nzhao721 commited on
Commit
7163353
·
verified ·
1 Parent(s): 0c8da51

Add files using upload-large-folder tool

Browse files
.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ data/train.jsonl filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ language:
4
+ - en
5
+ pretty_name: "Same Content, Different Format: worked-example data"
6
+ configs:
7
+ - config_name: train
8
+ data_files: data/train.jsonl
9
+ - config_name: holdout
10
+ data_files: data/holdout.jsonl
11
+ tags:
12
+ - math
13
+ - worked-examples
14
+ - gsm8k
15
+ - continual-pretraining
16
+ ---
17
+
18
+ # Same Content, Different Format: worked-example data
19
+
20
+ Data for the worked-example experiment of *Same Content, Different Format: QA Augmentation and Faded Worked Examples
21
+ Under a Fixed Training Budget*. The six trained models are in
22
+ [nzhao721/p1-worked-examples](https://huggingface.co/nzhao721/p1-worked-examples).
23
+
24
+ ## Configs
25
+
26
+ ### `train` (10,000 rows)
27
+
28
+ The training instances: 5,000 GSM8K training families with two instances each (3,333 answer-augmented, 3,333
29
+ rephrased, 1,667 self-verification and 1,667 FOBAR), written by Claude Sonnet 5.5 with MetaMathQA's prompts and
30
+ generation code (Yu et al., 2024; commit fe667b14).
31
+
32
+ | Field | Meaning |
33
+ |---|---|
34
+ | `instance_id`, `family_id` | instance and family (one family = all rewrites of one GSM8K seed question) |
35
+ | `type` | `GSM_AnsAug`, `GSM_Rephrased`, `GSM_SV` or `GSM_FOBAR` |
36
+ | `question`, `steps`, `final_answer` | the training question, its step-by-step solution and its final answer |
37
+ | `seed_question` | the original GSM8K question of the family |
38
+ | `writer_model` | the model that wrote the instance |
39
+ | `judge_sonnet`, `judge_argon` | the two LLM judges' audits of the solution (Claude Sonnet 5.5 and Gemini 4 Argon) |
40
+
41
+ Every instance's final answer matches the gold answer, and neither judge flagged a reasoning error or a slip.
42
+
43
+ ### `holdout` (5,606 rows)
44
+
45
+ The evaluation holdout: every original question plus every rewrite whose verifying answer matched the gold and that
46
+ both judges accepted, 3,247 GSM8K and 2,359 MATH prompts across 1,488 families. No family is shared with training.
47
+ `prompt` is the direct-condition prompt (`Problem: q\nAnswer:`); the open condition uses the problem alone.
48
+ `final_answer` is the full gold answer.
49
+
50
+ ## Tokenized training streams (`tokenized/`)
51
+
52
+ The exact token streams each arm was trained on, one folder per arm (`bare`, `complete`, `fade_ordered`,
53
+ `fade_shuffled`, `bare_pm`, `complete_pm`): `shard-00000.npy` (token ids, GPT-NeoX OLMo tokenizer, packed into
54
+ 2,048-token windows in training order), `label_mask-00000.npy` (1 where a token carries loss) and `manifest.json`.
55
+ `tokenized/pack_summary.json` gives each arm's windows, tokens, loss tokens and passes. The two fade arms draw on the
56
+ same 50,000 documents in different orders.
57
+
58
+ ## Licensing
59
+
60
+ GSM8K and MATH are released under the MIT license; MetaMathQA's prompts and code are used as released. The rewrites
61
+ were written by Claude Sonnet 5.5.
data/holdout.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
data/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f459ab54b23663e7d1ac6123caf9cfe01d6e7971301cb15b14c672faa7cbc10
3
+ size 14843328
tokenized/bare/label_mask-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:959b857b87db255809df4c3575b78360d659fcb3f3e3b9178285e57d1b65d2a3
3
+ size 200015872
tokenized/bare/manifest.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "bare",
3
+ "windows": 97664,
4
+ "steps": 1526,
5
+ "tokens": 200015872,
6
+ "loss_tokens": 198821225,
7
+ "docs_used": 2975368,
8
+ "passes": 297.54,
9
+ "corpus_docs": 10000
10
+ }
tokenized/bare/shard-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:98c5c4020f1cd4baa48443eb9ae5a9dd8ec380b7b6ee4e09f92c0fe5cabe9bf0
3
+ size 800063488
tokenized/bare_pm/label_mask-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5cd5d050d2cddd8be00e93e0699358a2251c0bd01d30d7be60b71b7df0e97b65
3
+ size 12976128
tokenized/bare_pm/manifest.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "bare_pm",
3
+ "windows": 6336,
4
+ "steps": 99,
5
+ "tokens": 12976128,
6
+ "loss_tokens": 12898903,
7
+ "docs_used": 194300,
8
+ "passes": 19.43,
9
+ "corpus_docs": 10000,
10
+ "matched_to": "fade_ordered passes"
11
+ }
tokenized/bare_pm/shard-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0acbb4cd4133ff3a3d0c2c6e315a495f1e5cdd914707ccff0460bb427f30a910
3
+ size 51904512
tokenized/complete/label_mask-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e28007e2fccc4f4c7c5f1fe2b838cb0b3e81fea838f3a4d437c5c24b7e2037df
3
+ size 200015872
tokenized/complete/manifest.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "complete",
3
+ "windows": 97664,
4
+ "steps": 1526,
5
+ "tokens": 200015872,
6
+ "loss_tokens": 196918109,
7
+ "docs_used": 971639,
8
+ "passes": 97.16,
9
+ "corpus_docs": 10000
10
+ }
tokenized/complete/shard-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2080569b68e67b23c05ad05fbc4a5dad1b1f8dfbe61092de558cad3b16bd2b58
3
+ size 800063488
tokenized/complete_pm/label_mask-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c452f51481e3035a584ec86da1241c02319d27f2dbdba0337dbdbb0e17e7c609
3
+ size 39976960
tokenized/complete_pm/manifest.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "complete_pm",
3
+ "windows": 19520,
4
+ "steps": 305,
5
+ "tokens": 39976960,
6
+ "loss_tokens": 39358917,
7
+ "docs_used": 194300,
8
+ "passes": 19.43,
9
+ "corpus_docs": 10000,
10
+ "matched_to": "fade_ordered passes"
11
+ }
tokenized/complete_pm/shard-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e6c0f91c558b87ba5dc4a9bb7b4bf1165c0173945563524301222a5815e74ad1
3
+ size 159907840
tokenized/fade_ordered/label_mask-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61e7c341ab503767e9e85ab4f0b9e7f56dcff49ae2fca614026e8f62fb3aa2fb
3
+ size 200015872
tokenized/fade_ordered/manifest.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "fade_ordered",
3
+ "windows": 97664,
4
+ "steps": 1526,
5
+ "tokens": 200015872,
6
+ "loss_tokens": 128116644,
7
+ "docs_used": 971708,
8
+ "stage_windows": [
9
+ 19533,
10
+ 19533,
11
+ 19533,
12
+ 19533,
13
+ 19532
14
+ ],
15
+ "levels": [
16
+ 1.0,
17
+ 0.75,
18
+ 0.5,
19
+ 0.25,
20
+ 0.0
21
+ ],
22
+ "corpus_docs": 50000,
23
+ "passes": 19.43
24
+ }
tokenized/fade_ordered/shard-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fcc50f9531285faf221318bd6eb3a81639b8274a11660ed6285383eb08adfbae
3
+ size 800063488
tokenized/fade_shuffled/label_mask-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d675a12dde9b3e9f37973305a88821fa8bf90c5837e4cc886ab20360aab9afb
3
+ size 200015872
tokenized/fade_shuffled/manifest.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "fade_shuffled",
3
+ "windows": 97664,
4
+ "steps": 1526,
5
+ "tokens": 200015872,
6
+ "loss_tokens": 128119401,
7
+ "docs_used": 971726,
8
+ "corpus_docs": 50000,
9
+ "passes": 19.43
10
+ }
tokenized/fade_shuffled/shard-00000.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f878dac1553780548b63074bf9b0752cde13006b1f2484077da3e7bc3076508
3
+ size 800063488
tokenized/pack_summary.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "seed": 0,
3
+ "budget_tokens": 200000000.0,
4
+ "steps": 1526,
5
+ "windows": 97664,
6
+ "instances": 10000,
7
+ "arms": {
8
+ "fade_ordered": {
9
+ "arm": "fade_ordered",
10
+ "windows": 97664,
11
+ "steps": 1526,
12
+ "tokens": 200015872,
13
+ "loss_tokens": 128116644,
14
+ "docs_used": 971708,
15
+ "stage_windows": [
16
+ 19533,
17
+ 19533,
18
+ 19533,
19
+ 19533,
20
+ 19532
21
+ ],
22
+ "levels": [
23
+ 1.0,
24
+ 0.75,
25
+ 0.5,
26
+ 0.25,
27
+ 0.0
28
+ ],
29
+ "corpus_docs": 50000,
30
+ "passes": 19.43
31
+ },
32
+ "fade_shuffled": {
33
+ "arm": "fade_shuffled",
34
+ "windows": 97664,
35
+ "steps": 1526,
36
+ "tokens": 200015872,
37
+ "loss_tokens": 128119401,
38
+ "docs_used": 971726,
39
+ "corpus_docs": 50000,
40
+ "passes": 19.43
41
+ },
42
+ "bare_pm": {
43
+ "arm": "bare_pm",
44
+ "windows": 6336,
45
+ "steps": 99,
46
+ "tokens": 12976128,
47
+ "loss_tokens": 12898903,
48
+ "docs_used": 194300,
49
+ "passes": 19.43,
50
+ "corpus_docs": 10000,
51
+ "matched_to": "fade_ordered passes"
52
+ },
53
+ "complete_pm": {
54
+ "arm": "complete_pm",
55
+ "windows": 19520,
56
+ "steps": 305,
57
+ "tokens": 39976960,
58
+ "loss_tokens": 39358917,
59
+ "docs_used": 194300,
60
+ "passes": 19.43,
61
+ "corpus_docs": 10000,
62
+ "matched_to": "fade_ordered passes"
63
+ }
64
+ }
65
+ }