maxgfr commited on
Commit
dc89f28
·
verified ·
1 Parent(s): 1c878e5

skycmd 84c6d854: skycmd LFM3 System 1 / System 2 questions

Browse files
Files changed (4) hide show
  1. MANIFEST.json +13 -0
  2. README.md +51 -0
  3. data/calib.parquet +3 -0
  4. data/test.parquet +3 -0
MANIFEST.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "commit": "84c6d854cdc2a9f6d951ebb55780060a01f38ca6",
3
+ "source_files": {
4
+ "calib.jsonl": {
5
+ "sha256": "fa59fb8ef1c170b24bc9c558b4cda4ae2d15d8c10bb4ca634de4d67ebccd079f",
6
+ "rows": 600
7
+ },
8
+ "test.jsonl": {
9
+ "sha256": "d65bdb1b659441333e5c88a16fdcaff763fb7131fa7df8c543ac3a9d332f69d4",
10
+ "rows": 1000
11
+ }
12
+ }
13
+ }
README.md ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ language:
4
+ - en
5
+ pretty_name: skycmd LFM3 System 1 / System 2 questions
6
+ task_categories:
7
+ - question-answering
8
+ tags:
9
+ - skycmd
10
+ configs:
11
+ - config_name: calib
12
+ data_files: data/calib.parquet
13
+ - config_name: test
14
+ data_files: data/test.parquet
15
+ ---
16
+
17
+ # skycmd LFM3 System 1 / System 2 questions
18
+
19
+ The calibration and test questions of the LFM3-1.7B System 1 / System 2 router (`docs/lfm3-s1s2-prereg.md`): GSM8K and MATH, drawn with a fixed seed at pinned revisions.
20
+
21
+ Part of [skycmd](https://github.com/maxgfr/skycmd), tiny decision models trained from scratch in Rust that run in the
22
+ browser. Written by `scripts/lfm3-s1s2.sh data` at commit [`84c6d854`](https://github.com/maxgfr/skycmd/commit/84c6d854cdc2a9f6d951ebb55780060a01f38ca6).
23
+ Licences of every set: [DATA_LICENSES.md](https://github.com/maxgfr/skycmd/blob/main/DATA_LICENSES.md).
24
+
25
+ Each row is one question: `id`, `task` (`gsm8k` or `math`), `question` and `answer` (the final
26
+ answer, compared after normalization).
27
+
28
+ ## Sources and licences
29
+
30
+ Each config is one source and keeps **its source's licence** (share-alike sources stay share-alike).
31
+ Attribution goes to the upstream authors; the conversion is ours.
32
+
33
+ | config | upstream | licence | how it is used | rows |
34
+ |---|---|---|---|---:|
35
+ | `calib` | openai/gsm8k (`main`, train, `740312ad`) and DigitalLearningGmbH/MATH-lighteval (train, `0530c786`) | mit | 300 + 300 calibration questions, seed 1234 | 600 |
36
+ | `test` | openai/gsm8k (`main`, test) and HuggingFaceH4/MATH-500 (`6e4ed1a2`) | mit | 500 + 500 test questions, seed 1234 | 1,000 |
37
+
38
+ ## Files
39
+
40
+ | file | rows |
41
+ |---|---:|
42
+ | `data/calib.parquet` | 600 |
43
+ | `data/test.parquet` | 1,000 |
44
+
45
+ ```python
46
+ from datasets import load_dataset
47
+ ds = load_dataset("maxgfr/skycmd-lfm3-math", "calib")
48
+ ```
49
+
50
+ skycmd wrote no row with a language model: the targets come from the upstream labels (some upstream
51
+ sets carry their own teacher soft labels, see their cards), the games' oracles or our generators.
data/calib.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a5df6739f5a24699f6ccbb84809efedea61c2276f80a0765a140488128360ca6
3
+ size 62615
data/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00e024399600adf4f12b89a876f11c8d224e7ef4e0afb75c7ddd0b82dd9729b7
3
+ size 101189