mirror of
https://github.com/handsomezhuzhu/vllm-npu-plugin.git
synced 2026-02-20 11:42:30 +00:00
feat: Add Ascend NPU attention backend for vLLM using FlashAttention operators.
This commit is contained in:
@@ -5,9 +5,9 @@ Implements the ``AttentionBackend``, ``AttentionMetadata``,
|
||||
``AttentionMetadataBuilder``, and ``AttentionImpl`` interfaces using
|
||||
Huawei Ascend NPU FlashAttention operators:
|
||||
|
||||
- ``torch_npu.npu_fusion_attention`` — fused multi-head attention
|
||||
- ``torch_npu._npu_flash_attention`` — prefill attention (TND layout)
|
||||
- ``torch_npu._npu_reshape_and_cache`` — KV cache update
|
||||
- ``torch_npu.npu_incre_flash_attention`` — paged-attention decode
|
||||
- ``torch_npu._npu_paged_attention`` — paged-attention decode
|
||||
"""
|
||||
|
||||
from dataclasses import dataclass
|
||||
@@ -319,30 +319,27 @@ class AscendAttentionBackendImpl(AttentionImpl):
|
||||
):
|
||||
"""Update KV cache with new key/value tensors.
|
||||
|
||||
Matches Huawei vllm-ascend: splits kv_cache[0]/[1] and writes via
|
||||
slot_mapping indices.
|
||||
Uses ``torch_npu._npu_reshape_and_cache`` for efficient in-place
|
||||
KV cache update, matching vllm-ascend reference.
|
||||
"""
|
||||
import torch_npu # noqa: F401
|
||||
|
||||
if kv_cache.numel() > 0:
|
||||
if self._key_cache is None:
|
||||
self._key_cache, self._value_cache = kv_cache[0], kv_cache[1]
|
||||
|
||||
slots = attn_metadata.slot_mapping
|
||||
key_to_cache = key[:attn_metadata.num_actual_tokens]
|
||||
val_to_cache = value[:attn_metadata.num_actual_tokens]
|
||||
|
||||
# Use pure-PyTorch indexing (ATB reshape_and_cache crashes on this env)
|
||||
block_size = self._key_cache.shape[1]
|
||||
block_idx = slots // block_size
|
||||
block_offset = slots % block_size
|
||||
self._key_cache[block_idx, block_offset] = key_to_cache
|
||||
self._value_cache[block_idx, block_offset] = val_to_cache
|
||||
num_actual = attn_metadata.num_actual_tokens
|
||||
torch_npu._npu_reshape_and_cache(
|
||||
key=key[:num_actual],
|
||||
value=value[:num_actual],
|
||||
key_cache=self._key_cache,
|
||||
value_cache=self._value_cache,
|
||||
slot_indices=slots,
|
||||
)
|
||||
|
||||
return key, value
|
||||
|
||||
# -----------------------------------------------------------------
|
||||
# Forward dispatch
|
||||
# -----------------------------------------------------------------
|
||||
|
||||
def forward(
|
||||
self,
|
||||
layer: nn.Module,
|
||||
@@ -399,15 +396,15 @@ class AscendAttentionBackendImpl(AttentionImpl):
|
||||
query, key, value, attn_metadata, output, num_tokens
|
||||
)
|
||||
else:
|
||||
# ChunkedPrefill or PrefillCacheHit
|
||||
# ChunkedPrefill — use npu_fused_infer_attention_score
|
||||
output = self._forward_chunked_prefill(
|
||||
query, key, value, attn_metadata, output, num_tokens
|
||||
query, attn_metadata, output, num_tokens
|
||||
)
|
||||
|
||||
return output
|
||||
|
||||
# -----------------------------------------------------------------
|
||||
# Decode path — paged attention (matches Huawei _npu_paged_attention)
|
||||
# Decode path — paged attention via _npu_paged_attention
|
||||
# -----------------------------------------------------------------
|
||||
|
||||
def _forward_decode(
|
||||
@@ -417,29 +414,24 @@ class AscendAttentionBackendImpl(AttentionImpl):
|
||||
output: torch.Tensor,
|
||||
num_tokens: int,
|
||||
) -> torch.Tensor:
|
||||
"""Decode-only via npu_incre_flash_attention."""
|
||||
"""Decode-only via _npu_paged_attention (matches vllm-ascend)."""
|
||||
import torch_npu # noqa: F401
|
||||
|
||||
q = query[:num_tokens].unsqueeze(2) # (B, H, 1, D) for BNSD
|
||||
|
||||
attn_out = torch_npu.npu_incre_flash_attention(
|
||||
q,
|
||||
self._key_cache,
|
||||
self._value_cache,
|
||||
torch_npu._npu_paged_attention(
|
||||
query=query,
|
||||
key_cache=self._key_cache,
|
||||
value_cache=self._value_cache,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
num_heads=self.num_heads,
|
||||
num_key_value_heads=self.num_kv_heads,
|
||||
scale_value=self.scale,
|
||||
block_table=attn_metadata.block_tables,
|
||||
actual_seq_lengths=attn_metadata.seq_lens_list,
|
||||
block_size=self._key_cache.shape[1],
|
||||
input_layout="BNSD",
|
||||
context_lens=attn_metadata.seq_lens,
|
||||
out=output,
|
||||
)
|
||||
|
||||
output[:num_tokens] = attn_out.squeeze(2)
|
||||
return output
|
||||
|
||||
# -----------------------------------------------------------------
|
||||
# Prefill without KV cache
|
||||
# Prefill without KV cache — _npu_flash_attention (TND layout)
|
||||
# -----------------------------------------------------------------
|
||||
|
||||
def _forward_prefill_no_cache(
|
||||
@@ -451,168 +443,72 @@ class AscendAttentionBackendImpl(AttentionImpl):
|
||||
output: torch.Tensor,
|
||||
num_tokens: int,
|
||||
) -> torch.Tensor:
|
||||
"""Prefill attention without KV cache (self-attention) via per-req loop."""
|
||||
"""Prefill attention without KV cache via _npu_flash_attention.
|
||||
|
||||
Uses TND layout and a pre-built causal mask from metadata.
|
||||
This matches vllm-ascend's _forward_prefill_no_cache.
|
||||
"""
|
||||
import torch_npu # noqa: F401
|
||||
|
||||
query_start_loc = attn_metadata.query_start_loc
|
||||
seq_lens = attn_metadata.seq_lens
|
||||
num_reqs = len(seq_lens)
|
||||
mask = attn_metadata.attn_mask
|
||||
|
||||
# Iterate and process each request independently to bypass TND issues
|
||||
for i in range(num_reqs):
|
||||
start = query_start_loc[i].item()
|
||||
end = query_start_loc[i + 1].item()
|
||||
q_len = end - start
|
||||
|
||||
# Extract q, k, v (BSND)
|
||||
q = query[start:end].unsqueeze(0)
|
||||
k = key[start:end].unsqueeze(0)
|
||||
v = value[start:end].unsqueeze(0)
|
||||
|
||||
# npu_fusion_attention: True = mask out (do NOT attend)
|
||||
# Upper triangle = future tokens = should be masked out
|
||||
attn_mask = torch.ones(
|
||||
q_len, q_len, dtype=torch.bool, device=query.device
|
||||
).triu_(diagonal=1).unsqueeze(0).unsqueeze(0)
|
||||
|
||||
# Run npu_fusion_attention (BSND)
|
||||
attn_out = torch_npu.npu_fusion_attention(
|
||||
q, k, v,
|
||||
head_num=self.num_heads,
|
||||
input_layout="BSND",
|
||||
scale=self.scale,
|
||||
atten_mask=attn_mask,
|
||||
pre_tockens=2147483647,
|
||||
next_tockens=0,
|
||||
torch_npu._npu_flash_attention(
|
||||
query=query,
|
||||
key=key,
|
||||
value=value,
|
||||
mask=mask,
|
||||
seq_len=attn_metadata.seq_lens,
|
||||
scale_value=self.scale,
|
||||
num_heads=self.num_heads,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
out=output,
|
||||
)
|
||||
|
||||
output[start:end] = attn_out[0]
|
||||
|
||||
return output
|
||||
return output[:num_tokens, :, :]
|
||||
|
||||
# -----------------------------------------------------------------
|
||||
# Chunked prefill — mixed prefill+decode via npu_fusion_attention
|
||||
# (npu_fused_infer_attention_score requires 4D on older CANN)
|
||||
# Chunked prefill — npu_fused_infer_attention_score (TND layout)
|
||||
# -----------------------------------------------------------------
|
||||
|
||||
def _forward_chunked_prefill(
|
||||
self,
|
||||
query: torch.Tensor,
|
||||
key: torch.Tensor,
|
||||
value: torch.Tensor,
|
||||
attn_metadata: AscendMetadata,
|
||||
output: torch.Tensor,
|
||||
num_tokens: int,
|
||||
) -> torch.Tensor:
|
||||
"""Chunked prefill: decode tokens via npu_incre_flash_attention,
|
||||
prefill tokens via npu_fusion_attention per request."""
|
||||
"""Chunked prefill / mixed prefill+decode via
|
||||
npu_fused_infer_attention_score, matching vllm-ascend's
|
||||
_forward_v1_style."""
|
||||
import torch_npu # noqa: F401
|
||||
|
||||
query_start_loc = attn_metadata.query_start_loc
|
||||
seq_lens = attn_metadata.seq_lens
|
||||
assert self._key_cache is not None
|
||||
assert attn_metadata.attn_mask is not None
|
||||
|
||||
# Per-request query lengths
|
||||
query_lens = query_start_loc[1:] - query_start_loc[:-1]
|
||||
num_block, block_size, _, _ = self._key_cache.shape
|
||||
key = self._key_cache.view(num_block, block_size, -1)
|
||||
value = self._value_cache.view(num_block, block_size, -1)
|
||||
|
||||
decode_mask = query_lens == 1
|
||||
prefill_mask = ~decode_mask
|
||||
num_decodes = decode_mask.sum().item()
|
||||
# Trim query to actual tokens (npu_fused_infer_attention_score
|
||||
# requires query.shape[0] == query_start_loc[-1])
|
||||
actual_num_tokens = attn_metadata.query_start_loc[-1]
|
||||
q = query[:actual_num_tokens]
|
||||
|
||||
# --- Decode tokens ---
|
||||
if num_decodes > 0 and self._key_cache is not None:
|
||||
decode_indices = torch.where(decode_mask)[0]
|
||||
decode_query = query[query_start_loc[decode_indices]]
|
||||
decode_block_tables = attn_metadata.block_tables[decode_indices]
|
||||
decode_seq_lens = seq_lens[decode_indices].tolist()
|
||||
|
||||
decode_out = torch_npu.npu_incre_flash_attention(
|
||||
decode_query.unsqueeze(2),
|
||||
self._key_cache,
|
||||
self._value_cache,
|
||||
num_heads=self.num_heads,
|
||||
out, _ = torch_npu.npu_fused_infer_attention_score(
|
||||
query=q,
|
||||
key=key,
|
||||
value=value,
|
||||
atten_mask=attn_metadata.attn_mask,
|
||||
block_table=attn_metadata.block_tables,
|
||||
input_layout="TND",
|
||||
block_size=block_size,
|
||||
actual_seq_lengths=attn_metadata.actual_seq_lengths_q,
|
||||
actual_seq_lengths_kv=attn_metadata.seq_lens_list,
|
||||
num_key_value_heads=self.num_kv_heads,
|
||||
scale_value=self.scale,
|
||||
block_table=decode_block_tables,
|
||||
actual_seq_lengths=decode_seq_lens,
|
||||
block_size=self._key_cache.shape[1],
|
||||
input_layout="BNSD",
|
||||
)
|
||||
|
||||
for i, idx in enumerate(decode_indices):
|
||||
token_pos = query_start_loc[idx].item()
|
||||
output[token_pos] = decode_out[i].squeeze(1)
|
||||
|
||||
# --- Prefill tokens (per-request via npu_fusion_attention) ---
|
||||
if prefill_mask.any():
|
||||
prefill_indices = torch.where(prefill_mask)[0]
|
||||
for idx in prefill_indices:
|
||||
start = query_start_loc[idx].item()
|
||||
end = query_start_loc[idx + 1].item()
|
||||
q_len = end - start
|
||||
kv_len = seq_lens[idx].item()
|
||||
q = query[start:end]
|
||||
|
||||
if self._key_cache is not None and kv_len > q_len:
|
||||
# Gather KV from paged cache
|
||||
block_table = attn_metadata.block_tables[idx]
|
||||
bs = self._key_cache.shape[1]
|
||||
num_blocks_needed = (kv_len + bs - 1) // bs
|
||||
block_ids = block_table[:num_blocks_needed]
|
||||
|
||||
gathered_k = self._key_cache[block_ids].reshape(
|
||||
-1, self.num_kv_heads, self.head_size
|
||||
)[:kv_len]
|
||||
gathered_v = self._value_cache[block_ids].reshape(
|
||||
-1, self.num_kv_heads, self.head_size
|
||||
)[:kv_len]
|
||||
|
||||
# npu_fusion_attention: True = mask out
|
||||
# For chunked prefill, mask future positions
|
||||
causal_mask = torch.ones(
|
||||
q_len, kv_len, dtype=torch.bool, device=query.device
|
||||
).triu_(diagonal=kv_len - q_len + 1)
|
||||
# logic for chunked prefill mask (non-square)?
|
||||
# If q_len < kv_len (prefill extension), mask logic is complex.
|
||||
# Usually: mask[i, j] = True if j <= i + (kv_len - q_len).
|
||||
# tril with diagonal adjustment.
|
||||
# diagonal=kv_len - q_len ensures main diagonal alignment.
|
||||
|
||||
attn_out = torch_npu.npu_fusion_attention(
|
||||
q.unsqueeze(0),
|
||||
gathered_k.unsqueeze(0),
|
||||
gathered_v.unsqueeze(0),
|
||||
head_num=self.num_heads,
|
||||
input_layout="BSND",
|
||||
num_heads=self.num_heads,
|
||||
scale=self.scale,
|
||||
sparse_mode=0,
|
||||
atten_mask=causal_mask.unsqueeze(0).unsqueeze(0),
|
||||
pre_tockens=kv_len,
|
||||
next_tockens=0,
|
||||
sparse_mode=3,
|
||||
)
|
||||
output[start:end] = attn_out[0].squeeze(0)
|
||||
else:
|
||||
# Full self-attention (no prior cache)
|
||||
k = key[start:end]
|
||||
v = value[start:end]
|
||||
|
||||
# npu_fusion_attention: True = mask out
|
||||
causal_mask = torch.ones(
|
||||
q_len, q_len, dtype=torch.bool, device=query.device
|
||||
).triu_(diagonal=1)
|
||||
|
||||
attn_out = torch_npu.npu_fusion_attention(
|
||||
q.unsqueeze(0),
|
||||
k.unsqueeze(0),
|
||||
v.unsqueeze(0),
|
||||
head_num=self.num_heads,
|
||||
input_layout="BSND",
|
||||
scale=self.scale,
|
||||
sparse_mode=0,
|
||||
atten_mask=causal_mask.unsqueeze(0).unsqueeze(0),
|
||||
pre_tockens=q_len,
|
||||
next_tockens=0,
|
||||
)
|
||||
output[start:end] = attn_out[0].squeeze(0)
|
||||
|
||||
output[:actual_num_tokens, :, :] = out[:actual_num_tokens, :, :]
|
||||
return output
|
||||
|
||||
|
||||
Reference in New Issue
Block a user