feat(data): add When2Call multi-turn tool-use dataset preset

Register nvidia/When2Call (train_sft split) in both dataset registries: the
offline preprocessing DATASET_CONFIGS (configs.py) and the on-policy
response-regeneration script's DATASET_CONFIGS.

When2Call is a multi-turn function-calling SFT dataset (CC-BY-4.0) that also
teaches when NOT to call a tool. Its conversations use the OpenAI `messages`
schema (mapped to `conversations`), and its tool schemas ride along in a
parallel `tools` column.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LRRGNt2T9ktWyFRboNUZNU
Signed-off-by: Ranran Haoran Zhang <ranzhang@redhat.com>

Author: WindChimeRan

scripts/response_regeneration/script.py

@@ -35,6 +35,12 @@
         "default_split": "train",
         "subset": "main",
     },
+    "when2call": {
+        "id": "nvidia/When2Call",
+        "prompt_field": "messages",
+        "default_split": "train",
+        "subset": "train_sft",
+    },
 }
 
 

src/speculators/data_generation/configs.py

@@ -47,6 +47,12 @@ def _normalize_nemotron(example: dict) -> dict:
     }
 
 
+def _normalize_when2call(example: dict) -> dict:
+    # When2Call already stores the dialogue as role/content ``messages``;
+    # the parallel ``tools`` column is preserved by the map for tool-aware runs.
+    return {"conversations": example["messages"]}
+
+
 def get_coco_dir():
     return os.getenv("COCO_DIR") or "coco/"
 
@@ -146,4 +152,11 @@ def _normalize_sharegpt4v_coco(example: dict) -> dict:
         hf_path="mlabonne/open-perfectblend",
         split="train",
     ),
+    "when2call": DatasetConfig(
+        name="when2call",
+        hf_path="nvidia/When2Call",
+        subset="train_sft",
+        split="train",
+        normalize_fn=_normalize_when2call,
+    ),
 }