Dataset Viewer
Auto-converted to Parquet Duplicate
dataset
stringclasses
1 value
instance_id
stringclasses
2 values
sentence_idx
int64
0
105
sentence
stringlengths
10
156
tags
listlengths
1
3
rel_pos
float64
0
0.99
n_sentences
int64
77
107
status
stringclasses
1 value
n_ok
int64
5
5
n_failed
int64
0
0
n_no_answer
int64
0
0
counts
unknown
top_answer
stringclasses
2 values
importance
float64
0
0
alpha
float64
0.1
0.1
rat
rat-0000
0
Okay, let's see.
[ "plan_generation" ]
0
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
1
I need to find a fourth word that's related to all three: cottage, Swiss, and cake.
[ "problem_setup" ]
0.009434
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
2
Hmm. Let me think about each word and their possible connections.
[ "plan_generation" ]
0.018868
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
3
First, cottage.
[ "fact_retrieval" ]
0.028302
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
4
A cottage is a small house, often in a rural area.
[ "fact_retrieval" ]
0.037736
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
5
It could be associated with things like countryside, vacation homes, or maybe even specific types of architecture.
[ "fact_retrieval" ]
0.04717
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
6
But how does that connect to Swiss and cake?
[ "uncertainty_management" ]
0.056604
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
7
Swiss. That could refer to Switzerland, the country, or something related to Swiss culture.
[ "fact_retrieval" ]
0.066038
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
8
Swiss cheese, for example, is a thing.
[ "fact_retrieval" ]
0.075472
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
9
Also, Swiss chocolate, maybe?
[ "fact_retrieval" ]
0.084906
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
10
Or Swiss people.
[ "fact_retrieval" ]
0.09434
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
11
But how does that link with cottage and cake?
[ "uncertainty_management" ]
0.103774
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
12
Cake. A sweet dessert.
[ "fact_retrieval" ]
0.113208
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
13
Maybe a type of cake?
[ "plan_generation" ]
0.122642
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
14
Like a Swiss cake?
[ "fact_retrieval" ]
0.132075
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
15
Or maybe a cottage cake?
[ "fact_retrieval" ]
0.141509
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
16
Wait, there's something called a cottage cheese cake, but that's a specific type.
[ "fact_retrieval" ]
0.150943
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
17
But does that connect Swiss?
[ "uncertainty_management" ]
0.160377
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
18
Wait, maybe there's a common term that links all three.
[ "plan_generation" ]
0.169811
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
19
Let me think of synonyms or related terms.
[ "plan_generation" ]
0.179245
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
20
For cottage, maybe "chalet"?
[ "fact_retrieval" ]
0.188679
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
21
Because a chalet is a type of cottage, especially in Switzerland.
[ "fact_retrieval" ]
0.198113
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
22
Oh, Swiss chalet!
[ "fact_retrieval" ]
0.207547
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
23
That's a term I've heard.
[ "fact_retrieval" ]
0.216981
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
24
So chalet is a type of cottage, specifically in Swiss areas.
[ "fact_retrieval" ]
0.226415
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
25
Then, what about cake?
[ "plan_generation" ]
0.235849
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
26
Maybe a Swiss cake?
[ "fact_retrieval" ]
0.245283
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
27
Or is there a connection between chalet and cake?
[ "uncertainty_management" ]
0.254717
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
28
Alternatively, maybe the word is "cheese." Because Swiss cheese is a type of cheese, and cottage cheese is another type.
[ "plan_generation", "fact_retrieval" ]
0.264151
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
29
But how does that relate to cake?
[ "uncertainty_management" ]
0.273585
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
30
Maybe a cheese cake?
[ "fact_retrieval" ]
0.283019
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
31
But then cottage cheese cake?
[ "fact_retrieval" ]
0.292453
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
32
But that's stretching it.
[ "uncertainty_management" ]
0.301887
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
33
Wait, but maybe the answer is "cheese" because cottage cheese and Swiss cheese are both types of cheese, and then maybe cake is part of a Swiss cheese cake?
[ "active_computation" ]
0.311321
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
34
But that seems a bit forced.
[ "uncertainty_management" ]
0.320755
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
35
Or maybe "cheese" is the link.
[ "result_consolidation" ]
0.330189
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
36
But is there a better word?
[ "uncertainty_management" ]
0.339623
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
37
Wait, another angle: cottage, Swiss, and cake.
[ "plan_generation" ]
0.349057
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
38
Maybe a word that's a type of dessert?
[ "plan_generation" ]
0.358491
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
39
Like "scone"?
[ "fact_retrieval" ]
0.367925
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
40
No, not really.
[ "uncertainty_management" ]
0.377358
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
41
Or "torte"?
[ "fact_retrieval" ]
0.386792
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
42
Torte is a type of cake, especially in Swiss cuisine.
[ "fact_retrieval" ]
0.396226
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
43
So a torte is a Swiss cake.
[ "fact_retrieval" ]
0.40566
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
44
Then, a cottage might be a place where you'd have a cottage cheese tart or something?
[ "uncertainty_management" ]
0.415094
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
45
Not sure. But if the word is "torte," then it connects to cake (as a type of cake) and Swiss (since Swiss torte is a thing).
[ "active_computation" ]
0.424528
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
46
And cottage?
[ "uncertainty_management" ]
0.433962
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
47
Maybe not directly.
[ "uncertainty_management" ]
0.443396
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
48
Unless there's a cottage torte?
[ "uncertainty_management" ]
0.45283
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
49
Not sure. Alternatively, maybe the word is "cheese." Because cottage cheese is a type of cheese, Swiss cheese is another, and maybe a cheese cake.
[ "plan_generation", "fact_retrieval" ]
0.462264
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
50
So cheese connects all three.
[ "result_consolidation" ]
0.471698
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
51
But does that make sense?
[ "uncertainty_management" ]
0.481132
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
52
The question says "related to all three," so maybe that's possible.
[ "fact_retrieval" ]
0.490566
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
53
But I'm not sure if that's the intended answer.
[ "uncertainty_management" ]
0.5
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
54
Wait, maybe the answer is "chocolate." Swiss chocolate is well-known, and cottage might be part of a chocolate cake?
[ "plan_generation", "fact_retrieval" ]
0.509434
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
55
But again, not sure.
[ "uncertainty_management" ]
0.518868
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
56
Wait, going back to chalet.
[ "plan_generation" ]
0.528302
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
57
Chalet is a type of cottage, Swiss chalet.
[ "fact_retrieval" ]
0.537736
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
58
Then, maybe a chalet has a cake?
[ "active_computation" ]
0.54717
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
59
Like a Swiss cake?
[ "fact_retrieval" ]
0.556604
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
60
But that's not a direct link.
[ "uncertainty_management" ]
0.566038
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
61
Alternatively, maybe "chocolate" again.
[ "plan_generation" ]
0.575472
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
62
Or "cheese" as before.
[ "plan_generation" ]
0.584906
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
63
Alternatively, think of a word that's part of all three.
[ "plan_generation" ]
0.59434
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
64
Like "bake." Cottage baking, Swiss baking, cake baking.
[ "fact_retrieval", "active_computation" ]
0.603774
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
65
But that's too vague.
[ "uncertainty_management" ]
0.613208
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
66
Wait, there's a type of cake called a "cottage cake," which might be a simple cake.
[ "fact_retrieval" ]
0.622642
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
67
And Swiss is related to Swiss cake.
[ "fact_retrieval" ]
0.632075
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
68
But I'm not sure if there's a specific term.
[ "uncertainty_management" ]
0.641509
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
69
Alternatively, maybe the answer is "cheese" as in cottage cheese, Swiss cheese, and cheese cake.
[ "plan_generation", "fact_retrieval" ]
0.650943
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
70
So cheese connects all three.
[ "result_consolidation" ]
0.660377
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
71
That seems plausible.
[ "uncertainty_management" ]
0.669811
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
72
But I'm not 100% sure.
[ "uncertainty_management" ]
0.679245
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
73
Let me check if there's another possibility.
[ "plan_generation" ]
0.688679
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
74
Another thought: "tart." Cottage tart, Swiss tart, and cake?
[ "plan_generation", "fact_retrieval" ]
0.698113
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
75
Maybe not.
[ "uncertainty_management" ]
0.707547
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
76
Or "recipe"?
[ "plan_generation" ]
0.716981
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
77
Cottage recipe, Swiss recipe, cake recipe.
[ "fact_retrieval" ]
0.726415
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
78
But that's too generic.
[ "uncertainty_management" ]
0.735849
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
79
Wait, maybe "alpine." Swiss is Alpine, cottage could be in the Alps, and cake?
[ "plan_generation", "fact_retrieval" ]
0.745283
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
80
Maybe Alpine cake?
[ "fact_retrieval" ]
0.754717
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
81
But not sure.
[ "uncertainty_management" ]
0.764151
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
82
Hmm. I'm leaning towards "cheese" as the connecting word.
[ "result_consolidation" ]
0.773585
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
83
Because cottage cheese and Swiss cheese are both cheeses, and a cheese cake could be a type of cake.
[ "fact_retrieval" ]
0.783019
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
84
So cheese connects all three.
[ "result_consolidation" ]
0.792453
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
85
Alternatively, maybe "chocolate" as Swiss chocolate and cottage chocolate?
[ "plan_generation", "fact_retrieval" ]
0.801887
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
86
But I think cheese is more likely.
[ "uncertainty_management" ]
0.811321
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
87
But I'm not sure.
[ "uncertainty_management" ]
0.820755
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
88
Let me think again.
[ "plan_generation" ]
0.830189
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
89
Wait, another angle: cottage, Swiss, and cake.
[ "problem_setup" ]
0.839623
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
90
Maybe the answer is "sourdough"?
[ "plan_generation" ]
0.849057
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
91
No, that's bread.
[ "uncertainty_management" ]
0.858491
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
92
Or "roll"?
[ "plan_generation" ]
0.867925
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
93
No. Maybe "bake"?
[ "uncertainty_management" ]
0.877358
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
94
No. Wait, perhaps "torte" as a Swiss cake.
[ "plan_generation", "fact_retrieval" ]
0.886792
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
95
Then, cottage could be part of a cottage torte?
[ "active_computation" ]
0.896226
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
96
Not sure. Alternatively, "butter" as in Swiss butter, cottage butter, and butter in cake.
[ "plan_generation", "fact_retrieval" ]
0.90566
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
97
But that's too generic.
[ "uncertainty_management" ]
0.915094
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
98
Alternatively, maybe the answer is "pastry." Because cottage pastry, Swiss pastry, and cake is a type of pastry.
[ "plan_generation", "fact_retrieval" ]
0.924528
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
rat
rat-0000
99
But again, that's a stretch.
[ "uncertainty_management" ]
0.933962
107
ok
5
0
0
{ "cheese": 5 }
cheese
0
0.1
End of preview. Expand in Data Studio

rat

Rollouts: 920 rows written (920 successful after retries, 0 still failed). finish_reason counts: {'stop': 920}. Positions: 184 sentences over 2 instances, 0 refused. Importance: 182 rows, 182 scored, alpha=0.1, mean 0.0000.

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

{
  "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": 5,
  "n_instances": 2,
  "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): 0. 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 2 --max-sentences 400 --n-rollouts 5 --max-tokens 16384 --margin 64 --min-room 256 --concurrency 64 --out-dir /scratch/mw4141/anchors_out/smoke --push-to-hub --push-org connections-dev --push-repo-name create-interpret-task-general-anchors-smoke --push-visibility private --config configs/general_full.json

git sha(s): 9a6589e5467700aebd5edca6d77de4dc14ae5f68. 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, , 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 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
alpha smoothing used
Downloads last month
19