enable skipping of SW attention layers when using FP8 KV cache (#33695)

Signed-off-by: Jonas Kuebler <kuebj@amazon.com>

tests/quantization/test_fp8.py

@@ -466,3 +466,26 @@ def test_fp8_reloading(
         weight_loader(param, torch.zeros(shape))  # cannot use empty
 
     method.process_weights_after_loading(layer)
+
+
+@pytest.mark.skipif(
+    not is_quant_method_supported("fp8"),
+    reason="FP8 is not supported on this GPU type.",
+)
+def test_kv_cache_dtype_skip_layers(vllm_runner, monkeypatch):
+    """Test that kv_cache_dtype_skip_layers skips quantization for specified layers."""
+    monkeypatch.setenv("VLLM_ALLOW_INSECURE_SERIALIZATION", "1")
+
+    with vllm_runner(
+        "facebook/opt-125m",
+        kv_cache_dtype="fp8",
+        kv_cache_dtype_skip_layers=["0", "2"],
+        enforce_eager=True,
+    ) as llm:
+
+        def check_layers(model):
+            for i, layer in enumerate(model.model.decoder.layers):
+                expected = "auto" if str(i) in ["0", "2"] else "fp8"
+                assert layer.self_attn.attn.kv_cache_dtype == expected
+
+        llm.apply_model(check_layers)

vllm/config/cache.py

@@ -87,6 +87,9 @@ class CacheConfig:
     It enables dynamic calculation of `k_scale` and `v_scale` when
     kv_cache_dtype is fp8. If `False`, the scales will be loaded from the model
     checkpoint if available. Otherwise, the scales will default to 1.0."""
+    kv_cache_dtype_skip_layers: list[str] = field(default_factory=list)
+    """Layer patterns to skip KV cache quantization. Accepts layer indices
+    (e.g., '0', '2', '4') or attention type names (e.g., 'sliding_window')."""
     cpu_kvcache_space_bytes: int | None = None
     """(CPU backend only) CPU key-value cache space."""
     mamba_page_size_padded: int | None = None

vllm/engine/arg_utils.py

@@ -597,6 +597,9 @@ class EngineArgs:
     attention_backend: AttentionBackendEnum | None = AttentionConfig.backend
 
     calculate_kv_scales: bool = CacheConfig.calculate_kv_scales
+    kv_cache_dtype_skip_layers: list[str] = get_field(
+        CacheConfig, "kv_cache_dtype_skip_layers"
+    )
     mamba_cache_dtype: MambaDType = CacheConfig.mamba_cache_dtype
     mamba_ssm_cache_dtype: MambaDType = CacheConfig.mamba_ssm_cache_dtype
     mamba_block_size: int | None = get_field(CacheConfig, "mamba_block_size")
@@ -1003,6 +1006,9 @@ class EngineArgs:
         cache_group.add_argument(
             "--calculate-kv-scales", **cache_kwargs["calculate_kv_scales"]
         )
+        cache_group.add_argument(
+            "--kv-cache-dtype-skip-layers", **cache_kwargs["kv_cache_dtype_skip_layers"]
+        )
         cache_group.add_argument(
             "--kv-sharing-fast-prefill", **cache_kwargs["kv_sharing_fast_prefill"]
         )
@@ -1578,6 +1584,7 @@ class EngineArgs:
             enable_prefix_caching=self.enable_prefix_caching,
             prefix_caching_hash_algo=self.prefix_caching_hash_algo,
             calculate_kv_scales=self.calculate_kv_scales,
+            kv_cache_dtype_skip_layers=self.kv_cache_dtype_skip_layers,
             kv_sharing_fast_prefill=self.kv_sharing_fast_prefill,
             mamba_cache_dtype=self.mamba_cache_dtype,
             mamba_ssm_cache_dtype=self.mamba_ssm_cache_dtype,

vllm/model_executor/layers/attention/attention.py

@@ -240,6 +240,31 @@ class Attention(nn.Module, AttentionLayerBase):
             and kv_cache_scheme.get("strategy") == "attn_head"
         )
 
+        # Skip quantization for specified layers
+        if cache_config is not None and cache_config.kv_cache_dtype_skip_layers:
+            from vllm.model_executor.models.utils import extract_layer_index
+
+            skip = False
+            # Check attention type
+            if (
+                sliding_window is not None
+                and "sliding_window" in cache_config.kv_cache_dtype_skip_layers
+            ):
+                skip = True
+            # Check layer index
+            layer_idx = extract_layer_index(prefix)
+            if str(layer_idx) in cache_config.kv_cache_dtype_skip_layers:
+                skip = True
+            if skip:
+                kv_cache_dtype = "auto"
+                calculate_kv_scales = False
+            logger.info(
+                "Layer %s: kv_cache_dtype=%s, sliding_window=%s",
+                prefix,
+                kv_cache_dtype,
+                sliding_window,
+            )
+
         self.kv_cache_torch_dtype = kv_cache_dtype_str_to_dtype(
             kv_cache_dtype, vllm_config.model_config
         )