File size: 8,136 Bytes
25d6a2b
 
21d9c9c
 
 
 
0c9e824
 
 
 
 
 
 
 
dac8b02
 
 
 
25d6a2b
 
 
 
 
 
 
 
 
 
 
 
5410627
25d6a2b
dac8b02
25d6a2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5410627
25d6a2b
5410627
25d6a2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dac8b02
 
 
 
 
 
 
 
 
25d6a2b
0c9e824
 
 
 
 
 
21d9c9c
0c9e824
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21d9c9c
 
 
 
 
 
 
 
 
0c9e824
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
---
configs:
- config_name: rat_importance
  data_files:
  - split: train
    path: data/rat_importance.jsonl
- config_name: rat_positions
  data_files:
  - split: train
    path: data/rat_positions.jsonl
- config_name: rat_rollouts
  data_files:
  - split: train
    path: data/rat_rollouts.jsonl
- config_name: simpleqa_verified_importance
  data_files:
  - split: train
    path: data/simpleqa_verified_importance.jsonl
- config_name: simpleqa_verified_positions
  data_files:
  - split: train
    path: data/simpleqa_verified_positions.jsonl
- config_name: simpleqa_verified_rollouts
  data_files:
  - split: train
    path: data/simpleqa_verified_rollouts.jsonl
---
<!-- config:simpleqa_verified -->
## simpleqa_verified

Rollouts: 102400 rows written (101900 successful after retries, 0 still failed). finish_reason counts: {'stop': 101897, 'length': 3}.
Positions: 1019 sentences over 10 instances, 0 refused.
Importance: 1009 rows, 1009 scored, alpha=0.1, mean 0.1947.

Method: Thought Anchors resampling importance. For sentence i, the set is `n_rollouts` continuations from the trace prefix before s_i; importance is KL(P_i || P_{i+1}) over bucketed answers.
Model: Qwen/Qwen3-8B (revision b968826d9c46dd6066d109eabc6255188de91218), thinking on.

### Parameters
```json
{
  "sampling": {
    "temperature": 0.9,
    "top_p": 0.95,
    "top_k": 20,
    "min_p": 0.0,
    "max_tokens": 32768,
    "seed": "not sent"
  },
  "rollout_caps": {
    "max_tokens": 16384,
    "margin": 64,
    "min_room": 256
  },
  "n_rollouts": 100,
  "n_instances": 10,
  "max_sentences": 400
}
```

### Sources and runs
Tagged traces: `connections-dev/create-interpret-task-general-tagged` @ 7a7f2b90ec22e5cdb5e9afce6fddf0bb4d9d4c12. Raw traces: `connections-dev/create-interpret-task-general` @ 4affe1fc0e49bbf908e169c827457c7a02d42fd8.
Skipped instances (too many sentences): 52.
Run commands:
```
/scratch/mw4141/code/create_interpret/anchors/cli.py rollout --tagged-repo connections-dev/create-interpret-task-general-tagged --samples-repo connections-dev/create-interpret-task-general --subset simpleqa_verified --n-instances 10 --max-sentences 400 --min-correct-frac 0.2 --max-correct-frac 0.8 --n-rollouts 100 --max-tokens 16384 --margin 64 --min-room 256 --concurrency 64 --out-dir /scratch/mw4141/anchors_out/full_simpleqa_verified --push-to-hub --push-org connections-dev --push-repo-name create-interpret-task-general-anchors --push-visibility private --config configs/general_full.json
/scratch/mw4141/code/create_interpret/anchors/cli.py rollout --tagged-repo connections-dev/create-interpret-task-general-tagged --samples-repo connections-dev/create-interpret-task-general --subset simpleqa_verified --n-instances 10 --max-sentences 400 --min-correct-frac 0.2 --max-correct-frac 0.8 --n-rollouts 100 --max-tokens 16384 --margin 64 --min-room 256 --concurrency 64 --out-dir /scratch/mw4141/anchors_out/full_simpleqa_verified --push-to-hub --push-org connections-dev --push-repo-name create-interpret-task-general-anchors --push-visibility private --config configs/general_full.json
```
git sha(s): 3df492e12210930d6418b45f2ec1cd9a6f05ee42, 3df492e12210930d6418b45f2ec1cd9a6f05ee42. Full manifest: manifests/simpleqa_verified.json.

### simpleqa_verified_rollouts columns
| Column | Description |
|---|---|
| instance_id / sample_idx | instance and the index of the tagged base trace the prefix was cut from |
| sentence_idx | the sentence the prefix stops before (the model writes its own replacement for it) |
| rollout_idx | 0..n_rollouts-1 within the set |
| content | the continuation only (starts right after the prefix; thinking, </think>, answer inline) |
| finish_reason / success / error | vLLM finish_reason; false if the request failed |
| max_tokens / prompt_tokens | per-request generation cap and prefix length in tokens |
| request_usage | usage block of the whole n-sample request |

### simpleqa_verified_positions columns
| Column | Description |
|---|---|
| sentence_idx / sentence / tags | position, its text and function tags from the tagged repo |
| n_sentences / rel_pos | trace length and sentence_idx / (n_sentences - 1) |
| prefix_chars / prompt_tokens / max_tokens | prefix length (chars, tokens) and the cap used |
| status / reason | ok, or refused (not enough room in the context window) with the reason |

### simpleqa_verified_importance columns
| Column | Description |
|---|---|
| counts | bucketed answer counts of the set at this sentence (<no_answer> = no extractable answer) |
| n_ok / n_failed / n_no_answer | completions used / failed requests / no extractable answer |
| top_answer | most common bucket |
| importance | KL(P_i || P_{i+1}); null for the last sentence or a missing set |
| alpha | smoothing used |

<!-- /config:simpleqa_verified -->

<!-- config:rat -->
## rat

Rollouts: 133400 rows written (133400 successful after retries, 0 still failed). finish_reason counts: {'stop': 133400}.
Positions: 1334 sentences over 10 instances, 0 refused.
Importance: 1324 rows, 1324 scored, alpha=0.1, mean 0.0858.

Method: Thought Anchors resampling importance. For sentence i, the set is `n_rollouts` continuations from the trace prefix before s_i; importance is KL(P_i || P_{i+1}) over bucketed answers.
Model: Qwen/Qwen3-8B (revision b968826d9c46dd6066d109eabc6255188de91218), thinking on.

### Parameters
```json
{
  "sampling": {
    "temperature": 0.9,
    "top_p": 0.95,
    "top_k": 20,
    "min_p": 0.0,
    "max_tokens": 32768,
    "seed": "not sent"
  },
  "rollout_caps": {
    "max_tokens": 16384,
    "margin": 64,
    "min_room": 256
  },
  "n_rollouts": 100,
  "n_instances": 10,
  "max_sentences": 400
}
```

### Sources and runs
Tagged traces: `connections-dev/create-interpret-task-general-tagged` @ 7a7f2b90ec22e5cdb5e9afce6fddf0bb4d9d4c12. Raw traces: `connections-dev/create-interpret-task-general` @ 4affe1fc0e49bbf908e169c827457c7a02d42fd8.
Skipped instances (too many sentences): 39.
Run commands:
```
/scratch/mw4141/code/create_interpret/anchors/cli.py rollout --tagged-repo connections-dev/create-interpret-task-general-tagged --samples-repo connections-dev/create-interpret-task-general --subset rat --n-instances 10 --max-sentences 400 --min-correct-frac 0.2 --max-correct-frac 0.8 --n-rollouts 100 --max-tokens 16384 --margin 64 --min-room 256 --concurrency 64 --out-dir /scratch/mw4141/anchors_out/full_rat --push-to-hub --push-org connections-dev --push-repo-name create-interpret-task-general-anchors --push-visibility private --config configs/general_full.json
```
git sha(s): 3df492e12210930d6418b45f2ec1cd9a6f05ee42. Full manifest: manifests/rat.json.

### rat_rollouts columns
| Column | Description |
|---|---|
| instance_id / sample_idx | instance and the index of the tagged base trace the prefix was cut from |
| sentence_idx | the sentence the prefix stops before (the model writes its own replacement for it) |
| rollout_idx | 0..n_rollouts-1 within the set |
| content | the continuation only (starts right after the prefix; thinking, </think>, answer inline) |
| finish_reason / success / error | vLLM finish_reason; false if the request failed |
| max_tokens / prompt_tokens | per-request generation cap and prefix length in tokens |
| request_usage | usage block of the whole n-sample request |

### rat_positions columns
| Column | Description |
|---|---|
| sentence_idx / sentence / tags | position, its text and function tags from the tagged repo |
| n_sentences / rel_pos | trace length and sentence_idx / (n_sentences - 1) |
| prefix_chars / prompt_tokens / max_tokens | prefix length (chars, tokens) and the cap used |
| status / reason | ok, or refused (not enough room in the context window) with the reason |

### rat_importance columns
| Column | Description |
|---|---|
| counts | bucketed answer counts of the set at this sentence (<no_answer> = no extractable answer) |
| n_ok / n_failed / n_no_answer | completions used / failed requests / no extractable answer |
| top_answer | most common bucket |
| importance | KL(P_i || P_{i+1}); null for the last sentence or a missing set |
| alpha | smoothing used |

<!-- /config:rat -->