Dataset Viewer
Auto-converted to Parquet Duplicate
item_id
string
task
string
var
int64
score
float64
success
int64
rounds
int64
conversations
list
sciworld_0
boil
0
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_1
boil
1
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_10
boil
10
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_2
boil
2
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_3
boil
3
0
0
1
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_4
boil
4
0
0
2
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_5
boil
5
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_6
boil
6
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_7
boil
7
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_8
boil
8
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_9
boil
9
0
0
1
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_90
change-the-state-of-matter-of
0
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_91
change-the-state-of-matter-of
1
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_100
change-the-state-of-matter-of
10
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_101
change-the-state-of-matter-of
11
0.01
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_102
change-the-state-of-matter-of
12
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_103
change-the-state-of-matter-of
13
0
0
3
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_92
change-the-state-of-matter-of
2
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_93
change-the-state-of-matter-of
3
0.03
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_94
change-the-state-of-matter-of
4
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_95
change-the-state-of-matter-of
5
0
0
4
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_96
change-the-state-of-matter-of
6
0
0
1
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_97
change-the-state-of-matter-of
7
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_98
change-the-state-of-matter-of
8
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_99
change-the-state-of-matter-of
9
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_60
freeze
0
0
0
3
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_61
freeze
1
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_70
freeze
10
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_71
freeze
11
0.05
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_72
freeze
12
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_73
freeze
13
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_62
freeze
2
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_63
freeze
3
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_64
freeze
4
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_66
freeze
6
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_67
freeze
7
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_68
freeze
8
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_69
freeze
9
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_30
melt
0
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
sciworld_31
melt
1
0
0
5
[ { "role": "user", "content": "You are an agent for science world. Every round I will give you an observation, you have to respond an action based on the observation to finish the given task. Here are the actions you may take: [{\"action\": \"open/close OBJ\", \"description\": \"open/close a container\"}, {\...
End of preview. Expand in Data Studio

student_partial_sciworld

Partial SciWorld trajectories: Qwen3-1.7B (base) acting for its first 5 turns on the 2,120-task training split. These are the student prefixes an online-ROSE run sees before the teacher takes over — captured separately so the take-over point can be studied, or teacher continuations generated offline.

What is here

rows 2,120 (one per training task)
file student_partial_sciworld.jsonl (11.4 MB)
model Qwen/Qwen3-1.7B, untrained base — the starting point of our ROSE/OPD runs
turns 5 (mean 4.96; a few episodes end early)
sampling temperature 1.0, max_new_tokens 512, context 16384
mean score 0.0352

Sampling matches what the student does inside the online-ROSE loop, so these prefixes are distributionally the same as the ones the teacher continues during training.

Schema

{
  "item_id": "sciworld_0",
  "task": "boil",
  "var": 0,
  "score": 0.0,          // ScienceWorld completion in [0,1] after 5 turns
  "success": 0,          // score == 1
  "rounds": 5,
  "conversations": [     // system + 24-action instruction + ack, then obs/action pairs
    {"role": "user", "content": "You are an agent for science world. ..."},
    {"role": "assistant", "content": "OK. I'll follow your instructions ..."},
    {"role": "user", "content": "<observation>"},
    {"role": "assistant", "content": "Thought: ...\nAction: ..."}
  ]
}

conversations is the full AgentGym-protocol dialogue, ready to re-render with any chat template.

Why 5 turns

Five turns is where online ROSE cuts. At that point the student has explored but has essentially completed nothing — mean score 0.0352, versus 0.0479 when the same base model runs the full 20 turns. The take-over lands after the student has committed to an approach but before the episode is decided.

For reference, on the same environment (200 held-out tasks, ≤20 turns, greedy):

Score Success Pass(>0)
Qwen3-1.7B base 0.0479 0.0000 0.1650
SFT on Qwen3-32B trajectories 0.2077 0.0450 0.4100
online ROSE (5 student + 5 teacher) 0.3127 0.0600 0.6600
Qwen3-32B teacher 0.4346 0.1100 0.9350

Environment

ScienceWorld via the AgentGym protocol. Tasks are (task_type, variation) pairs with seven variation families excluded (5-1 5-2 9-1 9-2 9-3 10-1 10-2), giving 4,639 games total; this split holds 2,120 of them. The prompt is the standard 24-action REACT instruction, and actions are parsed from the first non-empty line after Action:.

Scores are ScienceWorld's own completion signal, not a learned reward.

Downloads last month
8