scrub internal project references from generate.py

generate.py

@@ -1,6 +1,6 @@
 """Raw bundled inference CLI for min-spark (no transformers dependency).
 
-Mirrors spaces/min-spark-preview/loader.py's generation loop exactly: EOS
+Mirrors the Space loader's generation loop exactly: EOS
 prefix once, truncate to the last max_seq_len tokens, effort -> loop count.
 This is the second, self-contained integration path; the Transformers path is
 modeling_minspark.py. Prefer the Transformers path unless you want zero
@@ -36,7 +36,7 @@ def load_model(ckpt_path: str | None = None, device: str = "cpu") -> Meiosis:
 @torch.no_grad()
 def generate(model, tokenizer, prompt: str, *, loops: int, max_new: int,
              temperature: float, top_k: int, device: str):
-    """Yield decoded tokens one at a time (mirrors loader.py)."""
+    """Yield decoded tokens one at a time (mirrors the Space loader)."""
     ids = [EOS_ID] + tokenizer.encode(prompt).ids
     for _ in range(max_new):
         ctx = ids[-model.config.max_seq_len:]