moroqq commited on
Commit
8264655
·
verified ·
1 Parent(s): e7931d6

Add dataset card (cleaned)

Browse files
Files changed (1) hide show
  1. README.md +42 -55
README.md CHANGED
@@ -1,57 +1,44 @@
1
  ---
2
- dataset_info:
3
- features:
4
- - name: messages
5
- list:
6
- - name: role
7
- dtype: string
8
- - name: content
9
- dtype: string
10
- - name: metadata
11
- struct:
12
- - name: task_type
13
- dtype: string
14
- - name: description
15
- dtype: string
16
- - name: room_type
17
- dtype: string
18
- - name: difficulty
19
- dtype: string
20
- - name: num_steps
21
- dtype: int64
22
- - name: has_recovery_steps
23
- dtype: bool
24
- - name: has_close_open
25
- dtype: bool
26
- - name: has_admissible
27
- dtype: bool
28
- - name: num_actions
29
- dtype: int64
30
- - name: num_observations
31
- dtype: int64
32
- - name: num_detours
33
- dtype: int64
34
- - name: num_recovery_steps
35
- dtype: int64
36
- - name: has_open_action
37
- dtype: bool
38
- - name: subgoals
39
- list: string
40
- - name: total_chars
41
- dtype: int64
42
- - name: trajectory_outcome
43
- dtype: string
44
- - name: failure_reason
45
- dtype: string
46
- splits:
47
- - name: train
48
- num_bytes: 7654801
49
- num_examples: 2277
50
- download_size: 1475261
51
- dataset_size: 7654801
52
- configs:
53
- - config_name: default
54
- data_files:
55
- - split: train
56
- path: data/train-*
57
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: mit
3
+ task_categories:
4
+ - text-generation
5
+ - conversational
6
+ language:
7
+ - en
8
+ tags:
9
+ - synthetic
10
+ - sft
11
+ - agent
12
+ - alfworld
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  ---
14
+
15
+ # ALFWorld Trajectory Dataset v5 (Cleaned)
16
+
17
+ This dataset is a cleaned derivative of:
18
+
19
+ - `u-10bei/sft_alfworld_trajectory_dataset_v5`
20
+ - https://huggingface.co/datasets/u-10bei/sft_alfworld_trajectory_dataset_v5
21
+
22
+ ## What was removed?
23
+
24
+ We removed samples that contain the string **`Nothing happens.`** (case-insensitive) in any **user** message.
25
+
26
+ Rationale: those trajectories typically correspond to invalid / non-admissible actions (e.g. hallucinated object ids)
27
+ and can increase the probability of illegal actions after SFT.
28
+
29
+ ## Stats
30
+
31
+ - total rows: 2502
32
+ - removed rows: 225
33
+ - kept rows: 2277
34
+
35
+ ## Schema
36
+
37
+ Same as the source dataset:
38
+
39
+ - `messages`: list of `{role, content}`
40
+ - `metadata`: dict
41
+
42
+ ## Created
43
+
44
+ - 2026-02-21T09:41:34+00:00 (UTC)