James-Cuda commited on
Commit
8f5edd8
·
verified ·
1 Parent(s): 491be86

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +63 -9
README.md CHANGED
@@ -30,23 +30,77 @@ configs:
30
 
31
  # tinyrouter-m1
32
 
33
- Two **separate** Milestone-1 triage sets (same schema, different domain taxonomies):
34
 
35
- | config | domains | rows | meaning |
36
- | --- | ---: | ---: | --- |
37
- | **`5-domain`** | **5** | 328,282 | `math` / `code` / `knowledge` / `commonsense` / `instruction` over the full M1 pool |
38
- | **`20-domain`** | **20** | 5,000 | **exact** [GCI-Bench](https://huggingface.co/datasets/Glint-Research/GCI_Bench) topics |
39
 
40
- 20-domain labels (official GCI topics):
 
 
 
41
 
42
- `agriculture`, `archaeology`, `architecture`, `astronomy`, `automotive`, `aviation`, `chemistry`, `cooking`, `energy`, `finance`, `gardening`, `hardware`, `marine`, `music`, `photography`, `physiology`, `sports`, `textiles`, `weather`, `wildlife`
43
 
44
- Schema: `id | domain | difficulty | prompt` (difficulty 1–5).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
45
 
46
  ```python
47
  from datasets import load_dataset
48
  ds5 = load_dataset("James-Cuda/tinyrouter-m1", "5-domain", split="train")
49
- ds20 = load_dataset("James-Cuda/tinyrouter-m1", "20-domain", split="train") # GCI-Bench topics
 
 
50
  ```
51
 
52
  Built by `scripts/build_tinyrouter_m1_slim.py`.
 
30
 
31
  # tinyrouter-m1
32
 
33
+ Two **separate** Milestone-1 triage datasets. Same schema; different domain taxonomies.
34
 
35
+ ## Overview
 
 
 
36
 
37
+ | config | #domains | train | test | total | difficulty levels present |
38
+ | --- | ---: | ---: | ---: | ---: | --- |
39
+ | **`5-domain`** | 5 | 311,868 | 16,414 | 328,282 | 1, 2, 3, 4, 5 |
40
+ | **`20-domain`** | 20 | 4,750 | 250 | 5,000 | 2, 3, 4 |
41
 
42
+ ## Schema (both)
43
 
44
+ | column | type | meaning |
45
+ | --- | --- | --- |
46
+ | `id` | string | stable row id (`source:split:idx`) |
47
+ | `domain` | string | closed domain label for that config |
48
+ | `difficulty` | int | 1–5 |
49
+ | `prompt` | string | user / problem text |
50
+
51
+ ## `5-domain` — detailed
52
+
53
+ Full Milestone-1 pool remeshed to 5 router buckets.
54
+
55
+ **Difficulty:** 1=118, 2=92,844, 3=218,087, 4=14,812, 5=2,421
56
+
57
+ | domain | description / sources | train | test | total |
58
+ | --- | --- | ---: | ---: | ---: |
59
+ | `math` | GSM8K, MATH-500, AIME, AQuA-RAT, … | 102,658 | 5,442 | 108,100 |
60
+ | `code` | HumanEval, MBPP, BigCodeBench, … | 3,009 | 158 | 3,167 |
61
+ | `knowledge` | MMLU, MMLU-Pro, GPQA, ARC, TruthfulQA, GCI, … | 134,537 | 7,109 | 141,646 |
62
+ | `commonsense` | HellaSwag | 57,027 | 2,923 | 59,950 |
63
+ | `instruction` | Dolly-15k, IFEval | 14,637 | 782 | 15,419 |
64
+
65
+ ## `20-domain` — detailed
66
+
67
+ [GCI-Bench](https://huggingface.co/datasets/Glint-Research/GCI_Bench) rows only
68
+ (GPL-3.0). `domain` = official GCI `topic`; topic labels from upstream `topicLabel`.
69
+ Difficulty: easy→2 / medium→3 / hard→4.
70
+
71
+ **Difficulty:** 2=1,447, 3=2,278, 4=1,275
72
+
73
+ | domain | topic label | train | test | total | difficulty (2/3/4) |
74
+ | --- | --- | ---: | ---: | ---: | --- |
75
+ | `agriculture` | Agriculture & Crops | 228 | 22 | 250 | 75/120/55 |
76
+ | `archaeology` | Archaeology | 237 | 13 | 250 | 69/121/60 |
77
+ | `architecture` | Architecture & Construction | 237 | 13 | 250 | 76/121/53 |
78
+ | `astronomy` | Astronomy & Space | 242 | 8 | 250 | 69/115/66 |
79
+ | `automotive` | Automotive Mechanics | 241 | 9 | 250 | 60/130/60 |
80
+ | `aviation` | Aviation | 241 | 9 | 250 | 74/117/59 |
81
+ | `chemistry` | Chemistry Lab Processes | 242 | 8 | 250 | 72/114/64 |
82
+ | `cooking` | Culinary Science | 241 | 9 | 250 | 82/105/63 |
83
+ | `energy` | Renewable Energy | 232 | 18 | 250 | 81/103/66 |
84
+ | `finance` | Personal Finance | 239 | 11 | 250 | 74/104/72 |
85
+ | `gardening` | Home Gardening | 235 | 15 | 250 | 75/110/65 |
86
+ | `hardware` | Computer Hardware | 237 | 13 | 250 | 66/113/71 |
87
+ | `marine` | Marine Biology | 237 | 13 | 250 | 87/94/69 |
88
+ | `music` | Music Theory & Instruments | 239 | 11 | 250 | 82/113/55 |
89
+ | `photography` | Photography | 236 | 14 | 250 | 60/111/79 |
90
+ | `physiology` | Human Physiology | 235 | 15 | 250 | 66/113/71 |
91
+ | `sports` | Sports Training | 233 | 17 | 250 | 76/120/54 |
92
+ | `textiles` | Textile & Fashion | 245 | 5 | 250 | 75/108/67 |
93
+ | `weather` | Weather & Climate | 239 | 11 | 250 | 70/109/71 |
94
+ | `wildlife` | Wildlife & Animal Behavior | 234 | 16 | 250 | 58/137/55 |
95
+
96
+ ## Load
97
 
98
  ```python
99
  from datasets import load_dataset
100
  ds5 = load_dataset("James-Cuda/tinyrouter-m1", "5-domain", split="train")
101
+ ds20 = load_dataset("James-Cuda/tinyrouter-m1", "20-domain", split="train")
102
+ print(ds5[0]["domain"], ds5[0]["difficulty"])
103
+ print(ds20[0]["domain"], ds20[0]["difficulty"])
104
  ```
105
 
106
  Built by `scripts/build_tinyrouter_m1_slim.py`.