Support encoder-only models without KV-Cache (#21270)

Signed-off-by: Max de Bayser <maxdebayser@gmail.com>
Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
Co-authored-by: Russell Bryant <rbryant@redhat.com>
This commit is contained in:
Maximilien de Bayser
2025-07-26 21:09:52 +08:00
committed by GitHub
co-authored by Russell Bryant
parent f27fdfc3ed
commit 1cd6eaba54
17 changed files with 352 additions and 99 deletions
+84 -7
View File
@@ -130,6 +130,8 @@ class FlashAttentionMetadata:
prefix_scheduler_metadata: Optional[torch.Tensor] = None
max_num_splits: int = 0
causal: bool = True
def _get_sliding_window_configs(
vllm_config: VllmConfig) -> set[Optional[tuple[int, int]]]:
@@ -213,6 +215,7 @@ class FlashAttentionMetadataBuilder(
seq_lens_cpu = common_attn_metadata.seq_lens_cpu
block_table_tensor = common_attn_metadata.block_table_tensor
slot_mapping = common_attn_metadata.slot_mapping
causal = common_attn_metadata.causal
# the overhead of the aot schedule is not worth it for spec-decode
aot_schedule = self.aot_schedule and not fast_build
@@ -288,7 +291,7 @@ class FlashAttentionMetadataBuilder(
max_query_len=max_query_len,
seqlens=seq_lens,
max_seq_len=max_seq_len,
causal=True)
causal=causal)
if self.use_full_cuda_graph:
assert scheduler_metadata is not None
@@ -326,7 +329,7 @@ class FlashAttentionMetadataBuilder(
suffix_kv_lens=suffix_kv_lens,
prefix_scheduler_metadata=prefix_scheduler_metadata,
max_num_splits=max_num_splits,
)
causal=causal)
return attn_metadata
def can_run_in_cudagraph(
@@ -375,11 +378,14 @@ class FlashAttentionImpl(AttentionImpl):
FlashAttentionBackend.validate_head_size(head_size)
if attn_type != AttentionType.DECODER:
raise NotImplementedError("Encoder self-attention and "
"encoder/decoder cross-attention "
"are not implemented for "
if attn_type not in [
AttentionType.DECODER, AttentionType.ENCODER_ONLY
]:
raise NotImplementedError("Encoder/decoder cross-attention "
"is not implemented for "
"FlashAttentionImpl")
self.attn_type = attn_type
self.vllm_flash_attn_version = get_flash_attn_version()
if is_quantized_kv_cache(self.kv_cache_dtype) \
and not flash_attn_supports_fp8():
@@ -422,6 +428,8 @@ class FlashAttentionImpl(AttentionImpl):
# Profiling run.
return output
attn_type = self.attn_type
# IMPORTANT!
# NOTE(woosuk): With piece-wise CUDA graphs, this method is executed in
# eager-mode PyTorch. Thus, we need to be careful about any CPU overhead
@@ -432,6 +440,18 @@ class FlashAttentionImpl(AttentionImpl):
# performance to make sure it does not introduce any overhead.
num_actual_tokens = attn_metadata.num_actual_tokens
# Handle encoder attention differently - no KV cache needed
if attn_type in (AttentionType.ENCODER_ONLY, ):
# For encoder attention,
# we use direct Q, K, V tensors without caching
return self._forward_encoder_attention(query[:num_actual_tokens],
key[:num_actual_tokens],
value[:num_actual_tokens],
output[:num_actual_tokens],
attn_metadata, layer)
# For decoder and cross-attention, use KV cache as before
key_cache, value_cache = kv_cache.unbind(0)
if self.kv_sharing_target_layer_name is None:
@@ -483,7 +503,7 @@ class FlashAttentionImpl(AttentionImpl):
seqused_k=seqused_k,
max_seqlen_k=max_seqlen_k,
softmax_scale=self.scale,
causal=True,
causal=attn_metadata.causal,
alibi_slopes=self.alibi_slopes,
window_size=self.sliding_window,
block_table=block_table,
@@ -524,6 +544,63 @@ class FlashAttentionImpl(AttentionImpl):
)
return output
def _forward_encoder_attention(
self,
query: torch.Tensor,
key: torch.Tensor,
value: torch.Tensor,
output: torch.Tensor,
attn_metadata: FlashAttentionMetadata,
layer: torch.nn.Module,
) -> torch.Tensor:
"""Forward pass for encoder attention without KV cache.
Args:
query: shape = [num_encoder_tokens, num_heads, head_size]
key: shape = [num_encoder_tokens, num_kv_heads, head_size]
value: shape = [num_encoder_tokens, num_kv_heads, head_size]
output: shape = [num_encoder_tokens, num_heads, head_size]
attn_metadata: Encoder attention metadata
layer: The attention layer
"""
# For encoder attention, process FP8 quantization if needed
if self.kv_cache_dtype.startswith("fp8"):
raise NotImplementedError(
"quantization is not supported for encoder attention")
# Use encoder-specific metadata for sequence information
cu_seqlens_q = attn_metadata.query_start_loc
cu_seqlens_k = attn_metadata.query_start_loc
max_seqlen_q = attn_metadata.max_query_len
max_seqlen_k = attn_metadata.max_query_len
descale_shape = (
cu_seqlens_q.shape[0] - 1, # type: ignore[union-attr]
self.num_kv_heads)
# Call flash attention directly on Q, K, V tensors
flash_attn_varlen_func(
q=query,
k=key,
v=value,
out=output,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_q,
max_seqlen_k=max_seqlen_k,
softmax_scale=self.scale,
causal=False, # Encoder attention is bidirectional
alibi_slopes=self.alibi_slopes,
window_size=self.sliding_window,
softcap=self.logits_soft_cap,
fa_version=self.vllm_flash_attn_version,
q_descale=layer._q_scale.expand(descale_shape),
k_descale=layer._k_scale.expand(descale_shape),
v_descale=layer._v_scale.expand(descale_shape),
)
return output
def use_cascade_attention(
common_prefix_len: int,
+3
View File
@@ -59,6 +59,8 @@ class CommonAttentionMetadata:
block_table_tensor: torch.Tensor
slot_mapping: torch.Tensor
causal: bool = True
M = TypeVar("M")
@@ -395,6 +397,7 @@ def make_local_attention_virtual_batches(
max_query_len=seqlens_q_local.max(),
block_table_tensor=block_table_local,
slot_mapping=common_attn_metadata.slot_mapping,
causal=True,
)
+6
View File
@@ -111,6 +111,12 @@ class EngineCore:
"compatibility may not be maintained.",
vllm_config.scheduler_config.scheduler_cls)
if len(kv_cache_config.kv_cache_groups) == 0:
# Encoder models without KV cache don't support
# chunked prefill. But do SSM models?
logger.info("Disabling chunked prefill for model without KVCache")
vllm_config.scheduler_config.chunked_prefill_enabled = False
self.scheduler: SchedulerInterface = Scheduler(
vllm_config=vllm_config,
kv_cache_config=kv_cache_config,
+1
View File
@@ -330,6 +330,7 @@ class EagleProposer:
max_query_len=new_query_len_per_req.max().item(),
block_table_tensor=common_attn_metadata.block_table_tensor,
slot_mapping=common_attn_metadata.slot_mapping[token_indices],
causal=True,
)
return spec_common_attn_metadata, token_indices
+4
View File
@@ -4,6 +4,7 @@ from contextlib import contextmanager
from typing import Any
import torch
import torch.nn as nn
from vllm.config import VllmConfig
from vllm.logger import init_logger
@@ -59,6 +60,9 @@ class CPUModelRunner(GPUModelRunner):
self.scheduler_config,
self.lora_config, self.device)
def get_model(self) -> nn.Module:
return self.model
def warming_up_model(self) -> None:
logger.info("Warming up model for the compilation...")
# Only generate graph for the generic shape
+147 -39
View File
@@ -126,6 +126,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
self.is_multimodal_model = model_config.is_multimodal_model
self.is_pooling_model = model_config.pooler_config is not None
self.is_encoder_only_model = False
self.model_supports_multimodal_raw_input = (
model_config.model_supports_multimodal_raw_input)
self.max_model_len = model_config.max_model_len
@@ -735,6 +736,21 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
spec_decode_common_attn_metadata = None
attn_metadata: dict[str, Any] = {}
# Prepare encoder attention metadata separately
# (encoder layers are not in KV cache groups)
if self.is_encoder_only_model:
common_attn_metadata, encoder_attn_metadata = \
self._build_encoder_only_attn_metadata(
scheduler_output)
# Add encoder attention metadata for all encoder layers
attention_layers = get_layers_from_vllm_config(
self.vllm_config, Attention)
for layer_name, attn_module in attention_layers.items():
if attn_module.attn_type == AttentionType.ENCODER_ONLY:
attn_metadata[layer_name] = encoder_attn_metadata
# Prepare the attention metadata for each KV cache group and make layers
# in the same group share the same metadata.
for kv_cache_group_id, kv_cache_group_spec in enumerate(
@@ -760,6 +776,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
max_query_len=max_num_scheduled_tokens,
block_table_tensor=blk_table_tensor,
slot_mapping=slot_mapping,
causal=True,
)
if self.speculative_config and \
@@ -2102,7 +2119,8 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
block_table_tensor=self.input_batch.block_table[
kv_cache_group_id].get_device_tensor()[:num_reqs],
slot_mapping=self.input_batch.
block_table[kv_cache_group_id].slot_mapping[:num_tokens])
block_table[kv_cache_group_id].slot_mapping[:num_tokens],
causal=True)
attn_metadata_i = self.attn_metadata_builders[
kv_cache_group_id].build_for_cudagraph_capture(
@@ -2466,6 +2484,49 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
logger.info("Graph capturing finished in %.0f secs, took %.2f GiB",
elapsed_time, cuda_graph_size / (1 << 30))
def _initialize_single_attn_backend(
self, kv_cache_spec: KVCacheSpec
) -> tuple[AttentionBackend, AttentionMetadataBuilder]:
if isinstance(kv_cache_spec, AttentionSpec):
attn_backend_i = get_attn_backend(
kv_cache_spec.head_size,
self.dtype,
kv_cache_spec.dtype,
kv_cache_spec.block_size,
self.model_config.is_attention_free,
use_mla=kv_cache_spec.use_mla,
)
if attn_backend_i is None:
error_msg = (f"Error with get_attn_backend: "
f"{kv_cache_spec.head_size=}, "
f"{self.dtype=}, {kv_cache_spec.dtype=}, "
f"{kv_cache_spec.block_size=}, "
f"{self.model_config.is_attention_free=}, "
f"{kv_cache_spec.use_mla=}")
logger.error(error_msg)
raise NotImplementedError(
"Non-Attention backend is not supported by V1 "
"GPUModelRunner.")
elif isinstance(kv_cache_spec, MambaSpec):
attn_backend_i = Mamba2AttentionBackend
else:
raise ValueError(
f"Unknown KV cache spec type: {type(kv_cache_spec)}")
attn_metadata_builder_i = attn_backend_i.get_builder_cls()(
kv_cache_spec,
self.vllm_config,
self.device,
)
if (self.full_cuda_graph
and not attn_metadata_builder_i.full_cudagraph_supported):
raise ValueError(
f"Full CUDAGraph not supported for "
f"{attn_backend_i.__name__}. Turn off CompilationConfig."
f"full_cuda_graph or use a different attention backend.")
return attn_backend_i, attn_metadata_builder_i
def initialize_attn_backend(self, kv_cache_config: KVCacheConfig) -> None:
"""
Initialize the attention backends and attention metadata builders.
@@ -2476,48 +2537,45 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
for i, kv_cache_group_spec in enumerate(
kv_cache_config.kv_cache_groups):
kv_cache_spec = kv_cache_group_spec.kv_cache_spec
if isinstance(kv_cache_spec, AttentionSpec):
attn_backend_i = get_attn_backend(
kv_cache_spec.head_size,
self.dtype,
kv_cache_spec.dtype,
kv_cache_spec.block_size,
self.model_config.is_attention_free,
use_mla=kv_cache_spec.use_mla,
)
if attn_backend_i is None:
error_msg = (f"Error with get_attn_backend: "
f"{kv_cache_spec.head_size=}, "
f"{self.dtype=}, {kv_cache_spec.dtype=}, "
f"{kv_cache_spec.block_size=}, "
f"{self.model_config.is_attention_free=}, "
f"{kv_cache_spec.use_mla=}")
logger.error(error_msg)
raise NotImplementedError(
"Non-Attention backend is not supported by V1 "
"GPUModelRunner.")
elif isinstance(kv_cache_spec, MambaSpec):
attn_backend_i = Mamba2AttentionBackend
else:
raise ValueError(
f"Unknown KV cache spec type: {type(kv_cache_spec)}")
attn_metadata_builder_i = attn_backend_i.get_builder_cls()(
kv_cache_spec,
self.vllm_config,
self.device,
)
if (self.full_cuda_graph
and not attn_metadata_builder_i.full_cudagraph_supported):
raise ValueError(
f"Full CUDAGraph not supported for "
f"{attn_backend_i.__name__}. Turn off CompilationConfig."
f"full_cuda_graph or use a different attention backend.")
attn_backend_i, attn_metadata_builder_i = \
self._initialize_single_attn_backend(kv_cache_spec)
self.attn_backends.append(attn_backend_i)
self.attn_metadata_builders.append(attn_metadata_builder_i)
if len(self.attn_backends) > 0:
return
# Check if model is encoder-only
block_size = self.vllm_config.cache_config.block_size
use_mla = self.vllm_config.model_config.use_mla
attn_specs = list[AttentionSpec]()
attn_layers = get_layers_from_vllm_config(self.vllm_config, Attention)
for attn_module in attn_layers.values():
if attn_module.attn_type == AttentionType.ENCODER_ONLY:
assert attn_module.sliding_window is None, "Sliding "
"window attention is not supported for encoder-only models"
attn_specs.append(
FullAttentionSpec(block_size=block_size,
num_kv_heads=attn_module.num_kv_heads,
head_size=attn_module.head_size,
dtype=self.kv_cache_dtype,
use_mla=use_mla))
else:
raise ValueError("Expected only encoder-only layers")
if len(attn_specs) > 0:
assert len(attn_specs) == len(attn_layers), \
"All or none of the layers are expected to be encoder-only"
attn_backend, attn_metadata_builder = \
self._initialize_single_attn_backend(attn_specs[0])
self.attn_backends.append(attn_backend)
self.attn_metadata_builders.append(attn_metadata_builder)
self.is_encoder_only_model = True
def may_reinitialize_input_batch(self,
kv_cache_config: KVCacheConfig) -> None:
"""
@@ -2833,3 +2891,53 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
page_size_padded=page_size_padded)
return kv_cache_spec
def _build_encoder_only_attn_metadata(
self, scheduler_output: "SchedulerOutput") -> \
tuple[CommonAttentionMetadata, Any]:
"""Prepare encoder attention metadata for encoder-only models.
Args:
scheduler_output: Scheduler output
Returns:
dict[str, Any]: Encoder attention metadata
"""
num_reqs = self.input_batch.num_reqs
total_num_scheduled_tokens = scheduler_output.total_num_scheduled_tokens
# Get the number of scheduled tokens for each request.
req_ids = self.input_batch.req_ids
tokens = [scheduler_output.num_scheduled_tokens[i] for i in req_ids]
max_num_scheduled_tokens = max(tokens)
# Use the first attention metadata builder
# to create encoder attention metadata
builder = self.attn_metadata_builders[0]
dummy_block_table = torch.zeros((num_reqs, 1),
dtype=torch.int32,
device=self.device)
dummy_slot_mapping = torch.zeros((total_num_scheduled_tokens, ),
dtype=torch.int32,
device=self.device)
common_metadata = CommonAttentionMetadata(
query_start_loc=self.query_start_loc[:num_reqs + 1],
query_start_loc_cpu=self.query_start_loc_cpu[:num_reqs + 1],
seq_lens=self.seq_lens[:num_reqs],
seq_lens_cpu=self.seq_lens_cpu[:num_reqs],
num_computed_tokens_cpu=self.input_batch.
num_computed_tokens_cpu_tensor[:num_reqs],
num_reqs=num_reqs,
num_actual_tokens=total_num_scheduled_tokens,
max_query_len=max_num_scheduled_tokens,
block_table_tensor=dummy_block_table,
slot_mapping=dummy_slot_mapping,
causal=False,
)
return common_metadata, builder.build(
common_prefix_len=0, # No cascade for encoder
common_attn_metadata=common_metadata,
)