feat(recipes): 3 GRPO env recipes + catalog size 134->137 (v0.71.30)

src/soup_cli/recipes/catalog.py

@@ -4111,6 +4111,105 @@ training:
   moe_aux_loss_coeff: 0.01
   gradient_checkpointing: true
 
+output: ./output
+""",
+    ),
+    # v0.71.30 — bundled openenv rollout envs (out-of-the-box GRPO).
+    "grpo-env-calculator": RecipeMeta(
+        model="HuggingFaceTB/SmolLM2-135M-Instruct",
+        task="grpo",
+        size="135M",
+        tags=("grpo", "openenv", "rollout", "calculator", "reasoning"),
+        description="SmolLM2-135M GRPO on the bundled calculator env (openenv rollout)",
+        yaml_str="""\
+base: HuggingFaceTB/SmolLM2-135M-Instruct
+task: grpo
+
+data:
+  train: ./data/seed_prompts.jsonl
+  format: auto
+  max_length: 512
+
+training:
+  epochs: 1
+  lr: 1e-6
+  batch_size: 4
+  lora:
+    r: 16
+    alpha: 32
+    target_modules: auto
+  grpo_beta: 0.04
+  num_generations: 4
+  reward_fn: verifiable
+  verifiable_domain: math
+  rollout_backend: openenv
+  rollout_func: soup_cli.envs.calculator:rollout
+
+output: ./output
+""",
+    ),
+    "grpo-env-retrieval-qa": RecipeMeta(
+        model="HuggingFaceTB/SmolLM2-135M-Instruct",
+        task="grpo",
+        size="135M",
+        tags=("grpo", "openenv", "rollout", "retrieval", "qa"),
+        description="SmolLM2-135M GRPO on the bundled retrieval-QA env (openenv rollout)",
+        yaml_str="""\
+base: HuggingFaceTB/SmolLM2-135M-Instruct
+task: grpo
+
+data:
+  train: ./data/seed_prompts.jsonl
+  format: auto
+  max_length: 512
+
+training:
+  epochs: 1
+  lr: 1e-6
+  batch_size: 4
+  lora:
+    r: 16
+    alpha: 32
+    target_modules: auto
+  grpo_beta: 0.04
+  num_generations: 4
+  reward_fn: accuracy
+  rollout_backend: openenv
+  rollout_func: soup_cli.envs.retrieval_qa:rollout
+
+output: ./output
+""",
+    ),
+    "grpo-env-guess-number": RecipeMeta(
+        model="HuggingFaceTB/SmolLM2-135M-Instruct",
+        task="grpo",
+        size="135M",
+        tags=("grpo", "openenv", "rollout", "deduction", "game"),
+        description="SmolLM2-135M GRPO on the bundled number-deduction env (openenv rollout)",
+        yaml_str="""\
+base: HuggingFaceTB/SmolLM2-135M-Instruct
+task: grpo
+
+data:
+  train: ./data/seed_prompts.jsonl
+  format: auto
+  max_length: 512
+
+training:
+  epochs: 1
+  lr: 1e-6
+  batch_size: 4
+  lora:
+    r: 16
+    alpha: 32
+    target_modules: auto
+  grpo_beta: 0.04
+  num_generations: 4
+  reward_fn: verifiable
+  verifiable_domain: math
+  rollout_backend: openenv
+  rollout_func: soup_cli.envs.guess_number:rollout
+
 output: ./output
 """,
     ),

tests/test_recipes.py

@@ -260,7 +260,7 @@ class TestV025NewRecipes:
             assert cfg.base == recipe.model
             assert cfg.task == recipe.task
 
-    def test_catalog_size_is_134(self):
+    def test_catalog_size_is_137(self):
         """Total catalog size — grew with each release.
 
         v0.25.0 shipped 43 recipes (29 + 9 Part A + 2 Part B tools + 3 Part E MLX).
@@ -272,10 +272,11 @@ class TestV025NewRecipes:
         v0.62.0 added 3 (raft-llama3-8b, ra-dit-retriever, ra-dit-llama3-8b) -> 116.
         v0.71.24 added 17 (2026 model-family expansion) -> 133.
         v0.71.25 added 1 (qwen2.5-coder-7b-sft) -> 134.
+        v0.71.30 added 3 (grpo-env-calculator/retrieval-qa/guess-number) -> 137.
         """
         from soup_cli.recipes.catalog import RECIPES
 
-        assert len(RECIPES) == 134
+        assert len(RECIPES) == 137
 
     def test_new_recipes_searchable(self):
         """Search returns the new recipes via keyword/task filter."""

tests/test_v07130.py

@@ -456,6 +456,38 @@ class TestGrpoPrmWiring:
         assert captured["spec"] == "accuracy"
 
 
+# ---------------------------------------------------------------------------
+# Task 6 — recipes
+# ---------------------------------------------------------------------------
+_NEW_RECIPES = ["grpo-env-calculator", "grpo-env-retrieval-qa", "grpo-env-guess-number"]
+
+
+class TestRecipes:
+    @pytest.mark.parametrize("name", _NEW_RECIPES)
+    def test_recipe_resolves(self, name):
+        from soup_cli.recipes.catalog import get_recipe
+
+        recipe = get_recipe(name)
+        assert recipe is not None
+        assert recipe.task == "grpo"
+
+    @pytest.mark.parametrize("name", _NEW_RECIPES)
+    def test_recipe_yaml_parses(self, name):
+        from soup_cli.config.loader import load_config_from_string
+        from soup_cli.recipes.catalog import get_recipe
+
+        recipe = get_recipe(name)
+        cfg = load_config_from_string(recipe.yaml_str)
+        assert cfg.task == "grpo"
+        assert cfg.training.rollout_backend == "openenv"
+        assert cfg.training.rollout_func.startswith("soup_cli.envs.")
+
+    def test_catalog_size_is_137(self):
+        from soup_cli.recipes.catalog import RECIPES
+
+        assert len(RECIPES) == 137
+
+
 class TestNoTopLevelTorch:
     def test_prm_reward_has_no_top_level_torch(self):
         import soup_cli.utils.prm_reward as mod