Fix super3 data prep rl subgroup bootstrap

docs/nemotron/super3/README.md

@@ -51,8 +51,16 @@ $ uv run nemotron super3 data prep sft --run YOUR-CLUSTER
 $ uv run nemotron super3 sft --run YOUR-CLUSTER
 
 // Stage 2: Reinforcement Learning
-$ uv run nemotron super3 data prep rl --run YOUR-CLUSTER
-$ uv run nemotron super3 rl --run YOUR-CLUSTER
+$ uv run nemotron super3 data prep rl rlvr --run YOUR-CLUSTER
+$ uv run nemotron super3 data prep rl swe1 --run YOUR-CLUSTER
+$ uv run nemotron super3 data prep rl swe2 --run YOUR-CLUSTER
+$ uv run nemotron super3 data prep rl rlhf --run YOUR-CLUSTER
+$ uv run nemotron super3 rl rlvr -c rlvr1 --run YOUR-CLUSTER
+$ uv run nemotron super3 rl rlvr -c rlvr2 --run YOUR-CLUSTER
+$ uv run nemotron super3 rl rlvr -c rlvr3 --run YOUR-CLUSTER
+$ uv run nemotron super3 rl swe1 --run YOUR-CLUSTER
+$ uv run nemotron super3 rl swe2 --run YOUR-CLUSTER
+$ uv run nemotron super3 rl rlhf --run YOUR-CLUSTER
 
 // Stage 3: Evaluation
 $ uv run nemotron super3 eval --run YOUR-CLUSTER

src/nemotron/cli/commands/super3/data/prep/_typer_group.py

@@ -22,7 +22,7 @@
 from __future__ import annotations
 
 from nemotron.cli.commands.super3.data.prep.pretrain import META as PRETRAIN_META, pretrain
-from nemotron.cli.commands.super3.data.prep.rl import META as RL_META, rl
+from nemotron.cli.commands.super3.data.prep.rl import rl_app
 from nemotron.cli.commands.super3.data.prep.sft import META as SFT_META, sft
 from nemo_runspec.recipe_typer import RecipeTyper
 
@@ -43,4 +43,4 @@
 
 prep_app.add_recipe_command(pretrain, meta=PRETRAIN_META)
 prep_app.add_recipe_command(sft, meta=SFT_META)
-prep_app.add_recipe_command(rl, meta=RL_META)
+prep_app.add_typer(rl_app, name="rl")

src/nemotron/cli/commands/super3/data/prep/rl/_typer_group.py

@@ -12,7 +12,10 @@
 
 from nemo_runspec.recipe_typer import RecipeTyper
 
+from nemotron.cli.commands.super3.data.prep.rl.rlhf import META as RLHF_META, rlhf
 from nemotron.cli.commands.super3.data.prep.rl.rlvr import META as RLVR_META, rlvr
+from nemotron.cli.commands.super3.data.prep.rl.swe1 import META as SWE1_META, swe1
+from nemotron.cli.commands.super3.data.prep.rl.swe2 import META as SWE2_META, swe2
 
 rl_app = RecipeTyper(
     name="rl",
@@ -22,5 +25,6 @@
 )
 
 rl_app.add_recipe_command(rlvr, meta=RLVR_META, rich_help_panel="RL Sub-Stages")
-
-# TODO: Add swe1, swe2, rlhf sub-stage data prep commands as needed.
+rl_app.add_recipe_command(swe1, meta=SWE1_META, rich_help_panel="RL Sub-Stages")
+rl_app.add_recipe_command(swe2, meta=SWE2_META, rich_help_panel="RL Sub-Stages")
+rl_app.add_recipe_command(rlhf, meta=RLHF_META, rich_help_panel="RL Sub-Stages")

src/nemotron/cli/commands/super3/data/prep/rl/rlhf.py

@@ -0,0 +1,43 @@
+# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""RLHF data prep command (super3 stage3_rlhf sub-stage)."""
+
+from __future__ import annotations
+
+import typer
+
+from nemo_runspec import parse as parse_runspec
+from nemo_runspec.recipe_config import parse_recipe_config
+from nemo_runspec.recipe_typer import RecipeMeta
+
+from nemotron.cli.commands.super3.data.prep.rl._base import _execute_data_prep_rl
+
+SCRIPT_PATH = "src/nemotron/recipes/super3/stage2_rl/stage3_rlhf/data_prep.py"
+SPEC = parse_runspec(SCRIPT_PATH)
+
+META = RecipeMeta(
+    name=SPEC.name,
+    script_path=SCRIPT_PATH,
+    config_dir=str(SPEC.config_dir),
+    default_config=SPEC.config.default,
+    input_artifacts={"data": "Raw RLHF preference JSONL"},
+    output_artifacts={"data": "Split train/val JSONL"},
+)
+
+
+def rlhf(ctx: typer.Context) -> None:
+    """Prepare data for RLHF (stage3 — human preference prompts)."""
+    cfg = parse_recipe_config(ctx)
+    _execute_data_prep_rl(cfg, script_path=SCRIPT_PATH, spec=SPEC)

src/nemotron/cli/commands/super3/data/prep/rl/swe1.py

@@ -0,0 +1,43 @@
+# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""SWE1 data prep command (super3 stage2_swe1 sub-stage)."""
+
+from __future__ import annotations
+
+import typer
+
+from nemo_runspec import parse as parse_runspec
+from nemo_runspec.recipe_config import parse_recipe_config
+from nemo_runspec.recipe_typer import RecipeMeta
+
+from nemotron.cli.commands.super3.data.prep.rl._base import _execute_data_prep_rl
+
+SCRIPT_PATH = "src/nemotron/recipes/super3/stage2_rl/stage2_swe1/data_prep.py"
+SPEC = parse_runspec(SCRIPT_PATH)
+
+META = RecipeMeta(
+    name=SPEC.name,
+    script_path=SCRIPT_PATH,
+    config_dir=str(SPEC.config_dir),
+    default_config=SPEC.config.default,
+    input_artifacts={"data": "Raw SWE-Pivot JSONL"},
+    output_artifacts={"data": "Split train/val JSONL"},
+)
+
+
+def swe1(ctx: typer.Context) -> None:
+    """Prepare data for SWE1 (stage2.1 — SWE-Pivot prompts)."""
+    cfg = parse_recipe_config(ctx)
+    _execute_data_prep_rl(cfg, script_path=SCRIPT_PATH, spec=SPEC)