,
161
+ "success": 0.7282051282051282,
162
+ "evidence_recall": 0.829059829059829,
163
+ "reached_answer": 0.8666666666666667,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.09230769230769231,
166
+ "efficiency": 0.03891120409654046,
167
+ "gate_pass": 0.958974358974359,
168
+ "gated_acc": 0.7282051282051282,
169
+ "gated_abstained": 0.041025641025641026,
170
+ "n_calls": 13.543589743589743,
171
+ "n_edges_seen": 120.45641025641025,
172
+ "n_repeats": 0.020512820512820513,
173
+ "n_malformed": 0.005128205128205128,
174
+ "n_tool_errors": 0.005128205128205128,
175
+ "median_n_edges_seen": 116.0,
176
+ "median_edges_to_answer": 32.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8666666666666667,
179
+ "censored": 0.09230769230769231
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.7,
186
+ "p50_s": 4.5,
187
+ "p75_s": 5.3,
188
+ "questions_per_minute": 13.2,
189
+ "turns_per_episode": 15.3,
190
+ "s_per_turn": 0.3
191
+ }
192
+ }
data/runs/q2bpool32-s112--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s113--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s113--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s113--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 146100,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 113,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.7928571428571428,
71
+ "path_precision": 0.6198809523809524,
72
+ "em_strict": 0.8571428571428571,
73
+ "em_claim": 0.8571428571428571,
74
+ "success": 0.8571428571428571,
75
+ "evidence_recall": 0.8928571428571429,
76
+ "reached_answer": 0.9571428571428572,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.15714285714285714,
79
+ "efficiency": 0.032269701594644234,
80
+ "gate_pass": 1.0,
81
+ "gated_acc": 0.8571428571428571,
82
+ "gated_abstained": 0.0,
83
+ "n_calls": 12.32857142857143,
84
+ "n_edges_seen": 112.14285714285714,
85
+ "n_repeats": 0.02857142857142857,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.014285714285714285,
88
+ "median_n_edges_seen": 107.0,
89
+ "median_edges_to_answer": 25.0,
90
+ "median_n_calls": 13.0,
91
+ "reached_rate": 0.9571428571428572,
92
+ "censored": 0.014285714285714285
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.0064327485380116945,
98
+ "halluc_soft": 0.0064327485380116945,
99
+ "path_recall": 0.4166666666666667,
100
+ "path_precision": 0.42857142857142855,
101
+ "em_strict": 0.6140350877192983,
102
+ "em_claim": 0.6140350877192983,
103
+ "success": 0.5964912280701754,
104
+ "evidence_recall": 0.7105263157894737,
105
+ "reached_answer": 0.7368421052631579,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.10526315789473684,
108
+ "efficiency": 0.04624964714708834,
109
+ "gate_pass": 0.9649122807017544,
110
+ "gated_acc": 0.5964912280701754,
111
+ "gated_abstained": 0.03508771929824561,
112
+ "n_calls": 15.333333333333334,
113
+ "n_edges_seen": 120.14035087719299,
114
+ "n_repeats": 0.017543859649122806,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.08771929824561403,
117
+ "median_n_edges_seen": 117.0,
118
+ "median_edges_to_answer": 28.0,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.7368421052631579,
121
+ "censored": 0.12280701754385964
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.012867647058823529,
127
+ "halluc_soft": 0.0,
128
+ "path_recall": 0.6372549019607844,
129
+ "path_precision": 0.6051470588235294,
130
+ "em_strict": 0.7941176470588235,
131
+ "em_claim": 0.7941176470588235,
132
+ "success": 0.75,
133
+ "evidence_recall": 0.8480392156862745,
134
+ "reached_answer": 0.8676470588235294,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.07352941176470588,
137
+ "efficiency": 0.047614599319693925,
138
+ "gate_pass": 0.9411764705882353,
139
+ "gated_acc": 0.75,
140
+ "gated_abstained": 0.058823529411764705,
141
+ "n_calls": 13.455882352941176,
142
+ "n_edges_seen": 121.20588235294117,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.014705882352941176,
145
+ "n_tool_errors": 0.014705882352941176,
146
+ "median_n_edges_seen": 111.0,
147
+ "median_edges_to_answer": 25.0,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.8676470588235294,
150
+ "censored": 0.07352941176470588
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.006367521367521368,
156
+ "halluc_soft": 0.00188034188034188,
157
+ "path_recall": 0.6286324786324786,
158
+ "path_precision": 0.5588217338217338,
159
+ "em_strict": 0.764102564102564,
160
+ "em_claim": 0.764102564102564,
161
+ "success": 0.7435897435897436,
162
+ "evidence_recall": 0.8239316239316239,
163
+ "reached_answer": 0.8615384615384616,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.11282051282051282,
166
+ "efficiency": 0.04170718847563241,
167
+ "gate_pass": 0.9692307692307692,
168
+ "gated_acc": 0.7435897435897436,
169
+ "gated_abstained": 0.03076923076923077,
170
+ "n_calls": 13.6,
171
+ "n_edges_seen": 117.64102564102564,
172
+ "n_repeats": 0.020512820512820513,
173
+ "n_malformed": 0.005128205128205128,
174
+ "n_tool_errors": 0.035897435897435895,
175
+ "median_n_edges_seen": 112.0,
176
+ "median_edges_to_answer": 27.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8615384615384616,
179
+ "censored": 0.06666666666666667
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.5,
186
+ "p50_s": 3.9,
187
+ "p75_s": 5.1,
188
+ "questions_per_minute": 15.41,
189
+ "turns_per_episode": 14.8,
190
+ "s_per_turn": 0.28
191
+ }
192
+ }
data/runs/q2bpool32-s113--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s114--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s114--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s114--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 146284,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 114,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.8,
71
+ "path_precision": 0.6223809523809524,
72
+ "em_strict": 0.8714285714285714,
73
+ "em_claim": 0.8714285714285714,
74
+ "success": 0.8714285714285714,
75
+ "evidence_recall": 0.8928571428571429,
76
+ "reached_answer": 0.9142857142857143,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.1,
79
+ "efficiency": 0.03134302176995173,
80
+ "gate_pass": 0.9857142857142858,
81
+ "gated_acc": 0.8714285714285714,
82
+ "gated_abstained": 0.014285714285714285,
83
+ "n_calls": 12.757142857142858,
84
+ "n_edges_seen": 111.25714285714285,
85
+ "n_repeats": 0.02857142857142857,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.04285714285714286,
88
+ "median_n_edges_seen": 112.0,
89
+ "median_edges_to_answer": 21.5,
90
+ "median_n_calls": 14.0,
91
+ "reached_rate": 0.9142857142857143,
92
+ "censored": 0.04285714285714286
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.013157894736842105,
98
+ "halluc_soft": 0.008771929824561403,
99
+ "path_recall": 0.41228070175438597,
100
+ "path_precision": 0.42534113060428846,
101
+ "em_strict": 0.6140350877192983,
102
+ "em_claim": 0.6140350877192983,
103
+ "success": 0.5964912280701754,
104
+ "evidence_recall": 0.6973684210526315,
105
+ "reached_answer": 0.7017543859649122,
106
+ "abstained": 0.017543859649122806,
107
+ "answer_named_before_seen": 0.10526315789473684,
108
+ "efficiency": 0.044965610801762,
109
+ "gate_pass": 0.9473684210526315,
110
+ "gated_acc": 0.5964912280701754,
111
+ "gated_abstained": 0.05263157894736842,
112
+ "n_calls": 15.368421052631579,
113
+ "n_edges_seen": 118.21052631578948,
114
+ "n_repeats": 0.017543859649122806,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.05263157894736842,
117
+ "median_n_edges_seen": 118.0,
118
+ "median_edges_to_answer": 30.0,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.7017543859649122,
121
+ "censored": 0.15789473684210525
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.0024509803921568623,
127
+ "halluc_soft": 0.0,
128
+ "path_recall": 0.6715686274509803,
129
+ "path_precision": 0.6508578431372549,
130
+ "em_strict": 0.7941176470588235,
131
+ "em_claim": 0.7941176470588235,
132
+ "success": 0.7941176470588235,
133
+ "evidence_recall": 0.8333333333333333,
134
+ "reached_answer": 0.8823529411764706,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.07352941176470588,
137
+ "efficiency": 0.042866355812142064,
138
+ "gate_pass": 0.9852941176470589,
139
+ "gated_acc": 0.7941176470588235,
140
+ "gated_abstained": 0.014705882352941176,
141
+ "n_calls": 12.897058823529411,
142
+ "n_edges_seen": 120.88235294117646,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.029411764705882353,
145
+ "n_tool_errors": 0.0,
146
+ "median_n_edges_seen": 125.5,
147
+ "median_edges_to_answer": 32.5,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.8823529411764706,
150
+ "censored": 0.058823529411764705
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.004700854700854701,
156
+ "halluc_soft": 0.002564102564102564,
157
+ "path_recall": 0.6418803418803419,
158
+ "path_precision": 0.5747150997150997,
159
+ "em_strict": 0.7692307692307693,
160
+ "em_claim": 0.7692307692307693,
161
+ "success": 0.764102564102564,
162
+ "evidence_recall": 0.814957264957265,
163
+ "reached_answer": 0.841025641025641,
164
+ "abstained": 0.005128205128205128,
165
+ "answer_named_before_seen": 0.09230769230769231,
166
+ "efficiency": 0.03934340274268059,
167
+ "gate_pass": 0.9743589743589743,
168
+ "gated_acc": 0.764102564102564,
169
+ "gated_abstained": 0.02564102564102564,
170
+ "n_calls": 13.569230769230769,
171
+ "n_edges_seen": 116.64615384615385,
172
+ "n_repeats": 0.020512820512820513,
173
+ "n_malformed": 0.010256410256410256,
174
+ "n_tool_errors": 0.03076923076923077,
175
+ "median_n_edges_seen": 116.0,
176
+ "median_edges_to_answer": 24.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.841025641025641,
179
+ "censored": 0.08205128205128205
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.7,
186
+ "p50_s": 4.8,
187
+ "p75_s": 5.4,
188
+ "questions_per_minute": 12.61,
189
+ "turns_per_episode": 15.6,
190
+ "s_per_turn": 0.27
191
+ }
192
+ }
data/runs/q2bpool32-s114--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s115--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s115--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s115--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 149482,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 115,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.8214285714285714,
71
+ "path_precision": 0.6457086167800453,
72
+ "em_strict": 0.8714285714285714,
73
+ "em_claim": 0.8714285714285714,
74
+ "success": 0.8714285714285714,
75
+ "evidence_recall": 0.9071428571428571,
76
+ "reached_answer": 0.9142857142857143,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.11428571428571428,
79
+ "efficiency": 0.03556590366628406,
80
+ "gate_pass": 0.9857142857142858,
81
+ "gated_acc": 0.8714285714285714,
82
+ "gated_abstained": 0.014285714285714285,
83
+ "n_calls": 12.014285714285714,
84
+ "n_edges_seen": 106.27142857142857,
85
+ "n_repeats": 0.0,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.0,
88
+ "median_n_edges_seen": 98.5,
89
+ "median_edges_to_answer": 33.0,
90
+ "median_n_calls": 13.0,
91
+ "reached_rate": 0.9142857142857143,
92
+ "censored": 0.04285714285714286
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.012280701754385965,
98
+ "halluc_soft": 0.0035087719298245606,
99
+ "path_recall": 0.40789473684210525,
100
+ "path_precision": 0.4203425229741019,
101
+ "em_strict": 0.631578947368421,
102
+ "em_claim": 0.631578947368421,
103
+ "success": 0.5964912280701754,
104
+ "evidence_recall": 0.7105263157894737,
105
+ "reached_answer": 0.7894736842105263,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.10526315789473684,
108
+ "efficiency": 0.044506451230286445,
109
+ "gate_pass": 0.9473684210526315,
110
+ "gated_acc": 0.5964912280701754,
111
+ "gated_abstained": 0.05263157894736842,
112
+ "n_calls": 15.385964912280702,
113
+ "n_edges_seen": 118.54385964912281,
114
+ "n_repeats": 0.017543859649122806,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.0,
117
+ "median_n_edges_seen": 105.0,
118
+ "median_edges_to_answer": 31.0,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.7894736842105263,
121
+ "censored": 0.14035087719298245
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.020016339869281047,
127
+ "halluc_soft": 0.01388888888888889,
128
+ "path_recall": 0.6225490196078431,
129
+ "path_precision": 0.5445261437908497,
130
+ "em_strict": 0.7352941176470589,
131
+ "em_claim": 0.7352941176470589,
132
+ "success": 0.6911764705882353,
133
+ "evidence_recall": 0.8137254901960784,
134
+ "reached_answer": 0.8382352941176471,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.07352941176470588,
137
+ "efficiency": 0.047119302254997435,
138
+ "gate_pass": 0.9117647058823529,
139
+ "gated_acc": 0.6911764705882353,
140
+ "gated_abstained": 0.08823529411764706,
141
+ "n_calls": 13.544117647058824,
142
+ "n_edges_seen": 117.29411764705883,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.029411764705882353,
145
+ "n_tool_errors": 0.029411764705882353,
146
+ "median_n_edges_seen": 117.0,
147
+ "median_edges_to_answer": 31.0,
148
+ "median_n_calls": 15.0,
149
+ "reached_rate": 0.8382352941176471,
150
+ "censored": 0.14705882352941177
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.010569800569800569,
156
+ "halluc_soft": 0.005868945868945869,
157
+ "path_recall": 0.6311965811965812,
158
+ "path_precision": 0.5445482295482296,
159
+ "em_strict": 0.7538461538461538,
160
+ "em_claim": 0.7538461538461538,
161
+ "success": 0.7282051282051282,
162
+ "evidence_recall": 0.8170940170940172,
163
+ "reached_answer": 0.8512820512820513,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.09743589743589744,
166
+ "efficiency": 0.04220817194926173,
167
+ "gate_pass": 0.9487179487179487,
168
+ "gated_acc": 0.7282051282051282,
169
+ "gated_abstained": 0.05128205128205128,
170
+ "n_calls": 13.533333333333333,
171
+ "n_edges_seen": 113.7025641025641,
172
+ "n_repeats": 0.010256410256410256,
173
+ "n_malformed": 0.010256410256410256,
174
+ "n_tool_errors": 0.010256410256410256,
175
+ "median_n_edges_seen": 108.0,
176
+ "median_edges_to_answer": 31.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8512820512820513,
179
+ "censored": 0.1076923076923077
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.3,
186
+ "p50_s": 4.5,
187
+ "p75_s": 5.2,
188
+ "questions_per_minute": 13.46,
189
+ "turns_per_episode": 15.4,
190
+ "s_per_turn": 0.29
191
+ }
192
+ }
data/runs/q2bpool32-s115--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s116--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s116--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s116--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 143934,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 116,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0023809523809523803,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.7785714285714286,
71
+ "path_precision": 0.6300566893424037,
72
+ "em_strict": 0.8571428571428571,
73
+ "em_claim": 0.8571428571428571,
74
+ "success": 0.8428571428571429,
75
+ "evidence_recall": 0.9071428571428571,
76
+ "reached_answer": 0.9142857142857143,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.14285714285714285,
79
+ "efficiency": 0.033819132916973325,
80
+ "gate_pass": 0.9714285714285714,
81
+ "gated_acc": 0.8428571428571429,
82
+ "gated_abstained": 0.02857142857142857,
83
+ "n_calls": 12.17142857142857,
84
+ "n_edges_seen": 112.58571428571429,
85
+ "n_repeats": 0.02857142857142857,
86
+ "n_malformed": 0.014285714285714285,
87
+ "n_tool_errors": 0.0,
88
+ "median_n_edges_seen": 100.0,
89
+ "median_edges_to_answer": 22.5,
90
+ "median_n_calls": 13.5,
91
+ "reached_rate": 0.9142857142857143,
92
+ "censored": 0.02857142857142857
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.007017543859649121,
98
+ "halluc_soft": 0.0035087719298245606,
99
+ "path_recall": 0.44298245614035087,
100
+ "path_precision": 0.44314954051796157,
101
+ "em_strict": 0.7543859649122807,
102
+ "em_claim": 0.7543859649122807,
103
+ "success": 0.7192982456140351,
104
+ "evidence_recall": 0.75,
105
+ "reached_answer": 0.7719298245614035,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.07017543859649122,
108
+ "efficiency": 0.04512994781274882,
109
+ "gate_pass": 0.9649122807017544,
110
+ "gated_acc": 0.7192982456140351,
111
+ "gated_abstained": 0.03508771929824561,
112
+ "n_calls": 15.105263157894736,
113
+ "n_edges_seen": 124.01754385964912,
114
+ "n_repeats": 0.0,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.0,
117
+ "median_n_edges_seen": 119.0,
118
+ "median_edges_to_answer": 37.0,
119
+ "median_n_calls": 15.0,
120
+ "reached_rate": 0.7719298245614035,
121
+ "censored": 0.12280701754385964
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.007352941176470588,
127
+ "halluc_soft": 0.007352941176470588,
128
+ "path_recall": 0.6568627450980392,
129
+ "path_precision": 0.5675420168067227,
130
+ "em_strict": 0.7794117647058824,
131
+ "em_claim": 0.7794117647058824,
132
+ "success": 0.75,
133
+ "evidence_recall": 0.8284313725490197,
134
+ "reached_answer": 0.8529411764705882,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.07352941176470588,
137
+ "efficiency": 0.04651722218574099,
138
+ "gate_pass": 0.9705882352941176,
139
+ "gated_acc": 0.75,
140
+ "gated_abstained": 0.029411764705882353,
141
+ "n_calls": 13.191176470588236,
142
+ "n_edges_seen": 114.82352941176471,
143
+ "n_repeats": 0.04411764705882353,
144
+ "n_malformed": 0.0,
145
+ "n_tool_errors": 0.014705882352941176,
146
+ "median_n_edges_seen": 111.5,
147
+ "median_edges_to_answer": 23.5,
148
+ "median_n_calls": 14.5,
149
+ "reached_rate": 0.8529411764705882,
150
+ "censored": 0.058823529411764705
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.005470085470085469,
156
+ "halluc_soft": 0.0035897435897435893,
157
+ "path_recall": 0.6380341880341881,
158
+ "path_precision": 0.5536223036223036,
159
+ "em_strict": 0.8,
160
+ "em_claim": 0.8,
161
+ "success": 0.7743589743589744,
162
+ "evidence_recall": 0.8337606837606838,
163
+ "reached_answer": 0.8512820512820513,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.09743589743589744,
166
+ "efficiency": 0.04155342275971898,
167
+ "gate_pass": 0.9692307692307692,
168
+ "gated_acc": 0.7743589743589744,
169
+ "gated_abstained": 0.03076923076923077,
170
+ "n_calls": 13.384615384615385,
171
+ "n_edges_seen": 116.70769230769231,
172
+ "n_repeats": 0.02564102564102564,
173
+ "n_malformed": 0.005128205128205128,
174
+ "n_tool_errors": 0.005128205128205128,
175
+ "median_n_edges_seen": 113.0,
176
+ "median_edges_to_answer": 28.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8512820512820513,
179
+ "censored": 0.06666666666666667
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 4.0,
186
+ "p50_s": 4.3,
187
+ "p75_s": 5.0,
188
+ "questions_per_minute": 13.84,
189
+ "turns_per_episode": 14.7,
190
+ "s_per_turn": 0.3
191
+ }
192
+ }
data/runs/q2bpool32-s116--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s117--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s117--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s117--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 145789,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 117,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.7714285714285715,
71
+ "path_precision": 0.5951360544217688,
72
+ "em_strict": 0.8428571428571429,
73
+ "em_claim": 0.8428571428571429,
74
+ "success": 0.8428571428571429,
75
+ "evidence_recall": 0.8571428571428571,
76
+ "reached_answer": 0.8857142857142857,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.15714285714285714,
79
+ "efficiency": 0.03302699817579546,
80
+ "gate_pass": 1.0,
81
+ "gated_acc": 0.8428571428571429,
82
+ "gated_abstained": 0.0,
83
+ "n_calls": 12.442857142857143,
84
+ "n_edges_seen": 102.2,
85
+ "n_repeats": 0.0,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.0,
88
+ "median_n_edges_seen": 89.0,
89
+ "median_edges_to_answer": 23.0,
90
+ "median_n_calls": 14.0,
91
+ "reached_rate": 0.8857142857142857,
92
+ "censored": 0.04285714285714286
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.0,
98
+ "halluc_soft": 0.0,
99
+ "path_recall": 0.4298245614035088,
100
+ "path_precision": 0.4236006683375104,
101
+ "em_strict": 0.5964912280701754,
102
+ "em_claim": 0.5964912280701754,
103
+ "success": 0.5964912280701754,
104
+ "evidence_recall": 0.7192982456140351,
105
+ "reached_answer": 0.7017543859649122,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.07017543859649122,
108
+ "efficiency": 0.04568131378370273,
109
+ "gate_pass": 1.0,
110
+ "gated_acc": 0.5964912280701754,
111
+ "gated_abstained": 0.0,
112
+ "n_calls": 14.508771929824562,
113
+ "n_edges_seen": 126.2280701754386,
114
+ "n_repeats": 0.017543859649122806,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.0,
117
+ "median_n_edges_seen": 113.0,
118
+ "median_edges_to_answer": 34.0,
119
+ "median_n_calls": 15.0,
120
+ "reached_rate": 0.7017543859649122,
121
+ "censored": 0.07017543859649122
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.011029411764705883,
127
+ "halluc_soft": 0.003676470588235294,
128
+ "path_recall": 0.5882352941176471,
129
+ "path_precision": 0.5484593837535015,
130
+ "em_strict": 0.6911764705882353,
131
+ "em_claim": 0.6911764705882353,
132
+ "success": 0.6764705882352942,
133
+ "evidence_recall": 0.8137254901960784,
134
+ "reached_answer": 0.7941176470588235,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.07352941176470588,
137
+ "efficiency": 0.044371951454872015,
138
+ "gate_pass": 0.9558823529411765,
139
+ "gated_acc": 0.6764705882352942,
140
+ "gated_abstained": 0.04411764705882353,
141
+ "n_calls": 14.014705882352942,
142
+ "n_edges_seen": 120.41176470588235,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.0,
145
+ "n_tool_errors": 0.029411764705882353,
146
+ "median_n_edges_seen": 109.0,
147
+ "median_edges_to_answer": 28.0,
148
+ "median_n_calls": 15.0,
149
+ "reached_rate": 0.7941176470588235,
150
+ "censored": 0.08823529411764706
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.0038461538461538464,
156
+ "halluc_soft": 0.001282051282051282,
157
+ "path_recall": 0.6076923076923076,
158
+ "path_precision": 0.5287179487179486,
159
+ "em_strict": 0.717948717948718,
160
+ "em_claim": 0.717948717948718,
161
+ "success": 0.7128205128205128,
162
+ "evidence_recall": 0.8017094017094017,
163
+ "reached_answer": 0.8,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.10256410256410256,
166
+ "efficiency": 0.04068214080465659,
167
+ "gate_pass": 0.9846153846153847,
168
+ "gated_acc": 0.7128205128205128,
169
+ "gated_abstained": 0.015384615384615385,
170
+ "n_calls": 13.594871794871795,
171
+ "n_edges_seen": 115.57435897435897,
172
+ "n_repeats": 0.010256410256410256,
173
+ "n_malformed": 0.0,
174
+ "n_tool_errors": 0.010256410256410256,
175
+ "median_n_edges_seen": 101.0,
176
+ "median_edges_to_answer": 28.0,
177
+ "median_n_calls": 15.0,
178
+ "reached_rate": 0.8,
179
+ "censored": 0.06666666666666667
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.4,
186
+ "p50_s": 4.7,
187
+ "p75_s": 5.9,
188
+ "questions_per_minute": 12.9,
189
+ "turns_per_episode": 15.4,
190
+ "s_per_turn": 0.3
191
+ }
192
+ }
data/runs/q2bpool32-s117--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s118--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s118--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s118--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 144631,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 118,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0,
69
+ "halluc_soft": 0.0,
70
+ "path_recall": 0.8071428571428572,
71
+ "path_precision": 0.5976077097505669,
72
+ "em_strict": 0.8428571428571429,
73
+ "em_claim": 0.8428571428571429,
74
+ "success": 0.8428571428571429,
75
+ "evidence_recall": 0.9,
76
+ "reached_answer": 0.9285714285714286,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.12857142857142856,
79
+ "efficiency": 0.03636460900561034,
80
+ "gate_pass": 0.9857142857142858,
81
+ "gated_acc": 0.8428571428571429,
82
+ "gated_abstained": 0.014285714285714285,
83
+ "n_calls": 12.542857142857143,
84
+ "n_edges_seen": 109.98571428571428,
85
+ "n_repeats": 0.014285714285714285,
86
+ "n_malformed": 0.0,
87
+ "n_tool_errors": 0.014285714285714285,
88
+ "median_n_edges_seen": 107.0,
89
+ "median_edges_to_answer": 21.0,
90
+ "median_n_calls": 13.0,
91
+ "reached_rate": 0.9285714285714286,
92
+ "censored": 0.1
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.008771929824561403,
98
+ "halluc_soft": 0.002923976608187134,
99
+ "path_recall": 0.3684210526315789,
100
+ "path_precision": 0.38350041771094406,
101
+ "em_strict": 0.6140350877192983,
102
+ "em_claim": 0.6140350877192983,
103
+ "success": 0.5789473684210527,
104
+ "evidence_recall": 0.6929824561403509,
105
+ "reached_answer": 0.7368421052631579,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.07017543859649122,
108
+ "efficiency": 0.04614022534679111,
109
+ "gate_pass": 0.9649122807017544,
110
+ "gated_acc": 0.5789473684210527,
111
+ "gated_abstained": 0.03508771929824561,
112
+ "n_calls": 15.192982456140351,
113
+ "n_edges_seen": 125.47368421052632,
114
+ "n_repeats": 0.0,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.03508771929824561,
117
+ "median_n_edges_seen": 119.0,
118
+ "median_edges_to_answer": 27.5,
119
+ "median_n_calls": 16.0,
120
+ "reached_rate": 0.7368421052631579,
121
+ "censored": 0.17543859649122806
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.011642156862745098,
127
+ "halluc_soft": 0.001838235294117647,
128
+ "path_recall": 0.6911764705882353,
129
+ "path_precision": 0.6207107843137255,
130
+ "em_strict": 0.7794117647058824,
131
+ "em_claim": 0.7794117647058824,
132
+ "success": 0.7352941176470589,
133
+ "evidence_recall": 0.8725490196078431,
134
+ "reached_answer": 0.8676470588235294,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.08823529411764706,
137
+ "efficiency": 0.04051561181252608,
138
+ "gate_pass": 0.9558823529411765,
139
+ "gated_acc": 0.7352941176470589,
140
+ "gated_abstained": 0.04411764705882353,
141
+ "n_calls": 12.735294117647058,
142
+ "n_edges_seen": 119.42647058823529,
143
+ "n_repeats": 0.0,
144
+ "n_malformed": 0.0,
145
+ "n_tool_errors": 0.0,
146
+ "median_n_edges_seen": 106.5,
147
+ "median_edges_to_answer": 23.5,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.8676470588235294,
150
+ "censored": 0.029411764705882353
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.006623931623931625,
156
+ "halluc_soft": 0.0014957264957264956,
157
+ "path_recall": 0.6384615384615384,
158
+ "path_precision": 0.5430789580789581,
159
+ "em_strict": 0.7538461538461538,
160
+ "em_claim": 0.7538461538461538,
161
+ "success": 0.7282051282051282,
162
+ "evidence_recall": 0.82991452991453,
163
+ "reached_answer": 0.8512820512820513,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.09743589743589744,
166
+ "efficiency": 0.04066962604313636,
167
+ "gate_pass": 0.9692307692307692,
168
+ "gated_acc": 0.7282051282051282,
169
+ "gated_abstained": 0.03076923076923077,
170
+ "n_calls": 13.384615384615385,
171
+ "n_edges_seen": 117.8051282051282,
172
+ "n_repeats": 0.005128205128205128,
173
+ "n_malformed": 0.0,
174
+ "n_tool_errors": 0.015384615384615385,
175
+ "median_n_edges_seen": 108.0,
176
+ "median_edges_to_answer": 24.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8512820512820513,
179
+ "censored": 0.09743589743589744
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 3.9,
186
+ "p50_s": 4.7,
187
+ "p75_s": 5.7,
188
+ "questions_per_minute": 12.68,
189
+ "turns_per_episode": 14.9,
190
+ "s_per_turn": 0.3
191
+ }
192
+ }
data/runs/q2bpool32-s118--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s119--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s119--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s119--tsmc_core_agentic/summary.json ADDED
@@ -0,0 +1,192 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vllm:Qwen/Qwen3.5-2B",
3
+ "benchmark": "data/benchmarks/tsmc/tsmc_core_agentic.jsonl",
4
+ "agentic": true,
5
+ "shard": null,
6
+ "ids_file": null,
7
+ "adapter": "models/traj_sft_2b_flash20g2/final,models/traj_grpo_2b_flash20d1/checkpoint-100,models/traj_grpo_2b_flash20d2/checkpoint-120",
8
+ "backend": "tsmc",
9
+ "kg_fingerprint": "ba22334639149e9d",
10
+ "kb_path": "None",
11
+ "source_text": "absent",
12
+ "chunks_dir": null,
13
+ "chat_template": "plain_chatml",
14
+ "max_model_len": 16384,
15
+ "timing": {
16
+ "driver": "batched",
17
+ "sequential": false,
18
+ "wall_ms_semantics": "since_batch_start",
19
+ "concurrency": null,
20
+ "batch_size": null,
21
+ "n_episodes": 195,
22
+ "num_samples": 1,
23
+ "group_size": null,
24
+ "total_wall_ms": 145314,
25
+ "cache_hits": null,
26
+ "api_calls": null
27
+ },
28
+ "usage": {
29
+ "prompt_tokens": null,
30
+ "completion_tokens": null,
31
+ "reasoning_tokens": null,
32
+ "cached_prompt_tokens": null,
33
+ "service_tier": null,
34
+ "served_tiers": null,
35
+ "tier_fallbacks": null,
36
+ "truncated": null,
37
+ "refused": null
38
+ },
39
+ "policy": {
40
+ "prompt_version": "agent_v1",
41
+ "max_turns": 20,
42
+ "turn_max_tokens": 1024,
43
+ "final_max_tokens": 900,
44
+ "max_edges_seen": 400,
45
+ "max_repeats": 3,
46
+ "repeat_echo": true,
47
+ "page_size": 40,
48
+ "hub_threshold": 150,
49
+ "hub_summary_limit": 400,
50
+ "connect_max_hops": 3,
51
+ "find_degree_tiebreak": false,
52
+ "tools": null,
53
+ "temperature": 0.7,
54
+ "reasoning_effort": null,
55
+ "seed": 119,
56
+ "give_anchor": false,
57
+ "relation_vocab": true,
58
+ "forced_prefill": false,
59
+ "user_note": null,
60
+ "gold_scaffold": false,
61
+ "citation_repair": false,
62
+ "offer_revision": false
63
+ },
64
+ "aggregate": {
65
+ "easy": {
66
+ "n": 70,
67
+ "format_rate": 1.0,
68
+ "halluc_strict": 0.0035714285714285713,
69
+ "halluc_soft": 0.0035714285714285713,
70
+ "path_recall": 0.7857142857142857,
71
+ "path_precision": 0.6229931972789116,
72
+ "em_strict": 0.8571428571428571,
73
+ "em_claim": 0.8571428571428571,
74
+ "success": 0.8428571428571429,
75
+ "evidence_recall": 0.9142857142857143,
76
+ "reached_answer": 0.9285714285714286,
77
+ "abstained": 0.0,
78
+ "answer_named_before_seen": 0.18571428571428572,
79
+ "efficiency": 0.030892668502389954,
80
+ "gate_pass": 0.9857142857142858,
81
+ "gated_acc": 0.8428571428571429,
82
+ "gated_abstained": 0.014285714285714285,
83
+ "n_calls": 12.514285714285714,
84
+ "n_edges_seen": 103.45714285714286,
85
+ "n_repeats": 0.014285714285714285,
86
+ "n_malformed": 0.014285714285714285,
87
+ "n_tool_errors": 0.014285714285714285,
88
+ "median_n_edges_seen": 99.5,
89
+ "median_edges_to_answer": 21.0,
90
+ "median_n_calls": 13.5,
91
+ "reached_rate": 0.9285714285714286,
92
+ "censored": 0.014285714285714285
93
+ },
94
+ "hard": {
95
+ "n": 57,
96
+ "format_rate": 1.0,
97
+ "halluc_strict": 0.013742690058479532,
98
+ "halluc_soft": 0.0,
99
+ "path_recall": 0.35964912280701755,
100
+ "path_precision": 0.3783625730994152,
101
+ "em_strict": 0.49122807017543857,
102
+ "em_claim": 0.49122807017543857,
103
+ "success": 0.47368421052631576,
104
+ "evidence_recall": 0.6578947368421053,
105
+ "reached_answer": 0.6666666666666666,
106
+ "abstained": 0.0,
107
+ "answer_named_before_seen": 0.08771929824561403,
108
+ "efficiency": 0.04741264099514376,
109
+ "gate_pass": 0.9473684210526315,
110
+ "gated_acc": 0.47368421052631576,
111
+ "gated_abstained": 0.05263157894736842,
112
+ "n_calls": 15.403508771929825,
113
+ "n_edges_seen": 124.54385964912281,
114
+ "n_repeats": 0.03508771929824561,
115
+ "n_malformed": 0.0,
116
+ "n_tool_errors": 0.05263157894736842,
117
+ "median_n_edges_seen": 117.0,
118
+ "median_edges_to_answer": 32.5,
119
+ "median_n_calls": 15.0,
120
+ "reached_rate": 0.6666666666666666,
121
+ "censored": 0.22807017543859648
122
+ },
123
+ "medium": {
124
+ "n": 68,
125
+ "format_rate": 1.0,
126
+ "halluc_strict": 0.007352941176470588,
127
+ "halluc_soft": 0.0,
128
+ "path_recall": 0.6519607843137255,
129
+ "path_precision": 0.5833333333333333,
130
+ "em_strict": 0.7647058823529411,
131
+ "em_claim": 0.7647058823529411,
132
+ "success": 0.7647058823529411,
133
+ "evidence_recall": 0.8480392156862745,
134
+ "reached_answer": 0.8823529411764706,
135
+ "abstained": 0.0,
136
+ "answer_named_before_seen": 0.08823529411764706,
137
+ "efficiency": 0.045530267708825234,
138
+ "gate_pass": 0.9852941176470589,
139
+ "gated_acc": 0.7647058823529411,
140
+ "gated_abstained": 0.014705882352941176,
141
+ "n_calls": 13.647058823529411,
142
+ "n_edges_seen": 119.80882352941177,
143
+ "n_repeats": 0.014705882352941176,
144
+ "n_malformed": 0.0,
145
+ "n_tool_errors": 0.029411764705882353,
146
+ "median_n_edges_seen": 109.0,
147
+ "median_edges_to_answer": 36.5,
148
+ "median_n_calls": 14.0,
149
+ "reached_rate": 0.8823529411764706,
150
+ "censored": 0.058823529411764705
151
+ },
152
+ "ALL": {
153
+ "n": 195,
154
+ "format_rate": 1.0,
155
+ "halluc_strict": 0.007863247863247862,
156
+ "halluc_soft": 0.001282051282051282,
157
+ "path_recall": 0.6145299145299145,
158
+ "path_precision": 0.5376556776556777,
159
+ "em_strict": 0.717948717948718,
160
+ "em_claim": 0.717948717948718,
161
+ "success": 0.7076923076923077,
162
+ "evidence_recall": 0.8162393162393162,
163
+ "reached_answer": 0.8358974358974359,
164
+ "abstained": 0.0,
165
+ "answer_named_before_seen": 0.12307692307692308,
166
+ "efficiency": 0.04082597710815696,
167
+ "gate_pass": 0.9743589743589743,
168
+ "gated_acc": 0.7076923076923077,
169
+ "gated_abstained": 0.02564102564102564,
170
+ "n_calls": 13.753846153846155,
171
+ "n_edges_seen": 115.32307692307693,
172
+ "n_repeats": 0.020512820512820513,
173
+ "n_malformed": 0.005128205128205128,
174
+ "n_tool_errors": 0.03076923076923077,
175
+ "median_n_edges_seen": 106.0,
176
+ "median_edges_to_answer": 27.0,
177
+ "median_n_calls": 14.0,
178
+ "reached_rate": 0.8358974358974359,
179
+ "censored": 0.09230769230769231
180
+ }
181
+ },
182
+ "latency": {
183
+ "dir": "latency",
184
+ "n": 15,
185
+ "p25_s": 4.1,
186
+ "p50_s": 4.6,
187
+ "p75_s": 5.1,
188
+ "questions_per_minute": 12.96,
189
+ "turns_per_episode": 15.9,
190
+ "s_per_turn": 0.29
191
+ }
192
+ }
data/runs/q2bpool32-s119--tsmc_core_agentic/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s120--tsmc_core_agentic/completions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/runs/q2bpool32-s120--tsmc_core_agentic/results.jsonl ADDED
The diff for this file is too large to render. See raw diff