Cut HF completions before EOS so the last grid row is parsed

marc.py

@@ -560,9 +560,10 @@ class HFModel:
                 kwargs["temperature"] = temperature
             out = self.model.generate(**data, **kwargs)
             sequences = out[:, width:].tolist()
-            # Preserve EOS and <think> spellings as before; remove only batch padding
-            # after the first EOS so batching does not change the parser's input.
-            sequences = [s[:next((i + 1 for i, t in enumerate(s) if t in eos), len(s))]
+            # Cut each sequence before its first EOS (dropping batch padding too). A kept
+            # EOS is decoded onto the last grid row ("0 1<|im_end|>"), which then fails
+            # _ROW and loses that row. <think> spellings are preserved.
+            sequences = [s[:next((i for i, t in enumerate(s) if t in eos), len(s))]
                          for s in sequences]
             decoded = self.tok.batch_decode(sequences, skip_special_tokens=False)
             if len(decoded) != len(indices) * n: