Enable quantized attention in NemotronH models (#31898)

Signed-off-by: Roi Koren <roik@nvidia.com>
2026-01-07 19:37:19 +02:00
parent 30399cc725
commit bf184a6621
2 changed files with 4 additions and 0 deletions
--- a/vllm/model_executor/models/nemotron_h.py
+++ b/vllm/model_executor/models/nemotron_h.py
@@ -483,6 +483,7 @@ class NemotronHAttention(nn.Module):
            self.scaling,
            num_kv_heads=self.num_kv_heads,
            cache_config=cache_config,
+            quant_config=quant_config,
            prefix=f"{prefix}.attn",
        )