corbanyax commited on
Commit
dfb3de0
·
verified ·
1 Parent(s): 6f4d4c7

Initial release: adapters/data, card and cross-references

Browse files
README.md ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc0-1.0
3
+ language:
4
+ - en
5
+ pretty_name: Countdown-3 Puzzles with Hash-Defined Splits
6
+ size_categories:
7
+ - 100K<n<1M
8
+ task_categories:
9
+ - text-generation
10
+ - reinforcement-learning
11
+ tags:
12
+ - countdown
13
+ - arithmetic
14
+ - math-reasoning
15
+ - reinforcement-learning
16
+ - grpo
17
+ - verifiable-rewards
18
+ - synthetic
19
+ configs:
20
+ - config_name: default
21
+ data_files:
22
+ - split: pool
23
+ path: data/pool.parquet
24
+ - split: dev
25
+ path: data/dev.parquet
26
+ - split: dev200
27
+ path: data/dev200.parquet
28
+ - split: test_id
29
+ path: data/test_id.parquet
30
+ - split: test_ood_target
31
+ path: data/test_ood_target.parquet
32
+ - split: test_ood_4
33
+ path: data/test_ood_4.parquet
34
+ ---
35
+
36
+ # Countdown-3 Puzzles with Hash-Defined Splits
37
+
38
+ **Countdown-3** arithmetic puzzles for RL post-training with verifiable rewards: given three integers in [1, 50]
39
+ and a target, write an expression that uses every number exactly once with `+ - * /` and parentheses and equals the
40
+ target. The universe is **enumerated exhaustively** with exact rational arithmetic (no sampling), and splits are
41
+ defined by **SHA-256 of the number multiset**, so held-out splits use number combinations never seen in training.
42
+ Used to train and evaluate [Dress Rehearsal](https://huggingface.co/corbanyax/Dress-Rehearsal-Qwen3-0.6B-4bit-KV-LoRA) and the
43
+ [KV-cache precision study](https://huggingface.co/corbanyax/KV-Cache-Precision-GRPO-Qwen3-0.6B-LoRA).
44
+
45
+ ## Splits
46
+
47
+ | Split | Rows | Content |
48
+ |---|---|---|
49
+ | `pool` | 102,497 | training pool: every reachable integer target in [10, 100] for the remaining multisets |
50
+ | `dev` | 500 | first 500 multisets in hash order, one instance each |
51
+ | `dev200` | 200 | the first 200 problems of `dev`, the evaluation set used throughout the paper |
52
+ | `test_id` | 1,000 | next 1,000 multisets; evaluated **once**, for final policies only |
53
+ | `test_ood_target` | 1,000 | next 1,000 multisets with targets in [101, 200] |
54
+ | `test_ood_4` | 1,000 | 4 numbers in [1, 25], targets in [10, 100] |
55
+
56
+ Columns: `prompt_id` (e.g. `cd3:1-1-31=29`), `numbers` (list of int), `target` (int), `n_numbers` (int), `split`.
57
+
58
+ ```python
59
+ from datasets import load_dataset
60
+ ds = load_dataset("corbanyax/Countdown-3-Hash-Splits") # all splits
61
+ dev200 = load_dataset("corbanyax/Countdown-3-Hash-Splits", split="dev200")
62
+ ```
63
+
64
+ ## Construction
65
+
66
+ 3-number multisets (22,100) are ordered by SHA-256 of `cd3:a-b-c`. The first 500 form `dev`, the next 1,000
67
+ `test_id`, the next 1,000 `test_ood_target`; each contributes exactly one instance, the reachable target whose
68
+ instance key has the smallest SHA-256. All remaining multisets form `pool`. 4-number multisets are ordered the same
69
+ way and the first 1,000 form `test_ood_4`. Reachable values come from exact rational arithmetic over every split into
70
+ two sub-multisets; unary minus is not allowed. Rebuilding gives byte-identical files (SHA-256 of `countdown_3.parquet`
71
+ recorded in the project log).
72
+
73
+ ## Prompt and reward
74
+
75
+ System: `You are a careful assistant that solves arithmetic puzzles.` User: `Using the numbers {numbers}, create an arithmetic expression that equals
76
+ {target}. Use every number exactly once. You may use +, -, *, / and parentheses. Show brief reasoning. End with the
77
+ expression alone between answer tags, for example <answer>(100 - 75) * 60</answer>`. The format example uses numbers
78
+ above 50, so it never matches an instance. Reward 1 if the `<answer>` expression uses each number exactly once and
79
+ evaluates exactly to the target, else 0; no format credit.
80
+
81
+ The original per-file layout is kept under `raw/` (`countdown_3.parquet`, `countdown_4.parquet`).
82
+
83
+ ## Project family
84
+
85
+ | Item | Hugging Face | Kaggle mirror |
86
+ |---|---|---|
87
+ | Dress Rehearsal adapters (this method) | [corbanyax/Dress-Rehearsal-Qwen3-0.6B-4bit-KV-LoRA](https://huggingface.co/corbanyax/Dress-Rehearsal-Qwen3-0.6B-4bit-KV-LoRA) | [model](https://www.kaggle.com/models/atharvav75/dress-rehearsal-qwen3-kv4-grpo-adapters) |
88
+ | KV-cache precision study adapters (33 runs) | [corbanyax/KV-Cache-Precision-GRPO-Qwen3-0.6B-LoRA](https://huggingface.co/corbanyax/KV-Cache-Precision-GRPO-Qwen3-0.6B-LoRA) | [model](https://www.kaggle.com/models/atharvav75/kv-cache-precision-grpo-qwen3-lora-adapters) |
89
+ | Countdown-3 puzzles with hash-defined splits | [corbanyax/Countdown-3-Hash-Splits](https://huggingface.co/datasets/corbanyax/Countdown-3-Hash-Splits) | [dataset](https://www.kaggle.com/datasets/atharvav75/countdown-3-arithmetic-puzzles-hash-splits) |
90
+ | Full run logs, diagnostics, decision log | [corbanyax/KV-Cache-Precision-GRPO-Run-Logs](https://huggingface.co/datasets/corbanyax/KV-Cache-Precision-GRPO-Run-Logs) | [dataset](https://www.kaggle.com/datasets/atharvav75/dress-rehearsal-kv-precision-grpo-run-logs) |
91
+
92
+ Paper: *Beyond a Single Score: A Multi-Objective Evaluation of KV-Cache Quantization in GRPO Training* (under double-blind review, AAAI-27 Workshop on Multi-Objective and Multi-Task Optimization for Robust AI). All items are private until the review period ends.
data/dev.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b13d42c1a3170fe0d37963b9dedd4b80d6ea0179babf4a1f4a780b311c72474
3
+ size 8910
data/dev200.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ffa2d70a8062a1e36c910a2ad0ebdf281dd162a98518787d9b2823a312dca68
3
+ size 5067
data/pool.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e28f76880a5a1deab0d9224008b819aa6ac1bb68cc0ec7de5aecc1958ec99c04
3
+ size 1017261
data/test_id.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:481bed66202bf4e8f7571f65d2cdfbfc19a700b1a525b0e96731cc7f07e68363
3
+ size 15177
data/test_ood_4.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3809e5fe1a373d8081f88f7e0596189174c75570833f9379e47b966e303fdc3b
3
+ size 15799
data/test_ood_target.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2eca59e368d0f581c74f7d980b9313963ba5deb337b4666d99fb6226439cb0c7
3
+ size 15294
raw/countdown_3.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec313cb51c627fdaaba8595a2b2b43144b3ac79d7de4f07f4304dea2ae59025c
3
+ size 1056689
raw/countdown_4.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a748e480713b4d3890bf8b46583c9ea7d12dbb47de4220f6b08fed63058f34fc
3
+ size 15799