cleanup: strip verbose comments from physix/training/dataset.py

physix/training/dataset.py

@@ -1,19 +1,4 @@
-"""Build the prompt dataset for GRPO training.
-
-Responsibility: enumerate the curriculum of physical systems, simulate each
-one a configurable number of times, and emit a :class:`datasets.Dataset`
-whose rows contain everything the training loop needs:
-
-- ``prompt``: the chat-format string passed to the model
-- ``system_id``, ``state_variables``, ``parameters``, ``initial_conditions``,
-  ``timestamps``, ``observed``: the system context the scorer needs
-- ``previous_r_match``: 0.0 at turn-0 (we train on first-turn prompts; the
-  iterative refinement skill emerges from the model's general ability to
-  read history at inference time)
-
-Multi-turn prompts can be added later by extending this builder; the
-hackathon scope deliberately keeps it to turn-0 prompts.
-"""
+"""Build the prompt+context dataset for GRPO training."""
 
 from __future__ import annotations