Update comparison table: add BF16 variant, set BF16 as baseline

README.md

@@ -212,11 +212,12 @@ This is only a quick local parity check, not a full benchmark suite. It is inten
 
 The table below compares three locally validated TensorRT-LLM variants built for the same GPU family and the same local engine limits (`max_batch_size=1`, `max_seq_len=1024`, `max_num_tokens=256`).
 
-| Variant | Checkpoint | Engine | `short_chat_42_64` | `balanced_128_128` | `long_generation_42_192` | Quick-check overall | Quick-check change vs FP16 | Practical reading |
+| Variant | Checkpoint | Engine | `short_chat_42_64` | `balanced_128_128` | `long_generation_42_192` | Quick-check overall | Quick-check change vs BF16 | Practical reading |
 |---|---:|---:|---:|---:|---:|---:|---|---|
-| `FP16` | `3.4 GB` | `3.4 GB` | `105.48 tok/s` | `105.49 tok/s` | `105.70 tok/s` | `0.75` | `baseline` | Most conservative variant |
-| `FP8` | `2.1 GB` | `2.2 GB` | `166.72 tok/s` | `144.37 tok/s` | `151.36 tok/s` | `0.75` | `no drop on this quick-check` | Best balance in these local tests |
-| `NVFP4` | `1.6 GB` | `1.2 GB` | `199.58 tok/s` | `200.09 tok/s` | `200.28 tok/s` | `0.60` | `-15 pts on this quick-check` | Fastest and smallest, but with visible quality drop |
+| `BF16` | `3.4 GB` | `3.4 GB` | `105.27 tok/s` | `105.37 tok/s` | `105.57 tok/s` | `0.725` | `baseline` | Same size and speed as FP16, better numerical stability for training |
+| `FP16` | `3.4 GB` | `3.4 GB` | `105.48 tok/s` | `105.49 tok/s` | `105.70 tok/s` | `0.75` | `+2.5 pts on this quick-check` | Most conservative variant |
+| `FP8` | `2.1 GB` | `2.2 GB` | `166.72 tok/s` | `144.37 tok/s` | `151.36 tok/s` | `0.75` | `+2.5 pts on this quick-check` | Best balance in these local tests |
+| `NVFP4` | `1.6 GB` | `1.2 GB` | `199.58 tok/s` | `200.09 tok/s` | `200.28 tok/s` | `0.60` | `-12.5 pts on this quick-check` | Fastest and smallest, but with visible quality drop |
 
 This comparison is intentionally local and narrow. It should not be treated as a universal benchmark across all prompts, datasets, GPUs, or TensorRT-LLM versions.