Add test/train split and synthetic templates (#3)

- add test/train split (bf921180bf711231b895573283316bbc6fa0a096)

README.md

@@ -17,34 +17,54 @@ tags:
 configs:
 - config_name: eng
   data_files:
-  - split: original
-    path: data/eng/original-*.parquet
-  - split: synthetic
-    path: data/eng/synthetic-*.parquet
+  - split: test_original
+    path: data/eng/test_original-*.parquet
+  - split: test_synthetic
+    path: data/eng/test_synthetic-*.parquet
+  - split: train_original
+    path: data/eng/train_original-*.parquet
+  - split: train_synthetic
+    path: data/eng/train_synthetic-*.parquet
 - config_name: dan
   data_files:
-  - split: original
-    path: data/dan/original-*.parquet
-  - split: synthetic
-    path: data/dan/synthetic-*.parquet
+  - split: test_original
+    path: data/dan/test_original-*.parquet
+  - split: test_synthetic
+    path: data/dan/test_synthetic-*.parquet
+  - split: train_original
+    path: data/dan/train_original-*.parquet
+  - split: train_synthetic
+    path: data/dan/train_synthetic-*.parquet
 - config_name: nob
   data_files:
-  - split: original
-    path: data/nob/original-*.parquet
-  - split: synthetic
-    path: data/nob/synthetic-*.parquet
+  - split: test_original
+    path: data/nob/test_original-*.parquet
+  - split: test_synthetic
+    path: data/nob/test_synthetic-*.parquet
+  - split: train_original
+    path: data/nob/train_original-*.parquet
+  - split: train_synthetic
+    path: data/nob/train_synthetic-*.parquet
 - config_name: deu
   data_files:
-  - split: original
-    path: data/deu/original-*.parquet
-  - split: synthetic
-    path: data/deu/synthetic-*.parquet
+  - split: test_original
+    path: data/deu/test_original-*.parquet
+  - split: test_synthetic
+    path: data/deu/test_synthetic-*.parquet
+  - split: train_original
+    path: data/deu/train_original-*.parquet
+  - split: train_synthetic
+    path: data/deu/train_synthetic-*.parquet
 - config_name: isl
   data_files:
-  - split: original
-    path: data/isl/original-*.parquet
-  - split: synthetic
-    path: data/isl/synthetic-*.parquet
+  - split: test_original
+    path: data/isl/test_original-*.parquet
+  - split: test_synthetic
+    path: data/isl/test_synthetic-*.parquet
+  - split: train_original
+    path: data/isl/train_original-*.parquet
+  - split: train_synthetic
+    path: data/isl/train_synthetic-*.parquet
 ---
 
 # Multilingual GSM-Symbolic
@@ -62,12 +82,15 @@ Want to add your own language to the dataset? It only requires validating 100 te
 
 ## Dataset Structure
 
-Each language is a separate **config** (subset); each config has two **splits**:
+Each language is a separate **config** (subset); each config has up to four **splits**. Only english has a train set currently.:
 
 | Split | Description |
 |-------|-------------|
-| `original` | The 100 concrete GSM problems for that language |
-| `synthetic` | 20 generated variants per template (2 000 problems) |
+| `test_original` | The 100 concrete GSM problems for that language |
+| `test_synthetic` | 20 generated variants per template (2 000 problems) |
+| `train_original` | The 7.47k concrete GSM problems for that language from the GSM 8k test set |
+| `train_synthetic` | 20 generated variants per template (170k problems) |
+
 
 ### Fields
 

data/dan/{original-00000-of-00001.parquet → test_original-00000-of-00001.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f002789d0af6d45e4519bf057830d0176eb60e3307e3628670b1dc80c8e931a8
-size 38164
+oid sha256:d92a2401bc9954d934ad790d43194d9065c96fdf8f298e99d529118a5748d8bf
+size 38160

data/dan/{synthetic-00000-of-00001.parquet → test_synthetic-00000-of-00001.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0af2fc0e58892230602bbc463a215cdbf2a3e5e11659edf40eece8cff380c01d
-size 309265
+oid sha256:fe6a84f73b04913132ad1452acc1ecf5842bb4016b3ccfb189ddff38f7a1025a
+size 309561

data/eng/{synthetic-00000-of-00001.parquet → test_synthetic-00000-of-00001.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b33e25b148b9976d62b0029ade9b4b914c6b801ddafc241c3f4f08f2a3d751b
-size 305700
+oid sha256:91d6cb6f44d355927bb4c9550988179a73872e42c67b65ffd782301e0dff116c
+size 305709

data/eng/train_original-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3009304640672e465bdbebb74e3d4a417ffc9ab32c3a42a59811e33875f8d5e8
+size 64632

data/eng/train_synthetic-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5560819ab719d3f96265e3253f4c90cc19874ef97033646a8aa2baa400b96ff0
+size 515471

data/generation_log.json

@@ -2,20 +2,29 @@
   "eng": {
     "seed": 42,
     "n_synthetic_per_template": 20,
-    "commit": "47d503cf59d647bc54e86cac71e12d7d1b4b4e59",
+    "commit": "1b7bcc91cf87041d669ef54a87d4fb3eeb0b65f3",
     "package_version": "0.4.4",
-    "generated_at": "2026-05-02T15:05:22.332079+00:00",
-    "n_original": 100,
-    "n_synthetic": 2000
+    "generated_at": "2026-06-03T11:10:20.882780+00:00",
+    "splits": [
+      "test",
+      "train"
+    ],
+    "test_n_original": 100,
+    "test_n_synthetic": 2000,
+    "train_n_original": 191,
+    "train_n_synthetic": 3820
   },
   "dan": {
     "seed": 42,
     "n_synthetic_per_template": 20,
-    "commit": "47d503cf59d647bc54e86cac71e12d7d1b4b4e59",
+    "commit": "30331651b506d1aabacf9e8968ac8091c1f0c2ff",
     "package_version": "0.4.4",
-    "generated_at": "2026-05-02T14:55:27.525978+00:00",
-    "n_original": 100,
-    "n_synthetic": 2000
+    "generated_at": "2026-06-03T12:44:39.848650+00:00",
+    "splits": [
+      "test"
+    ],
+    "test_n_original": 100,
+    "test_n_synthetic": 2000
   },
   "isl": {
     "seed": 42,
@@ -29,11 +38,12 @@
   "deu": {
     "seed": 42,
     "n_synthetic_per_template": 20,
-    "commit": "47d503cf59d647bc54e86cac71e12d7d1b4b4e59",
+    "commit": "1b7bcc91cf87041d669ef54a87d4fb3eeb0b65f3",
     "package_version": "0.4.4",
-    "generated_at": "2026-05-02T15:00:28.154629+00:00",
-    "n_original": 100,
-    "n_synthetic": 2000
+    "generated_at": "2026-06-03T10:54:20.991288+00:00",
+    "splits": [
+      "train"
+    ]
   },
   "nob": {
     "seed": 42,

generate_hf_dataset.py

@@ -14,13 +14,17 @@ Produces parquet files and a generation log under ``hf_dataset/data/<lang>/``.
 No upload is performed — push to the Hub separately when ready.
 
 Usage:
-    uv run scripts/generate_hf_dataset.py --lang eng
-    uv run scripts/generate_hf_dataset.py --lang dan --seed 42
-    uv run scripts/generate_hf_dataset.py --lang eng --n 20
+    uv run hf_dataset/generate_hf_dataset.py --lang eng
+    uv run hf_dataset/generate_hf_dataset.py --lang dan --seed 42
+    uv run hf_dataset/generate_hf_dataset.py --lang eng --n 20
+    uv run hf_dataset/generate_hf_dataset.py --lang eng --split test
+    uv run hf_dataset/generate_hf_dataset.py --lang eng --split train
 
 Output per language:
-    hf_dataset/data/<lang>/original-00000-of-00001.parquet
-    hf_dataset/data/<lang>/synthetic-00000-of-00001.parquet
+    hf_dataset/data/<lang>/test_original-00000-of-00001.parquet
+    hf_dataset/data/<lang>/test_synthetic-00000-of-00001.parquet
+    hf_dataset/data/<lang>/train_original-00000-of-00001.parquet
+    hf_dataset/data/<lang>/train_synthetic-00000-of-00001.parquet
     hf_dataset/data/<lang>/generation_log.json
 """
 
@@ -44,6 +48,8 @@ DEFAULT_SEED = 42
 DEFAULT_N = 20
 HF_DATA_DIR = Path(__file__).resolve().parent / "data"
 GENERATION_LOG = HF_DATA_DIR / "generation_log.json"
+TEMPLATE_DATA_ROOT = Path(multilingual_gsm_symbolic.__file__).resolve().parent / "data" / "templates"
+VALID_DATA_SPLITS = ("test", "train")
 
 logging.basicConfig(level=logging.INFO, format="%(asctime)s %(message)s", datefmt="%H:%M:%S")
 logging.getLogger("multilingual_gsm_symbolic").setLevel(logging.WARNING)
@@ -61,7 +67,7 @@ def _extract_target(answer: str) -> str:
     return answer.split("####")[-1].strip()
 
 
-def build_original_rows(lang: str) -> list[dict]:
+def build_original_rows(lang: str, split: str = "test") -> list[dict]:
     return [
         {
             "question": template.question,
@@ -70,13 +76,13 @@ def build_original_rows(lang: str) -> list[dict]:
             "language": lang,
             "source_id": template.id_orig,
         }
-        for template in load_data(lang)
+        for template in load_data(lang, directory=TEMPLATE_DATA_ROOT / lang / split)
     ]
 
 
-def build_synthetic_rows(lang: str, n: int, rng: Random) -> list[dict]:
+def build_synthetic_rows(lang: str, n: int, rng: Random, split: str = "test") -> list[dict]:
     replacements = load_replacements(lang)
-    templates = load_data(lang)
+    templates = load_data(lang, directory=TEMPLATE_DATA_ROOT / lang / split)
     rows = []
     for i, template in enumerate(templates, 1):
         questions = template.generate_questions(n=n, replacements=replacements, rng=rng)
@@ -94,27 +100,34 @@ def build_synthetic_rows(lang: str, n: int, rng: Random) -> list[dict]:
     return rows
 
 
-def generate(lang: str, seed: int, n: int) -> None:
+def generate(lang: str, seed: int, n: int, splits: list[str]) -> None:
     out_dir = HF_DATA_DIR / lang
     out_dir.mkdir(parents=True, exist_ok=True)
 
-    log.info(f"[{lang}] Building original split...")
-    original_rows = build_original_rows(lang)
+    dataset = DatasetDict()
+    split_counts = {}
+    available = [candidate for candidate in VALID_DATA_SPLITS if (TEMPLATE_DATA_ROOT / lang / candidate).exists()]
+    for split in splits:
+        if split not in available:
+            print(f"[{lang}] Split '{split}' is not available for this language.")
+            continue
 
-    log.info(f"[{lang}] Building synthetic split (n={n}, seed={seed})...")
-    rng = Random(seed)
-    synthetic_rows = build_synthetic_rows(lang, n=n, rng=rng)
+        log.info(f"[{lang}] Building {split} original split...")
+        original_rows = build_original_rows(lang, split=split)
+
+        log.info(f"[{lang}] Building {split} synthetic split (n={n}, seed={seed})...")
+        rng = Random(seed)
+        synthetic_rows = build_synthetic_rows(lang, n=n, rng=rng, split=split)
+
+        dataset[f"{split}_original"] = Dataset.from_list(original_rows)
+        dataset[f"{split}_synthetic"] = Dataset.from_list(synthetic_rows)
+        split_counts[f"{split}_n_original"] = len(original_rows)
+        split_counts[f"{split}_n_synthetic"] = len(synthetic_rows)
 
-    dataset = DatasetDict(
-        {
-            "original": Dataset.from_list(original_rows),
-            "synthetic": Dataset.from_list(synthetic_rows),
-        }
-    )
     log.info(f"[{lang}] {dataset}")
 
-    dataset["original"].to_parquet(out_dir / "original-00000-of-00001.parquet")
-    dataset["synthetic"].to_parquet(out_dir / "synthetic-00000-of-00001.parquet")
+    for split_name, split_dataset in dataset.items():
+        split_dataset.to_parquet(out_dir / f"{split_name}-00000-of-00001.parquet")
     log.info(f"[{lang}] Parquet files written to {out_dir}")
 
     # Update the shared generation log
@@ -127,8 +140,8 @@ def generate(lang: str, seed: int, n: int) -> None:
         "commit": _git_commit(),
         "package_version": multilingual_gsm_symbolic.__version__,
         "generated_at": datetime.now(UTC).isoformat(),
-        "n_original": len(original_rows),
-        "n_synthetic": len(synthetic_rows),
+        "splits": splits,
+        **split_counts,
     }
     GENERATION_LOG.write_text(json.dumps(existing, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
     log.info(f"[{lang}] Generation log updated at {GENERATION_LOG}")
@@ -141,16 +154,23 @@ def main() -> None:
     parser.add_argument(
         "--n", type=int, default=DEFAULT_N, help=f"Synthetic questions per template (default: {DEFAULT_N})"
     )
+    parser.add_argument(
+        "--split",
+        choices=("test", "train", "both"),
+        default="test",
+        help="Dataset split to generate: test, train, or both (default: test)",
+    )
     args = parser.parse_args()
 
     available = available_languages()
     langs = list(available) if args.lang is None else [args.lang]
+    splits = list(VALID_DATA_SPLITS) if args.split == "both" else [args.split]
 
     for lang in langs:
         if lang not in available:
             print(f"Unknown language '{lang}'. Available: {', '.join(available)}", file=sys.stderr)
             sys.exit(1)
-        generate(lang=lang, seed=args.seed, n=args.n)
+        generate(lang=lang, seed=args.seed, n=args.n, splits=splits)
 
 
 if __name__ == "__main__":