[misc]: surface GenerationResult.peak_memory_mb in the MiniMax H3 examples

Every generation already computes its own peak. The worker calls
torch.cuda.max_memory_allocated() at multiproc_executor.py:699 and the
value reaches the caller as GenerationResult.peak_memory_mb, but both H3
examples print only the generation time and drop it.

On a memory-constrained device that number is the one that decides whether
a configuration runs at all, and it is what a reader has to reproduce to
check any memory claim about this model. Right now they cannot: the field
is populated, handed over, and discarded.

examples/inference/optimizations/spark_benchmark.py already prints it, so
this only brings the H3 examples in line.

Two lines each, guarded on the value being present, so nothing changes for
a backend that does not report it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: KyleNeverGivesUp

examples/inference/basic/basic_fasth3.py

@@ -343,6 +343,9 @@ def run(args: argparse.Namespace) -> list[float]:
             generation_time = getattr(result, "generation_time", None)
             if generation_time is not None:
                 print(f"Generation time: {float(generation_time):.3f}s")
+            peak_memory_mb = getattr(result, "peak_memory_mb", None)
+            if peak_memory_mb is not None:
+                print(f"Peak memory: {float(peak_memory_mb):.1f} MB")
             denoise_time = _denoise_seconds(result)
             if denoise_time is not None:
                 measured_denoise_times.append(denoise_time)

examples/inference/basic/basic_minimax_h3_t2v.py

@@ -114,10 +114,14 @@ def main() -> None:
             # machine-readable: benchmark harnesses parse this line to separate
             # generation from model-load time (last occurrence = steady state)
             print(f"Generation time: {result.generation_time:.2f}s")
+        if result.peak_memory_mb is not None:
+            print(f"Peak memory: {result.peak_memory_mb:.1f} MB")
         for _ in range(args.repeats - 1):
             result = generator.generate(request)
             if result.generation_time is not None:
                 print(f"Generation time: {result.generation_time:.2f}s")
+            if result.peak_memory_mb is not None:
+                print(f"Peak memory: {result.peak_memory_mb:.1f} MB")
     finally:
         generator.shutdown()