EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[feature] chunkprefill support pcp&dcp (#3801)

Browse Source 
### What this PR does / why we need it?
ChunkPrefill now can support Long Sequence Feature Pcp&Dcp

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
CI tests passed with self-test


- vLLM version: v0.11.0
- vLLM main:
83f478bb19

---------

Signed-off-by: Apocalypse990923-qshi <qiushixu@usc.edu>
Signed-off-by: Delphine-Nic <tanwenqin@huawei.com>
Co-authored-by: Delphine-Nic <tanwenqin@huawei.com>
Co-authored-by: Delphine-Nic <3834144971@qq.com>
This commit is contained in:
Apocalypse
2025-11-11 09:18:02 +08:00
committed by GitHub
parent 7ffbe73d54
commit 71866d5311
8 changed files with 1276 additions and 170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
tests/ut/attention/test_attention_v1.py
View File

@@ -83,6 +83,7 @@ class TestAscendAttentionMetadataBuilder(TestBase):
self.mock_vllm_config.compilation_config.cudagraph_mode = None self.mock_vllm_config.compilation_config.cudagraph_mode = None
self.mock_vllm_config.scheduler_config.max_num_seqs = 10 self.mock_vllm_config.scheduler_config.max_num_seqs = 10
self.mock_vllm_config.scheduler_config.decode_max_num_seqs = 10 self.mock_vllm_config.scheduler_config.decode_max_num_seqs = 10
self.mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
self.mock_device = 'cpu:0' self.mock_device = 'cpu:0'
self.builder = AscendAttentionMetadataBuilder(None, None, self.builder = AscendAttentionMetadataBuilder(None, None,
self.mock_vllm_config, self.mock_vllm_config,

3
tests/ut/attention/test_mla_v1.py
View File

@@ -484,6 +484,9 @@ class TestAscendMLAImpl(TestBase):
chunk_ctx.chunk_seq_lens = [torch.tensor([8])] chunk_ctx.chunk_seq_lens = [torch.tensor([8])]
chunk_ctx.chunk_seq_lens_npu = [torch.tensor([8])] chunk_ctx.chunk_seq_lens_npu = [torch.tensor([8])]
chunk_ctx.starts = [torch.tensor([0])] chunk_ctx.starts = [torch.tensor([0])]
chunk_ctx.max_chunk_num = 1
chunk_ctx.mask_for_non_zero_chunk = [True]
chunk_ctx.local_chunked_kv_lens = [[[[8]]]]
prefill_meta = MagicMock() prefill_meta = MagicMock()
prefill_meta.chunked_context = chunk_ctx prefill_meta.chunked_context = chunk_ctx

529
vllm_ascend/attention/attention_v1.py
View File

@@ -37,6 +37,8 @@ from vllm.v1.core.sched.output import SchedulerOutput
from vllm.v1.kv_cache_interface import AttentionSpec from vllm.v1.kv_cache_interface import AttentionSpec
from vllm_ascend.attention.utils import (AscendCommonAttentionMetadata, from vllm_ascend.attention.utils import (AscendCommonAttentionMetadata,
extract_req_dcp_by_chunk_pcp,
filter_chunked_req_indices,
split_decodes_and_prefills) split_decodes_and_prefills)
from vllm_ascend.compilation.acl_graph import (get_graph_params, from vllm_ascend.compilation.acl_graph import (get_graph_params,
update_graph_params_workspaces) update_graph_params_workspaces)
@@ -52,6 +54,7 @@ if prefill_context_parallel_enable():
get_prefill_context_model_parallel_rank, get_prefill_context_model_parallel_rank,
get_prefill_context_model_parallel_world_size get_prefill_context_model_parallel_world_size
) )
# isort: on # isort: on
@@ -155,9 +158,23 @@ class AscendPCPMetadata:
@dataclass @dataclass
class AscendMetadataForPrefill: class AscendMetadataForPrefill:
@dataclass
class ChunkedContextMetadata:
actual_chunk_seq_lengths: list[int]
mask_for_non_zero_chunk: Optional[list[bool]] = None
max_chunk_num: int = 0
local_chunked_kv_lens: Optional[list[Optional[list[Optional[list[
Optional[list[int]]]]]]]] = None
cp_kv_recover_idx_for_chunk: Optional[list[int]] = None
kv_inverse_idx_for_chunk: Optional[list[int]] = None
""" Prefill Specific Metadata for Ascend""" """ Prefill Specific Metadata for Ascend"""
pcp_metadata: Optional[AscendPCPMetadata] = None pcp_metadata: Optional[AscendPCPMetadata] = None
pcp_allgather_restore_idx: Optional[List[int]] = None pcp_allgather_restore_idx: Optional[List[int]] = None
chunked_context: Optional[ChunkedContextMetadata] = None
block_tables: torch.Tensor = None
actual_seq_lengths_q: torch.Tensor = None
@dataclass @dataclass
@@ -165,6 +182,7 @@ class AscendMetadataForDecode:
""" Decode Specific Metadata for Ascend""" """ Decode Specific Metadata for Ascend"""
num_computed_tokens_of_pcp_dcp: Optional[list[list[list[int]]]] = None num_computed_tokens_of_pcp_dcp: Optional[list[list[list[int]]]] = None
batch_seq_mask: torch.Tensor = None batch_seq_mask: torch.Tensor = None
block_tables: torch.Tensor = None
@dataclass @dataclass
@@ -237,13 +255,10 @@ class AscendAttentionMetadataBuilder:
self.max_num_blocks_per_req = cdiv( self.max_num_blocks_per_req = cdiv(
self.model_config.max_model_len, self.model_config.max_model_len,
AscendAttentionBackend.get_supported_block_size()[0]) AscendAttentionBackend.get_supported_block_size()[0])
decode_max_num_seqs = getattr(vllm_config.scheduler_config, self.batch_seq_mask_buf = torch.empty(
'decode_max_num_seqs', 0) vllm_config.scheduler_config.max_num_batched_tokens,
max_num_seqs = max(vllm_config.scheduler_config.max_num_seqs, dtype=torch.uint8,
decode_max_num_seqs) device=device)
self.batch_seq_mask_buf = torch.empty(max_num_seqs,
dtype=torch.uint8,
device=device)
self.pcp_size = get_prefill_context_model_parallel_world_size( self.pcp_size = get_prefill_context_model_parallel_world_size(
) if prefill_context_parallel_enable() else 1 ) if prefill_context_parallel_enable() else 1
self.pcp_rank = get_prefill_context_model_parallel_rank( self.pcp_rank = get_prefill_context_model_parallel_rank(
@@ -263,6 +278,27 @@ class AscendAttentionMetadataBuilder:
AscendAttentionMetadataBuilder.reorder_batch_threshold = self.decode_threshold AscendAttentionMetadataBuilder.reorder_batch_threshold = self.decode_threshold
scheduler_config = vllm_config.scheduler_config
self.block_size = vllm_config.cache_config.block_size
self.max_blocks = (vllm_config.model_config.max_model_len +
self.block_size - 1) // self.block_size
self.chunked_prefill_enabled = scheduler_config.chunked_prefill_enabled
if self.chunked_prefill_enabled:
self.chunked_prefill_workspace_size = min(
# Max sure there is enough for 8 full length request or at least
# 4 pages of cache per request
max(8 * self.model_config.max_model_len,
4 * scheduler_config.max_num_seqs * self.block_size),
128 * 1024)
assert self.chunked_prefill_workspace_size >= \
scheduler_config.max_num_seqs * self.block_size
self.chunked_prefill_workspace = torch.empty(
(self.chunked_prefill_workspace_size,
self.model_config.get_head_size()),
dtype=self.model_config.dtype,
device=device,
)
def reorder_batch(self, input_batch, def reorder_batch(self, input_batch,
scheduler_output: "SchedulerOutput") -> bool: scheduler_output: "SchedulerOutput") -> bool:
return False return False
@@ -279,9 +315,8 @@ class AscendAttentionMetadataBuilder:
num_reqs num_reqs
+ 1] + 1]
decode_threshold = 1
num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = \ num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = \
split_decodes_and_prefills(common_attn_metadata, decode_threshold=decode_threshold) split_decodes_and_prefills(common_attn_metadata, decode_threshold=self.decode_threshold)
assert num_decodes + num_prefills == num_reqs assert num_decodes + num_prefills == num_reqs
assert num_decode_tokens + num_prefill_tokens == num_actual_tokens assert num_decode_tokens + num_prefill_tokens == num_actual_tokens
@@ -302,6 +337,7 @@ class AscendAttentionMetadataBuilder:
query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu[: query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu[:
num_reqs num_reqs
+ 1] + 1]
num_computed_tokens_cpu = (seq_lens - query_lens)
if attn_state == AscendAttentionState.DecodeOnly and \ if attn_state == AscendAttentionState.DecodeOnly and \
common_attn_metadata.num_input_tokens > num_actual_tokens: common_attn_metadata.num_input_tokens > num_actual_tokens:
@@ -338,16 +374,35 @@ class AscendAttentionMetadataBuilder:
attn_mask = torch_npu.npu_format_cast(mask_nz.contiguous(), attn_mask = torch_npu.npu_format_cast(mask_nz.contiguous(),
ACL_FORMAT_FRACTAL_NZ) ACL_FORMAT_FRACTAL_NZ)
common_long_seq_metadata = common_attn_metadata.prefill_context_parallel_metadata
prefill_metadata = None prefill_metadata = None
if num_prefills > 0: decode_metadata = None
pcp_metadata = None if common_long_seq_metadata is not None:
common_long_seq_metadata = common_attn_metadata.prefill_context_parallel_metadata chunked_context_metadata = None
if common_long_seq_metadata is not None: if num_prefills > 0:
query_lens = query_lens[num_decode_tokens:]
context_lens_cpu = num_computed_tokens_cpu[
num_decodes:num_reqs]
max_context_len_cpu = context_lens_cpu.max().item()
pcp_size = get_prefill_context_model_parallel_world_size(
) if prefill_context_parallel_enable() else 1
if self.chunked_prefill_enabled and max_context_len_cpu > 0:
cp_kv_recover_idx_for_chunk = common_long_seq_metadata.cp_kv_recover_idx_for_chunk
kv_inverse_idx_for_chunk = torch.argsort(
cp_kv_recover_idx_for_chunk.to(torch.float32)
) if cp_kv_recover_idx_for_chunk is not None else None
chunked_context_metadata = \
AscendMetadataForPrefill.ChunkedContextMetadata(
actual_chunk_seq_lengths=torch.cumsum(query_lens * pcp_size, dim=0),
mask_for_non_zero_chunk=common_long_seq_metadata.mask_for_non_zero_chunk,
local_chunked_kv_lens=common_long_seq_metadata.local_chunked_kv_lens,
cp_kv_recover_idx_for_chunk=cp_kv_recover_idx_for_chunk,
kv_inverse_idx_for_chunk=kv_inverse_idx_for_chunk,
max_chunk_num=common_long_seq_metadata.max_chunk_num
)
attn_mask_seqlens = common_long_seq_metadata.attn_mask_seqlens attn_mask_seqlens = common_long_seq_metadata.attn_mask_seqlens
head_attn_nomask_seqlens = common_long_seq_metadata.head_attn_nomask_seqlens head_attn_nomask_seqlens = common_long_seq_metadata.head_attn_nomask_seqlens
tail_attn_nomask_seqlens = common_long_seq_metadata.tail_attn_nomask_seqlens tail_attn_nomask_seqlens = common_long_seq_metadata.tail_attn_nomask_seqlens
pcp_size = get_prefill_context_model_parallel_world_size(
) if prefill_context_parallel_enable() else 1
if pcp_size > 1: if pcp_size > 1:
attn_mask_seqlens = torch.cumsum(attn_mask_seqlens[0], attn_mask_seqlens = torch.cumsum(attn_mask_seqlens[0],
dim=0).tolist() dim=0).tolist()
@@ -355,6 +410,7 @@ class AscendAttentionMetadataBuilder:
head_attn_nomask_seqlens[1], dim=0).tolist() head_attn_nomask_seqlens[1], dim=0).tolist()
tail_attn_nomask_seqlens = torch.cumsum( tail_attn_nomask_seqlens = torch.cumsum(
tail_attn_nomask_seqlens[1], dim=0).tolist() tail_attn_nomask_seqlens[1], dim=0).tolist()
pcp_metadata = AscendPCPMetadata( pcp_metadata = AscendPCPMetadata(
q_head_idx=common_long_seq_metadata.q_head_idx_tensor, q_head_idx=common_long_seq_metadata.q_head_idx_tensor,
q_tail_idx=common_long_seq_metadata.q_tail_idx_tensor, q_tail_idx=common_long_seq_metadata.q_tail_idx_tensor,
@@ -371,16 +427,17 @@ class AscendAttentionMetadataBuilder:
tail_attn_nomask_seqlens=tail_attn_nomask_seqlens, tail_attn_nomask_seqlens=tail_attn_nomask_seqlens,
q_full_idx=common_long_seq_metadata.q_full_idx, q_full_idx=common_long_seq_metadata.q_full_idx,
pcp_prefill_mask=common_long_seq_metadata.pcp_prefill_mask) pcp_prefill_mask=common_long_seq_metadata.pcp_prefill_mask)
prefill_metadata = AscendMetadataForPrefill(
pcp_metadata=pcp_metadata,
pcp_allgather_restore_idx=common_long_seq_metadata.
pcp_allgather_restore_idx
if common_long_seq_metadata is not None else None)
decode_metadata = None prefill_metadata = AscendMetadataForPrefill(
if num_decodes > 0: pcp_metadata=pcp_metadata,
common_long_seq_metadata = common_attn_metadata.prefill_context_parallel_metadata pcp_allgather_restore_idx=common_long_seq_metadata.
if common_long_seq_metadata is not None: pcp_allgather_restore_idx
if common_long_seq_metadata is not None else None,
chunked_context=chunked_context_metadata,
block_tables=block_table[num_decodes:],
actual_seq_lengths_q=torch.cumsum(query_lens, dim=0))
if num_decodes > 0:
num_computed_tokens_of_pcp_dcp = common_long_seq_metadata.num_computed_tokens_of_pcp_dcp num_computed_tokens_of_pcp_dcp = common_long_seq_metadata.num_computed_tokens_of_pcp_dcp
assert num_computed_tokens_of_pcp_dcp is not None assert num_computed_tokens_of_pcp_dcp is not None
num_computed_tokens_array = np.array( num_computed_tokens_array = np.array(
@@ -397,7 +454,7 @@ class AscendAttentionMetadataBuilder:
num_computed_tokens_of_pcp_dcp=num_computed_tokens_array, num_computed_tokens_of_pcp_dcp=num_computed_tokens_array,
batch_seq_mask=self.batch_seq_mask_buf[:batch_seq_mask. batch_seq_mask=self.batch_seq_mask_buf[:batch_seq_mask.
shape[0]], shape[0]],
) block_tables=block_table[:num_decodes])
attn_metadata = AscendMetadata( attn_metadata = AscendMetadata(
num_actual_tokens=num_actual_tokens, num_actual_tokens=num_actual_tokens,
@@ -751,7 +808,8 @@ class AscendAttentionBackendImpl(AttentionImpl):
k_mask: torch.Tensor, k_mask: torch.Tensor,
v_mask: torch.Tensor, v_mask: torch.Tensor,
kv_seqlens_mask: List[int], kv_seqlens_mask: List[int],
mask: torch.Tensor) -> torch.Tensor: mask: torch.Tensor,
attn_metadata) -> torch.Tensor:
# nomask Attention # nomask Attention
if k_nomask is not None: if k_nomask is not None:
attn_out_nomask, attn_lse_nomask = torch.ops.npu.npu_fused_infer_attention_score( attn_out_nomask, attn_lse_nomask = torch.ops.npu.npu_fused_infer_attention_score(
@@ -786,30 +844,42 @@ class AscendAttentionBackendImpl(AttentionImpl):
softmax_lse_flag=True, softmax_lse_flag=True,
actual_seq_lengths_kv=kv_seqlens_mask, actual_seq_lengths_kv=kv_seqlens_mask,
actual_seq_lengths=q_seqlens) actual_seq_lengths=q_seqlens)
# update # update
output = attn_out_mask output = attn_out_mask
attn_lse = attn_lse_mask
if k_nomask is not None: if k_nomask is not None:
T = attn_out_mask.shape[0] if attn_metadata.prefill is not None and attn_metadata.prefill.chunked_context is None:
N = attn_out_mask.shape[1] output = self._npu_attn_out_lse_update(attn_lse_mask,
D = attn_out_mask.shape[2] attn_lse_nomask,
attn_out_mask,
attn_out_nomask)
attn_lse = None
else:
output, attn_lse = self._update_out_and_lse(
torch.stack([attn_out_nomask, attn_out_mask], dim=0),
torch.stack([attn_lse_nomask, attn_lse_mask], dim=0))
attn_out_mask, attn_lse_mask = self._out_lse_reshape( return output, attn_lse
attn_out_mask, attn_lse_mask)
attn_out_nomask, attn_lse_nomask = self._out_lse_reshape(
attn_out_nomask, attn_lse_nomask)
attn_out_mask = attn_out_mask.to(torch.float32)
attn_out_nomask = attn_out_nomask.to(torch.float32)
attn_lse_mask = attn_lse_mask.to(torch.float32)
attn_lse_nomask = attn_lse_nomask.to(torch.float32)
attn_output = [attn_out_nomask, attn_out_mask]
attn_lse = [attn_lse_nomask, attn_lse_mask]
update_type = 0
output, _ = torch_npu.npu_attention_update(attn_lse, attn_output,
update_type)
output = output.view(T, N, D)
def _npu_attn_out_lse_update(self, attn_lse_mask, attn_lse_nomask,
attn_out_mask, attn_out_nomask):
T = attn_out_mask.shape[0]
N = attn_out_mask.shape[1]
D = attn_out_mask.shape[2]
attn_out_mask, attn_lse_mask = self._out_lse_reshape(
attn_out_mask, attn_lse_mask)
attn_out_nomask, attn_lse_nomask = self._out_lse_reshape(
attn_out_nomask, attn_lse_nomask)
attn_out_mask = attn_out_mask.to(torch.float32)
attn_out_nomask = attn_out_nomask.to(torch.float32)
attn_lse_mask = attn_lse_mask.to(torch.float32)
attn_lse_nomask = attn_lse_nomask.to(torch.float32)
attn_output = [attn_out_nomask, attn_out_mask]
attn_lse = [attn_lse_nomask, attn_lse_mask]
update_type = 0
output, _ = torch_npu.npu_attention_update(attn_lse, attn_output,
update_type)
output = output.view(T, N, D)
return output return output
def _forward_prefill_cp(self, query: torch.Tensor, key: torch.Tensor, def _forward_prefill_cp(self, query: torch.Tensor, key: torch.Tensor,
@@ -831,7 +901,7 @@ class AscendAttentionBackendImpl(AttentionImpl):
mask = attn_metadata.prefill.pcp_metadata.pcp_prefill_mask mask = attn_metadata.prefill.pcp_metadata.pcp_prefill_mask
# 1. Attention calculation in the first half of Q in load balancing # 1. Attention calculation in the first half of Q in load balancing
output_head = self._attention_with_nomask_and_mask( output_heads, lse_heads = self._attention_with_nomask_and_mask(
q=torch.index_select(query, 0, q_head_idx), q=torch.index_select(query, 0, q_head_idx),
q_seqlens=attn_mask_seqlens, q_seqlens=attn_mask_seqlens,
k_nomask=torch.index_select(key, 0, kv_with_q_head_nomask_idx) k_nomask=torch.index_select(key, 0, kv_with_q_head_nomask_idx)
@@ -842,12 +912,13 @@ class AscendAttentionBackendImpl(AttentionImpl):
k_mask=torch.index_select(key, 0, kv_with_q_head_mask_idx), k_mask=torch.index_select(key, 0, kv_with_q_head_mask_idx),
v_mask=torch.index_select(value, 0, kv_with_q_head_mask_idx), v_mask=torch.index_select(value, 0, kv_with_q_head_mask_idx),
kv_seqlens_mask=attn_mask_seqlens, kv_seqlens_mask=attn_mask_seqlens,
mask=mask) mask=mask,
attn_metadata=attn_metadata)
# 2. the Attention calculation in the latter half of Q in load balancing # 2. the Attention calculation in the latter half of Q in load balancing
# pcp_rank0: Q3*KV0~KV2 + Q3*KV3 # pcp_rank0: Q3*KV0~KV2 + Q3*KV3
# pcp_rank1: Q2*KV0~KV1 + Q2*KV2 # pcp_rank1: Q2*KV0~KV1 + Q2*KV2
output_tail = self._attention_with_nomask_and_mask( output_tails, lse_tails = self._attention_with_nomask_and_mask(
q=torch.index_select(query, 0, q_tail_idx), q=torch.index_select(query, 0, q_tail_idx),
q_seqlens=attn_mask_seqlens, q_seqlens=attn_mask_seqlens,
k_nomask=torch.index_select(key, 0, kv_with_q_tail_nomask_idx), k_nomask=torch.index_select(key, 0, kv_with_q_tail_nomask_idx),
@@ -856,13 +927,17 @@ class AscendAttentionBackendImpl(AttentionImpl):
k_mask=torch.index_select(key, 0, kv_with_q_tail_mask_idx), k_mask=torch.index_select(key, 0, kv_with_q_tail_mask_idx),
v_mask=torch.index_select(value, 0, kv_with_q_tail_mask_idx), v_mask=torch.index_select(value, 0, kv_with_q_tail_mask_idx),
kv_seqlens_mask=attn_mask_seqlens, kv_seqlens_mask=attn_mask_seqlens,
mask=mask) mask=mask,
attn_metadata=attn_metadata)
# 3. Combine the output of the first half and second half.
q_full_idx = attn_metadata.prefill.pcp_metadata.q_full_idx q_full_idx = attn_metadata.prefill.pcp_metadata.q_full_idx
output = torch.index_select( output = torch.index_select(
torch.cat([output_head, output_tail], dim=0), 0, q_full_idx) torch.cat([output_heads, output_tails], dim=0), 0, q_full_idx)
return output attn_lse = None
if attn_metadata.prefill is not None and attn_metadata.prefill.chunked_context is not None:
attn_lse = torch.index_select(
torch.cat([lse_heads, lse_tails], dim=0), 0, q_full_idx)
return output, attn_lse
def _out_lse_reshape(self, attn_out: torch.Tensor, def _out_lse_reshape(self, attn_out: torch.Tensor,
attn_lse: torch.Tensor) -> torch.Tensor: attn_lse: torch.Tensor) -> torch.Tensor:
@@ -928,7 +1003,7 @@ class AscendAttentionBackendImpl(AttentionImpl):
'softmax_lse_flag': 'softmax_lse_flag':
True, True,
'block_table': 'block_table':
attn_metadata.block_tables, attn_metadata.decode_meta.block_tables,
'block_size': 'block_size':
self.key_cache.shape[1], self.key_cache.shape[1],
'actual_seq_lengths_kv': 'actual_seq_lengths_kv':
@@ -1029,8 +1104,24 @@ class AscendAttentionBackendImpl(AttentionImpl):
attn_out = self._npu_attention_update(attn_out_lse_list) attn_out = self._npu_attention_update(attn_out_lse_list)
return attn_out return attn_out
def _update_out_and_lse(self, out_list: torch.Tensor,
lse_list: torch.Tensor) -> torch.Tensor:
"""LSE_final = log(sum(exp(LSE_i))), O_final = sum(exp(LSE_i - LSE_final) * O_i)
Args:
out_list: shape = [N, batch_size, num_heads, head_size]
lse_list: shape = [N, batch_size, num_heads, 1]
Returns:
out_final: shape = [batch_size, num_heads, head_size]
lse_final: shape = [batch_size, num_heads, 1]
"""
lse_final = torch.logsumexp(lse_list, dim=0, keepdim=False)
out_final = torch.sum(torch.exp(lse_list - lse_final) * out_list,
dim=0)
return out_final, lse_final
def _forward_pcp_dcp(self, query: torch.Tensor, key: torch.Tensor, def _forward_pcp_dcp(self, query: torch.Tensor, key: torch.Tensor,
value: torch.Tensor, attn_metadata: AscendMetadata, value: torch.Tensor, kv_cache: Tuple[torch.Tensor],
attn_metadata: AscendMetadata,
output: torch.Tensor) -> torch.Tensor: output: torch.Tensor) -> torch.Tensor:
assert attn_metadata is not None assert attn_metadata is not None
has_decode = attn_metadata.num_decodes > 0 has_decode = attn_metadata.num_decodes > 0
@@ -1043,32 +1134,320 @@ class AscendAttentionBackendImpl(AttentionImpl):
decode_query, attn_metadata) decode_query, attn_metadata)
output[:num_decode_tokens] = output_decode output[:num_decode_tokens] = output_decode
if has_prefill: if has_prefill:
prefill_query = query[num_decode_tokens:] assert attn_metadata.prefill is not None
num_actual_tokens_pcp_padded = attn_metadata.num_actual_tokens_pcp_padded // self.pcp_size
prefill_query = query[
num_decode_tokens:num_actual_tokens_pcp_padded]
key = key[self.pcp_size * num_decode_tokens:] key = key[self.pcp_size * num_decode_tokens:]
value = value[self.pcp_size * num_decode_tokens:] value = value[self.pcp_size * num_decode_tokens:]
if self.pcp_size > 1: if self.pcp_size > 1:
output_prefill = self._forward_prefill_cp( # Scenario of Enabling PCP or PCP&DCP
attn_output_prefill, attn_lse_prefill = self._forward_prefill_cp(
prefill_query, key, value, attn_metadata) prefill_query, key, value, attn_metadata)
else: else:
max_prefill_seq_len = attn_metadata.seq_lens[ # Scenario of Enabling DCP Individually
attn_metadata.num_decode_tokens:].max().item() attn_output_prefill, attn_lse_prefill = torch.ops.npu.npu_fused_infer_attention_score(
if attn_metadata.attn_mask is not None: prefill_query,
attn_metadata.attn_mask = attn_metadata.attn_mask[: key,
max_prefill_seq_len, : value,
max_prefill_seq_len] num_heads=self.num_heads,
else: num_key_value_heads=self.num_kv_heads,
ValueError("Attn_metadata.attn_mask is required") input_layout="TND",
seq_lens_back = attn_metadata.seq_lens atten_mask=attn_metadata.attn_mask,
attn_metadata.seq_lens = attn_metadata.seq_lens[ scale=self.scale,
attn_metadata.num_decode_tokens:] sparse_mode=3,
output_prefill = self._forward_prefill_no_cache( antiquant_mode=0,
prefill_query, key, value, attn_metadata, antiquant_scale=None,
output[num_decode_tokens:], prefill_query.shape[0]) softmax_lse_flag=True,
attn_metadata.seq_lens = seq_lens_back actual_seq_lengths_kv=attn_metadata.prefill.
output[num_decode_tokens:output_prefill.shape[0] + actual_seq_lengths_q,
num_decode_tokens] = output_prefill actual_seq_lengths=attn_metadata.prefill.
actual_seq_lengths_q)
self._process_chunk_prefill(attn_output_prefill, attn_lse_prefill,
kv_cache, prefill_query, attn_metadata)
output[num_decode_tokens:attn_output_prefill.shape[0] +
num_decode_tokens] = attn_output_prefill
return output return output
def _process_chunk_prefill(self, current_attn_output_prefill,
current_attn_lse_prefill, kv_cache,
prefill_query, attn_metadata):
if attn_metadata.prefill is not None and attn_metadata.prefill.chunked_context is not None:
prefill_query_all = self._prefill_query_all_gather(
attn_metadata, prefill_query)
attn_output_full_chunk, attn_lse_full_chunk = self._compute_prefill_context(
prefill_query_all, kv_cache, attn_metadata)
self._update_chunk_attn_out_lse_with_current_attn_out_lse(
current_attn_output_prefill, current_attn_lse_prefill,
attn_output_full_chunk, attn_lse_full_chunk, prefill_query,
attn_metadata)
def _update_chunk_attn_out_lse_with_current_attn_out_lse(
self, current_attn_output_prefill, current_attn_lse_prefill,
attn_output_full_chunk, attn_lse_full_chunk, prefill_query,
attn_metadata):
if self.pcp_size > 1:
inverse_idx = attn_metadata.prefill.chunked_context.kv_inverse_idx_for_chunk
attn_output_full_chunk = torch.index_select(
attn_output_full_chunk, 0, inverse_idx)
attn_lse_full_chunk = torch.index_select(attn_lse_full_chunk, 0,
inverse_idx)
num_tokens = prefill_query.size(0)
attn_output_full_chunk = attn_output_full_chunk[
self.pcp_rank * num_tokens:(self.pcp_rank + 1) * num_tokens, :, :]
attn_lse_full_chunk = attn_lse_full_chunk[
self.pcp_rank * num_tokens:(self.pcp_rank + 1) * num_tokens, :, :]
assert attn_output_full_chunk.shape == current_attn_output_prefill.shape and attn_lse_full_chunk.shape == current_attn_lse_prefill.shape
seq_len = attn_metadata.query_lens.detach().clone()
filtered_indices = filter_chunked_req_indices(
seq_len,
attn_metadata.prefill.chunked_context.mask_for_non_zero_chunk)
attn_output_prefill_filtered = current_attn_output_prefill[
filtered_indices, :, :]
attn_lse_prefill_filtered = current_attn_lse_prefill[
filtered_indices, :, :]
attn_output_full_chunk = attn_output_full_chunk[filtered_indices, :, :]
attn_lse_full_chunk = attn_lse_full_chunk[filtered_indices, :, :]
attn_output_filtered = self._npu_attn_out_lse_update(
attn_lse_prefill_filtered, attn_lse_full_chunk,
attn_output_prefill_filtered, attn_output_full_chunk)
current_attn_output_prefill[
filtered_indices, :, :] = attn_output_filtered.to(
current_attn_output_prefill.dtype)
def _prefill_query_all_gather(self, attn_metadata, prefill_query):
prefill_query_all = get_pcp_group().all_gather(prefill_query.contiguous(),
0) \
if self.pcp_size > 1 else prefill_query
prefill_query_all = torch.index_select(prefill_query_all,
0,
attn_metadata.prefill.chunked_context.cp_kv_recover_idx_for_chunk) \
if self.pcp_size > 1 else prefill_query_all
return prefill_query_all
def _compute_prefill_context(self, query: torch.Tensor,
kv_cache: Tuple[torch.Tensor],
attn_metadata: AscendMetadata):
assert len(kv_cache) > 1
assert attn_metadata is not None
assert attn_metadata.prefill is not None
assert attn_metadata.prefill.chunked_context is not None
prefill_metadata = attn_metadata.prefill
local_chunked_kv_lens = attn_metadata.prefill.chunked_context.local_chunked_kv_lens
mask_for_non_zero_chunk = prefill_metadata.chunked_context.mask_for_non_zero_chunk
max_chunk_num = prefill_metadata.chunked_context.max_chunk_num
assert local_chunked_kv_lens is not None and mask_for_non_zero_chunk is not None and max_chunk_num > 0
iters = max_chunk_num
# Keep the causal mask; do not override to all-ones. [req_id][chunk_id][cp-rank][dcp_rank]
context_starts_rank = None
prefix_output_list = []
prefix_lse_list = []
for i in range(iters):
key, value, seq_lens_current_chunk_rank = self._load_kv_for_chunk(
attn_metadata, kv_cache, context_starts_rank, i,
local_chunked_kv_lens, prefill_metadata, query)
# 2. Attention computation
if seq_lens_current_chunk_rank is None or torch.all(
seq_lens_current_chunk_rank == 0).item():
prefix_output = torch.full(
(query.size(0), self.num_heads, self.head_size),
fill_value=0,
dtype=query.dtype,
device=query.device)
prefix_lse = torch.full((query.size(0), self.num_heads, 1),
fill_value=0,
dtype=torch.float32,
device=query.device)
else:
actual_seq_lengths_kv = torch.cumsum(
seq_lens_current_chunk_rank, dim=0).tolist()
prefix_output, prefix_lse = torch.ops.npu.npu_fused_infer_attention_score(
query,
key,
value,
num_heads=self.num_heads,
num_key_value_heads=self.num_kv_heads,
input_layout="TND", #
atten_mask=None,
scale=self.scale,
sparse_mode=0,
antiquant_mode=0,
antiquant_scale=None,
softmax_lse_flag=True,
actual_seq_lengths_kv=actual_seq_lengths_kv,
actual_seq_lengths=attn_metadata.prefill.chunked_context.
actual_chunk_seq_lengths)
prefix_output_list.append(prefix_output)
prefix_lse_list.append(prefix_lse)
# 3. update attn-out & lse
prefix_output, prefix_lse = self._update_attn_out_lse_in_chunks(
prefix_output_list, prefix_lse_list)
self._update_attn_out_lse_in_pcp(attn_metadata, prefix_output,
prefix_lse)
return prefix_output, prefix_lse
def _update_attn_out_lse_in_chunks(self, prefix_output_list,
prefix_lse_list):
# update output and lse
if len(prefix_output_list) > 1:
prefix_output, prefix_lse = self._update_out_and_lse(
torch.stack(prefix_output_list, dim=0),
torch.stack(prefix_lse_list, dim=0))
else:
prefix_output = prefix_output_list[0]
prefix_lse = prefix_lse_list[0]
return prefix_output, prefix_lse
def _update_attn_out_lse_in_pcp(self, attn_metadata, prefix_output,
prefix_lse):
# CP dimension all_gather and fusion
if self.pcp_size > 1:
# filter non-zero chunk part of prefix_output
current_seq_lens = attn_metadata.query_lens.detach().clone()
current_seq_lens.mul_(self.pcp_size) # q_full
current_seq_lens_cpu = current_seq_lens.cpu()
filtered_indices = filter_chunked_req_indices(
current_seq_lens_cpu,
attn_metadata.prefill.chunked_context.mask_for_non_zero_chunk)
prefix_output_filtered = prefix_output[filtered_indices, :, :]
prefix_lse_filtered = prefix_lse[filtered_indices, :, :]
out_lse_local = torch.cat(
[prefix_output_filtered, prefix_lse_filtered], dim=-1)
attn_out_lse_list = [
torch.empty_like(out_lse_local) for _ in range(self.pcp_size)
]
dist.all_gather(attn_out_lse_list,
out_lse_local,
group=self.pcp_group)
attn_out_lse_allgather = torch.stack(
attn_out_lse_list,
dim=0) # [pcp, batch_size, num_heads, head_size+1]
attn_out_allgather, attn_lse_allgather = torch.split(
attn_out_lse_allgather, [self.head_size, 1], dim=-1)
prefix_output_filtered, prefix_lse_filtered = self._update_out_and_lse(
attn_out_allgather, attn_lse_allgather)
prefix_output[filtered_indices, :, :] = prefix_output_filtered.to(
prefix_output.dtype)
prefix_lse[filtered_indices, :, :] = prefix_lse_filtered.to(
prefix_lse.dtype)
def _load_kv_for_chunk(self, attn_metadata, kv_cache, context_starts_rank,
i, local_chunked_kv_lens, prefill_metadata, query):
cache_key = kv_cache[0]
cache_value = kv_cache[1]
num_heads = cache_key.size(2)
head_size = kv_cache[0].size(-1)
# 1. Load current query's history key-value
seq_lens_current_chunk = attn_metadata.query_lens.detach().clone()
num_requests = len(seq_lens_current_chunk)
# Before dealing with a new chunk, set to zero, and accumulate the start positions as chunk prefill step increases
context_starts_rank = torch.zeros(
num_requests, dtype=torch.int32, device=query.device
) if context_starts_rank is None else context_starts_rank
# Calculate tokens each rank should process per request
seq_lens_current_chunk_rank = torch.zeros_like(seq_lens_current_chunk,
dtype=torch.int32,
device=query.device)
total_toks = 0
for req_idx in range(num_requests):
if i >= len(local_chunked_kv_lens[req_idx]):
continue
n_computed_acc = local_chunked_kv_lens[req_idx][i]
total_toks += n_computed_acc[self.pcp_rank][self.dcp_rank]
seq_lens_current_chunk_rank[req_idx] = n_computed_acc[
self.pcp_rank][self.dcp_rank]
if total_toks > 0:
key = torch.empty(total_toks,
num_heads,
head_size,
dtype=query.dtype,
device=query.device)
value = torch.empty(total_toks,
num_heads,
head_size,
dtype=query.dtype,
device=query.device)
torch_npu.atb.npu_paged_cache_load(
cache_key,
cache_value,
attn_metadata.prefill.block_tables,
seq_lens_current_chunk_rank.to(query.device),
seq_starts=
context_starts_rank, # slot offsets of current chunk in current iteration
key=key,
value=value,
)
else:
# If current rank has no tokens to process, create empty tensors
key = torch.empty(0,
self.num_heads,
self.head_size,
dtype=query.dtype,
device=query.device)
value = torch.empty(0,
self.num_heads,
self.head_size,
dtype=query.dtype,
device=query.device)
seq_lens_current_chunk_rank = torch.zeros(
(len(seq_lens_current_chunk), ),
dtype=torch.int32,
device=query.device)
for req_idx in range(num_requests):
# Before dealing with a new chunk, set to zero, and accumulate the start positions as chunk prefill step increases
if i >= len(local_chunked_kv_lens[req_idx]):
continue
context_starts_rank[req_idx] += local_chunked_kv_lens[req_idx][i][
self.pcp_rank][self.dcp_rank]
if self.dcp_size > 1:
req_dcp_sizes = extract_req_dcp_by_chunk_pcp(
local_chunked_kv_lens, i, self.dcp_size, self.pcp_rank)
assert len(req_dcp_sizes) == num_requests and all(
len(dcp_arr) == self.dcp_size for dcp_arr in req_dcp_sizes)
total_toks = np.sum(np.array(req_dcp_sizes))
kv_local = torch.cat([key, value], dim=-1)
head_dim = kv_local.size(-1)
kv_full = torch.empty((total_toks, num_heads, head_dim),
device=query.device,
dtype=query.dtype)
kv_full_list = [None for _ in range(self.dcp_size)]
dist.all_gather_object(kv_full_list,
kv_local,
group=self.dcp_group)
kv_full_list = [
kv for kv in kv_full_list if kv is not None and kv.numel() > 0
]
if len(kv_full_list) > 0:
kv_full = torch.cat(kv_full_list, dim=0)
key, value = kv_full.split([head_size, head_size], dim=-1)
if total_toks == 0:
return key, value, None
seq_lens_current_chunk_rank = torch.tensor(
np.sum(np.array(req_dcp_sizes), axis=1),
dtype=torch.int32,
device=query.device) # [reqs]
return key, value, seq_lens_current_chunk_rank
def forward( def forward(
self, self,
layer: AttentionLayer, layer: AttentionLayer,
@@ -1162,7 +1541,7 @@ class AscendAttentionBackendImpl(AttentionImpl):
if self.pcp_size * self.dcp_size > 1: if self.pcp_size * self.dcp_size > 1:
intermediate_output = self._forward_pcp_dcp( intermediate_output = self._forward_pcp_dcp(
query, key, value, attn_metadata, output) query, key, value, kv_cache, attn_metadata, output)
elif attn_type == AttentionType.ENCODER_ONLY: elif attn_type == AttentionType.ENCODER_ONLY:
# TODO(zzzwwjj): Deal with this `cum_seq_len` more elegantly. # TODO(zzzwwjj): Deal with this `cum_seq_len` more elegantly.
cum_seq_len = attn_metadata.query_start_loc[1:].tolist() cum_seq_len = attn_metadata.query_start_loc[1:].tolist()
@@ -1185,7 +1564,7 @@ class AscendAttentionBackendImpl(AttentionImpl):
intermediate_output = self._forward_prefill_no_cache( intermediate_output = self._forward_prefill_no_cache(
query, key, value, attn_metadata, output, num_tokens) query, key, value, attn_metadata, output, num_tokens)
elif attn_metadata.attn_state == \ elif attn_metadata.attn_state == \
AscendAttentionState.PrefillCacheHit: AscendAttentionState.PrefillCacheHit:
intermediate_output = self._forward_prefill_cache_hit( intermediate_output = self._forward_prefill_cache_hit(
query, attn_metadata, output) query, attn_metadata, output)
elif attn_metadata.attn_state == AscendAttentionState.DecodeOnly: elif attn_metadata.attn_state == AscendAttentionState.DecodeOnly:

515
vllm_ascend/attention/mla_v1.py
View File

@@ -5,6 +5,7 @@ from typing import (TYPE_CHECKING, ClassVar, List, NamedTuple, Optional, Tuple,
import numpy as np import numpy as np
import torch import torch
import torch.distributed as dist import torch.distributed as dist
import torch.nn.functional as F
import torch_npu import torch_npu
from torch import nn from torch import nn
from vllm.attention.backends.abstract import (AttentionBackend, from vllm.attention.backends.abstract import (AttentionBackend,
@@ -27,11 +28,14 @@ from vllm.v1.attention.backends.utils import AttentionCGSupport
from vllm_ascend import envs from vllm_ascend import envs
from vllm_ascend.ascend_config import get_ascend_config from vllm_ascend.ascend_config import get_ascend_config
from vllm_ascend.attention.attention_v1 import AscendAttentionState from vllm_ascend.attention.attention_v1 import AscendAttentionState
from vllm_ascend.attention.utils import (AscendCommonAttentionMetadata,
maybe_save_kv_layer_to_connector, # isort: off
split_decodes_and_prefills, from vllm_ascend.attention.utils import (
trans_rope_weight, transdata, AscendCommonAttentionMetadata, extract_req_dcp_by_chunk_pcp,
wait_for_kv_layer_from_connector) filter_chunked_req_indices, maybe_save_kv_layer_to_connector,
split_decodes_and_prefills, trans_rope_weight, transdata,
wait_for_kv_layer_from_connector)
# isort: on
from vllm_ascend.compilation.acl_graph import (get_graph_params, from vllm_ascend.compilation.acl_graph import (get_graph_params,
update_graph_params_workspaces) update_graph_params_workspaces)
from vllm_ascend.ops.weight_prefetch import maybe_npu_prefetch from vllm_ascend.ops.weight_prefetch import maybe_npu_prefetch
@@ -111,6 +115,10 @@ class AscendMLAPrefillMetadata:
workspace: torch.Tensor workspace: torch.Tensor
chunk_seq_lens: torch.Tensor chunk_seq_lens: torch.Tensor
chunk_seq_lens_npu: torch.Tensor chunk_seq_lens_npu: torch.Tensor
mask_for_non_zero_chunk: Optional[list[bool]] = None
max_chunk_num: int = 0
local_chunked_kv_lens: Optional[list[Optional[list[Optional[list[
Optional[list[int]]]]]]]] = None
attn_mask: torch.Tensor attn_mask: torch.Tensor
query_lens: torch.Tensor query_lens: torch.Tensor
@@ -125,6 +133,7 @@ class AscendMLAPrefillMetadata:
sin: torch.Tensor = None sin: torch.Tensor = None
cos: torch.Tensor = None cos: torch.Tensor = None
pcp_metadata: Optional[AscendPCPMetadata] = None pcp_metadata: Optional[AscendPCPMetadata] = None
cp_kv_recover_idx_for_chunk: Optional[list[int]] = None
@dataclass @dataclass
@@ -347,6 +356,10 @@ class AscendMLAMetadataBuilder:
num_actual_tokens_pcp_padded = long_seq_metadata.num_actual_tokens_pcp_padded if long_seq_metadata else None num_actual_tokens_pcp_padded = long_seq_metadata.num_actual_tokens_pcp_padded if long_seq_metadata else None
num_computed_tokens_of_pcp_dcp = long_seq_metadata.num_computed_tokens_of_pcp_dcp if long_seq_metadata else None num_computed_tokens_of_pcp_dcp = long_seq_metadata.num_computed_tokens_of_pcp_dcp if long_seq_metadata else None
cp_kv_recover_idx_for_chunk = long_seq_metadata.cp_kv_recover_idx_for_chunk if long_seq_metadata else None
local_chunked_kv_lens = long_seq_metadata.local_chunked_kv_lens if long_seq_metadata else None
mask_for_non_zero_chunk = long_seq_metadata.mask_for_non_zero_chunk if long_seq_metadata else None
max_chunk_num = long_seq_metadata.max_chunk_num if long_seq_metadata else 0
num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = \ num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = \
split_decodes_and_prefills(common_attn_metadata, decode_threshold=self.decode_threshold) split_decodes_and_prefills(common_attn_metadata, decode_threshold=self.decode_threshold)
@@ -359,14 +372,15 @@ class AscendMLAMetadataBuilder:
device = self.device device = self.device
block_table = (common_attn_metadata.block_table_tensor[:num_reqs]) block_table = (common_attn_metadata.block_table_tensor[:num_reqs])
input_positions = common_attn_metadata.positions[:
num_actual_tokens].long(
)
if num_actual_tokens_pcp_padded is None: if num_actual_tokens_pcp_padded is None:
num_actual_tokens_pcp_padded = num_actual_tokens num_actual_tokens_pcp_padded = num_actual_tokens
slot_mapping = common_attn_metadata.slot_mapping[: slot_mapping = common_attn_metadata.slot_mapping[:
num_actual_tokens_pcp_padded] num_actual_tokens_pcp_padded]
input_positions = common_attn_metadata.positions[:
num_actual_tokens_pcp_padded].long(
)
if self.cos_cache is None: if self.cos_cache is None:
self.cos_cache = model.model.layers[ self.cos_cache = model.model.layers[
@@ -408,7 +422,8 @@ class AscendMLAMetadataBuilder:
tail_attn_nomask_seqlens=common_long_seq_metadata. tail_attn_nomask_seqlens=common_long_seq_metadata.
tail_attn_nomask_seqlens, tail_attn_nomask_seqlens,
q_full_idx=common_long_seq_metadata.q_full_idx, q_full_idx=common_long_seq_metadata.q_full_idx,
pcp_prefill_mask=common_long_seq_metadata.pcp_prefill_mask, pcp_prefill_mask=common_long_seq_metadata.pcp_prefill_mask
if long_seq_metadata else None,
pcp_allgather_restore_idx=long_seq_metadata. pcp_allgather_restore_idx=long_seq_metadata.
pcp_allgather_restore_idx if long_seq_metadata else None) pcp_allgather_restore_idx if long_seq_metadata else None)
@@ -452,6 +467,9 @@ class AscendMLAMetadataBuilder:
chunk_seq_lens=chunk_seq_lens, chunk_seq_lens=chunk_seq_lens,
chunk_seq_lens_npu=chunk_seq_lens.npu(), chunk_seq_lens_npu=chunk_seq_lens.npu(),
workspace=self.chunked_prefill_workspace, workspace=self.chunked_prefill_workspace,
local_chunked_kv_lens=local_chunked_kv_lens,
mask_for_non_zero_chunk=mask_for_non_zero_chunk,
max_chunk_num=max_chunk_num,
) )
prefill_input_positions = input_positions[tokens_start:] prefill_input_positions = input_positions[tokens_start:]
cos = self.cos_cache[ cos = self.cos_cache[
@@ -474,7 +492,7 @@ class AscendMLAMetadataBuilder:
sin=sin, sin=sin,
cos=cos, cos=cos,
pcp_metadata=pcp_metadata, pcp_metadata=pcp_metadata,
) cp_kv_recover_idx_for_chunk=cp_kv_recover_idx_for_chunk)
decode_metadata = None decode_metadata = None
if num_decodes > 0: if num_decodes > 0:
@@ -887,8 +905,26 @@ class AscendMLAImpl(MLAAttentionImpl):
prefill_metadata = attn_metadata.prefill prefill_metadata = attn_metadata.prefill
if prefill_metadata is None or prefill_metadata.chunked_context is None: if prefill_metadata is None or prefill_metadata.chunked_context is None:
return prefix_output, prefix_lse return prefix_output, prefix_lse
local_chunked_kv_lens = prefill_metadata.chunked_context.local_chunked_kv_lens
mask_for_non_zero_chunk = prefill_metadata.chunked_context.mask_for_non_zero_chunk
max_chunk_num = prefill_metadata.chunked_context.max_chunk_num
if self.pcp_size * self.dcp_size > 1:
assert local_chunked_kv_lens is not None and mask_for_non_zero_chunk is not None and max_chunk_num > 0
if self.pcp_size > 1:
prefix_output = torch.zeros(q_nope.shape[0],
self.num_heads,
self.v_head_dim,
dtype=q_nope.dtype,
device=q_nope.device)
prefix_lse = torch.zeros(self.num_heads,
q_pe.shape[0],
dtype=torch.float32,
device=q_pe.device)
iters = len(prefill_metadata.chunked_context.seq_tot) iters = len(prefill_metadata.chunked_context.seq_tot)
if self.pcp_size * self.dcp_size > 1:
iters = max_chunk_num
current_seq_len = torch.tensor(prefill_metadata.query_lens, current_seq_len = torch.tensor(prefill_metadata.query_lens,
dtype=torch.int32) dtype=torch.int32)
@@ -896,60 +932,305 @@ class AscendMLAImpl(MLAAttentionImpl):
cache_k_pe = kv_c_and_k_pe_cache[1] cache_k_pe = kv_c_and_k_pe_cache[1]
num_heads = cache_k_pe.size(2) num_heads = cache_k_pe.size(2)
latent_kv_dim = kv_c_and_k_pe_cache[0].size(-1) latent_kv_dim = kv_c_and_k_pe_cache[0].size(-1)
# token -> request mapping for building per-token masks when CP>1
seq_len1 = torch.tensor(prefill_metadata.query_lens,
dtype=torch.int32,
device=q_nope.device)
seq_len1.mul_(
self.pcp_size) # q_full: already padded, divisible by cp_size
# Select mask: prefer CP prefill mask from metadata; fallback to cached prefill_mask; create if needed.
mask_local = None
if attn_metadata is not None and attn_metadata.prefill is not None and \
attn_metadata.prefill.pcp_metadata is not None and attn_metadata.prefill.pcp_metadata.pcp_prefill_mask is not None:
mask_local = attn_metadata.prefill.pcp_metadata.pcp_prefill_mask
else:
mask_local = self.prefill_mask
if mask_local is None:
mask_local = torch.triu(
torch.ones(512,
512,
device=q_nope.device,
dtype=q_nope.dtype), 1)
self.prefill_mask = mask_local
# Keep the causal mask; do not override to all-ones.
context_starts_rank = None
for i in range(iters): for i in range(iters):
toks = prefill_metadata.chunked_context.seq_tot[i] if self.pcp_size * self.dcp_size > 1:
## DCP mode: each rank processes its own (cp,dcp) historical context slice per request dimension
num_requests = len(seq_len1)
assert num_requests == len(local_chunked_kv_lens)
# Before dealing with a new chunk, set to zero, and accumulate the start positions as chunk prefill step increases
context_starts_rank = torch.zeros(
num_requests, dtype=torch.int32, device=q_nope.device
) if context_starts_rank is None else context_starts_rank
context_seq_len = prefill_metadata.chunked_context.chunk_seq_lens[ ## Calculate tokens each rank should process per request
i] seq_len2_rank = torch.zeros_like(seq_len1, dtype=torch.int32)
context_seq_len_npu = prefill_metadata.chunked_context.chunk_seq_lens_npu[ total_toks = 0
i]
seq_len = torch.stack([current_seq_len, context_seq_len])
kv_c_normed = torch.empty(toks,
num_heads,
latent_kv_dim,
dtype=q_nope.dtype,
device=q_nope.device)
k_pe = torch.empty(toks,
num_heads,
rope_dim,
dtype=q_nope.dtype,
device=q_nope.device)
torch_npu.atb.npu_paged_cache_load( for req_idx in range(num_requests):
cache_kv_c, if i >= len(local_chunked_kv_lens[req_idx]):
cache_k_pe, continue
prefill_metadata.block_table, n_computed_acc = local_chunked_kv_lens[req_idx][i]
context_seq_len_npu, total_toks += n_computed_acc[self.pcp_rank][self.dcp_rank]
seq_starts=prefill_metadata.chunked_context.starts[i], seq_len2_rank[req_idx] = n_computed_acc[self.pcp_rank][
key=kv_c_normed, self.dcp_rank]
value=k_pe,
) if total_toks > 0:
kv_c_normed = torch.empty(total_toks,
num_heads,
latent_kv_dim,
dtype=q_nope.dtype,
device=q_nope.device)
k_pe = torch.empty(total_toks,
num_heads,
rope_dim,
dtype=q_nope.dtype,
device=q_nope.device)
torch_npu.atb.npu_paged_cache_load(
cache_kv_c,
cache_k_pe,
prefill_metadata.block_table,
seq_len2_rank.to(q_nope.device),
seq_starts=
context_starts_rank, # slot offsets of current chunk in current iteration
key=kv_c_normed,
value=k_pe,
)
seq_len2 = seq_len2_rank.to(q_nope.device)
else:
# If current rank has no tokens to process, create empty tensors
kv_c_normed = torch.empty(0,
num_heads,
latent_kv_dim,
dtype=q_nope.dtype,
device=q_nope.device)
k_pe = torch.empty(0,
num_heads,
rope_dim,
dtype=q_nope.dtype,
device=q_nope.device)
seq_len2 = torch.zeros((len(seq_len1), ),
dtype=torch.int32,
device=q_nope.device)
seq_len = torch.stack([seq_len1.cpu(), seq_len2.cpu()])
for req_idx in range(num_requests):
# Before dealing with a new chunk, set to zero, and accumulate the start positions as chunk prefill step increases
if i >= len(local_chunked_kv_lens[req_idx]):
continue
context_starts_rank[req_idx] += local_chunked_kv_lens[
req_idx][i][self.pcp_rank][self.dcp_rank]
else:
# Original logic: ChunkPrefill-only mode
toks = prefill_metadata.chunked_context.seq_tot[i]
context_seq_len = prefill_metadata.chunked_context.chunk_seq_lens[
i]
context_seq_len_npu = prefill_metadata.chunked_context.chunk_seq_lens_npu[
i]
seq_len = torch.stack([current_seq_len, context_seq_len])
kv_c_normed = torch.empty(toks,
num_heads,
latent_kv_dim,
dtype=q_nope.dtype,
device=q_nope.device)
k_pe = torch.empty(toks,
num_heads,
rope_dim,
dtype=q_nope.dtype,
device=q_nope.device)
torch_npu.atb.npu_paged_cache_load(
cache_kv_c,
cache_k_pe,
prefill_metadata.block_table,
context_seq_len_npu,
seq_starts=prefill_metadata.chunked_context.starts[i],
key=kv_c_normed,
value=k_pe,
)
kv_c_normed = kv_c_normed.squeeze() kv_c_normed = kv_c_normed.squeeze()
kv_nope = self.kv_b_proj(kv_c_normed)[0].view( \ if self.dcp_size > 1:
-1, self.num_heads, self.qk_nope_head_dim + self.v_head_dim) # DCP mode: first all_gather within DCP group, let each rank in CP group share complete sequence blocks
k_nope, v = kv_nope\ # Step 1: DCP all_gather latent
.split([self.qk_nope_head_dim, self.v_head_dim], dim=-1) kv_c_k_pe_local = torch.cat(
k_pe = k_pe.expand((*k_nope.shape[:-1], -1)) [kv_c_normed, k_pe.squeeze()],
torch_npu.atb.npu_ring_mla( dim=-1) # [local_toks, latent_dim + rope_dim]
q_nope=q_nope,
q_rope=q_pe, # Step 2: use all_gather_into_tensor_uneven (gather + cat)
k_nope=k_nope, req_dcp_sizes = extract_req_dcp_by_chunk_pcp(
k_rope=k_pe, local_chunked_kv_lens, i, self.dcp_size, self.pcp_rank
value=v, ) # need to know num tokens of each rank in dcp group before all_gather # [reqs, dcp]
mask=self.prefill_mask, assert len(req_dcp_sizes) == num_requests and all(
seqlen=seq_len, len(dcp_arr) == self.dcp_size for dcp_arr in req_dcp_sizes)
head_num=self.num_heads, total_toks = np.sum(np.array(req_dcp_sizes))
kv_head_num=self.num_heads, latent_rope_dim = kv_c_k_pe_local.size(-1)
pre_out=prefix_output, kv_c_k_pe_full = torch.empty((total_toks, latent_rope_dim),
prev_lse=prefix_lse, device=kv_c_k_pe_local.device,
qk_scale=self.scale, dtype=kv_c_k_pe_local.dtype)
kernel_type="kernel_type_high_precision",
mask_type="no_mask", kv_c_k_pe_full_list = [None for _ in range(self.dcp_size)]
input_layout="type_bsnd", dist.all_gather_object(kv_c_k_pe_full_list,
calc_type="calc_type_default", kv_c_k_pe_local,
output=prefix_output, group=self.dcp_group)
softmax_lse=prefix_lse) kv_c_k_pe_full_list = [
kv_c_k_pe for kv_c_k_pe in kv_c_k_pe_full_list
if kv_c_k_pe is not None and kv_c_k_pe.numel() > 0
]
if len(kv_c_k_pe_full_list) > 0:
kv_c_k_pe_full = torch.cat(kv_c_k_pe_full_list, dim=0)
if len(kv_c_k_pe_full.shape) == 1:
assert total_toks == 1
kv_c_k_pe_full = kv_c_k_pe_full.unsqueeze(0)
assert kv_c_k_pe_full.shape[
0] == total_toks and kv_c_k_pe_full.shape[
1] == latent_rope_dim
kv_c_normed_full, k_pe_full = torch.split(
kv_c_k_pe_full, [latent_kv_dim, rope_dim], dim=-1)
# Step 3: process complete sequence with TP projection to get current rank's head slice
# Case that no kv_cache has been stored on this CP rank(after dcp all_gather), no need to do following computation.
if total_toks == 0:
continue
kv_nope = self.kv_b_proj(kv_c_normed_full)[0].view(
-1, self.num_heads,
self.qk_nope_head_dim + self.v_head_dim)
k_nope, v = kv_nope.split(
[self.qk_nope_head_dim, self.v_head_dim], dim=-1)
k_pe = k_pe_full.unsqueeze(1).expand((*k_nope.shape[:-1], -1))
seq_len2 = torch.tensor(np.sum(np.array(req_dcp_sizes),
axis=1),
dtype=torch.int32,
device=q_nope.device) # [reqs]
seq_len = torch.stack([seq_len1.cpu(), seq_len2.cpu()])
else:
# Non-DCP mode: use TP-split projection
kv_nope = self.kv_b_proj(kv_c_normed)[0].view(
-1, self.num_heads,
self.qk_nope_head_dim + self.v_head_dim)
k_nope, v = kv_nope.split(
[self.qk_nope_head_dim, self.v_head_dim], dim=-1)
k_pe = k_pe.expand((*k_nope.shape[:-1], -1))
if self.pcp_size > 1:
# Case that no kv_cache has been stored on this CP rank, no need to do following computation.
if torch.all(seq_len2 == 0).item():
continue
# PCP mode: first compute this rank's contribution to the chunk
if i == 0:
torch_npu.atb.npu_ring_mla(
q_nope=q_nope,
q_rope=q_pe,
k_nope=k_nope,
k_rope=k_pe,
value=v,
mask=mask_local,
seqlen=seq_len,
head_num=self.num_heads,
kv_head_num=self.num_heads,
pre_out=None,
prev_lse=None,
qk_scale=self.scale,
kernel_type="kernel_type_high_precision",
mask_type="no_mask",
input_layout="type_bsnd",
calc_type="calc_type_first_ring",
output=prefix_output,
softmax_lse=prefix_lse)
continue
torch_npu.atb.npu_ring_mla(
q_nope=q_nope,
q_rope=q_pe,
k_nope=k_nope,
k_rope=k_pe,
value=v,
mask=mask_local,
seqlen=seq_len,
head_num=self.num_heads,
kv_head_num=self.num_heads,
pre_out=prefix_output,
prev_lse=prefix_lse,
qk_scale=self.scale,
kernel_type="kernel_type_high_precision",
mask_type="no_mask",
input_layout="type_bsnd",
calc_type="calc_type_default",
output=prefix_output,
softmax_lse=prefix_lse)
else:
assert not torch.all(context_seq_len == 0).item()
# compute this chunk block then update prefix tensors to keep shapes consistent
torch_npu.atb.npu_ring_mla(
q_nope=q_nope,
q_rope=q_pe,
k_nope=k_nope,
k_rope=k_pe,
value=v,
mask=mask_local,
seqlen=seq_len,
head_num=self.num_heads,
kv_head_num=self.num_heads,
pre_out=prefix_output,
prev_lse=prefix_lse,
qk_scale=self.scale,
kernel_type="kernel_type_high_precision",
mask_type="no_mask",
input_layout="type_bsnd",
calc_type="calc_type_default",
output=prefix_output,
softmax_lse=prefix_lse)
# CP dimension all_gather and fusion
if self.pcp_size > 1:
# filter non-zero chunk part of prefix_output
seq_len1_cpu = seq_len1.cpu()
filtered_indices = filter_chunked_req_indices(
seq_len1_cpu, mask_for_non_zero_chunk)
prefix_output_filtered = prefix_output[filtered_indices, :, :]
prefix_lse_filtered = prefix_lse[:, filtered_indices]
# normalize prefix LSE to [bs, heads, 1] for stable updates
prefix_lse_filtered_bt = prefix_lse_filtered.permute(
1, 0).unsqueeze(-1).contiguous(
) if prefix_lse_filtered is not None else None
out_lse_local = torch.cat(
[prefix_output_filtered, prefix_lse_filtered_bt], dim=-1)
out_lse_list = [
torch.empty_like(out_lse_local) for _ in range(self.pcp_size)
]
dist.all_gather(out_lse_list, out_lse_local, group=self.pcp_group)
prefix_output_filtered = None
prefix_lse_filtered_bt = None
for r in range(self.pcp_size):
out_lse_r = out_lse_list[r]
if torch.all(out_lse_r == 0).item():
continue
out_r, lse_r = torch.split(out_lse_r, [self.v_head_dim, 1],
dim=-1)
token_mask = torch.ones([out_r.size(0)],
dtype=torch.uint8,
device=out_r.device)
prefix_output_filtered, prefix_lse_filtered_bt = self._update_out_and_lse(
prefix_output_filtered, prefix_lse_filtered_bt, out_r,
lse_r, token_mask)
# convert lse back to [heads, bs]
assert prefix_output_filtered is not None and prefix_lse_filtered_bt is not None
prefix_lse_filtered = prefix_lse_filtered_bt.squeeze(-1).permute(
1, 0).contiguous()
prefix_output[filtered_indices, :, :] = prefix_output_filtered.to(
prefix_output.dtype)
prefix_lse[:, filtered_indices] = prefix_lse_filtered.to(
prefix_lse.dtype)
return prefix_output, prefix_lse return prefix_output, prefix_lse
def _forward_prefill( def _forward_prefill(
@@ -1516,7 +1797,7 @@ class AscendMLAImpl(MLAAttentionImpl):
tail_attn_nomask_seqlens = attn_metadata.prefill.pcp_metadata.tail_attn_nomask_seqlens tail_attn_nomask_seqlens = attn_metadata.prefill.pcp_metadata.tail_attn_nomask_seqlens
mask = attn_metadata.prefill.pcp_metadata.pcp_prefill_mask mask = attn_metadata.prefill.pcp_metadata.pcp_prefill_mask
output_head = self._attention_with_mask_and_nomask( output_head, head_lse = self._attention_with_mask_and_nomask(
q_nope=torch.index_select(q_nope, 0, q_head_idx), q_nope=torch.index_select(q_nope, 0, q_head_idx),
q_pe=torch.index_select(q_pe, 0, q_head_idx), q_pe=torch.index_select(q_pe, 0, q_head_idx),
k_nope=k_nope, k_nope=k_nope,
@@ -1528,7 +1809,7 @@ class AscendMLAImpl(MLAAttentionImpl):
attn_nomask_seqlens=head_attn_nomask_seqlens, attn_nomask_seqlens=head_attn_nomask_seqlens,
mask=mask) mask=mask)
output_tail = self._attention_with_mask_and_nomask( output_tail, tail_lse = self._attention_with_mask_and_nomask(
q_nope=torch.index_select(q_nope, 0, q_tail_idx), q_nope=torch.index_select(q_nope, 0, q_tail_idx),
q_pe=torch.index_select(q_pe, 0, q_tail_idx), q_pe=torch.index_select(q_pe, 0, q_tail_idx),
k_nope=k_nope, k_nope=k_nope,
@@ -1544,7 +1825,83 @@ class AscendMLAImpl(MLAAttentionImpl):
output = torch.index_select( output = torch.index_select(
torch.cat([output_head, output_tail], dim=0), 0, q_full_idx) torch.cat([output_head, output_tail], dim=0), 0, q_full_idx)
output = output.reshape([num_tokens, self.num_heads * self.v_head_dim]) # Synchronize and reorder LSE for subsequent chunked context accumulation
attn_lse = torch.cat([head_lse, tail_lse], dim=1)
attn_lse = attn_lse[:, q_full_idx]
# Post-processing: keep [tokens, H, V] shape and perform chunked context accumulation if needed
if attn_metadata.prefill is not None and \
attn_metadata.prefill.chunked_context is not None:
# q all_gather
q_nope_full = get_pcp_group().all_gather(q_nope.contiguous(), 0)
q_pe_full = get_pcp_group().all_gather(q_pe.contiguous(), 0)
q_nope_full = torch.index_select(
q_nope_full, 0,
attn_metadata.prefill.cp_kv_recover_idx_for_chunk)
q_pe_full = torch.index_select(
q_pe_full, 0,
attn_metadata.prefill.cp_kv_recover_idx_for_chunk)
attn_output_pre = output.view(num_tokens, self.num_heads,
self.v_head_dim)
attn_output_pre_full, attn_lse_full = self._compute_prefill_context(
q_nope_full,
q_pe_full,
kv_c_and_k_pe_cache,
self.qk_rope_head_dim,
attn_metadata,
None,
None,
)
# reorder back && extract output + lse result of each cp rank
inverse_idx = torch.argsort(
attn_metadata.prefill.cp_kv_recover_idx_for_chunk)
attn_output_pre_full = torch.index_select(attn_output_pre_full, 0,
inverse_idx)
attn_lse_full = torch.index_select(attn_lse_full, 1, inverse_idx)
attn_output_pre_new = attn_output_pre_full[
self.pcp_rank * num_tokens:(self.pcp_rank + 1) *
num_tokens, :, :]
attn_lse_new = attn_lse_full[:, self.pcp_rank *
num_tokens:(self.pcp_rank + 1) *
num_tokens]
# update(output_origin, output_new)
assert attn_output_pre_new.shape == attn_output_pre.shape and attn_lse_new.shape == attn_lse.shape
seq_len = torch.tensor(attn_metadata.prefill.query_lens,
dtype=torch.int32)
mask_for_non_zero_chunk = attn_metadata.prefill.chunked_context.mask_for_non_zero_chunk
filtered_indices = filter_chunked_req_indices(
seq_len, mask_for_non_zero_chunk)
attn_output_pre_filtered = attn_output_pre[filtered_indices, :, :]
attn_lse_filtered = attn_lse[:, filtered_indices]
attn_output_pre_new = attn_output_pre_new[filtered_indices, :, :]
attn_lse_new = attn_lse_new[:, filtered_indices]
# normalize prefix LSE to [bs, heads, 1] for stable updates
attn_lse_filtered = attn_lse_filtered.permute(1, 0).unsqueeze(-1)
attn_lse_new = attn_lse_new.permute(1, 0).unsqueeze(-1)
token_mask = torch.ones([attn_lse_new.size(0)],
dtype=torch.uint8,
device=attn_lse_new.device)
attn_output_pre_filtered, attn_lse_filtered = self._update_out_and_lse(
attn_output_pre_filtered, attn_lse_filtered,
attn_output_pre_new, attn_lse_new, token_mask)
# convert lse back to [heads, bs]
attn_lse_filtered = attn_lse_filtered.squeeze(-1).permute(
1, 0).contiguous()
attn_output_pre[
filtered_indices, :, :] = attn_output_pre_filtered.to(
attn_output_pre.dtype)
attn_lse[:,
filtered_indices] = attn_lse_filtered.to(attn_lse.dtype)
attn_output_pre = attn_output_pre.to(q_nope.dtype)
output = attn_output_pre.reshape(
[num_tokens, self.num_heads * self.v_head_dim])
else:
output = output.reshape(
[num_tokens, self.num_heads * self.v_head_dim])
return output return output
@@ -1588,7 +1945,7 @@ class AscendMLAImpl(MLAAttentionImpl):
# nomask # nomask
if kv_nomask_idx.shape[0] == 0: if kv_nomask_idx.shape[0] == 0:
return attn_output return attn_output, attn_lse
k_nope_nomask = torch.index_select(k_nope, 0, kv_nomask_idx) k_nope_nomask = torch.index_select(k_nope, 0, kv_nomask_idx)
value_nomask = torch.index_select(value, 0, kv_nomask_idx) value_nomask = torch.index_select(value, 0, kv_nomask_idx)
@@ -1611,7 +1968,7 @@ class AscendMLAImpl(MLAAttentionImpl):
calc_type="calc_type_default", calc_type="calc_type_default",
output=attn_output, output=attn_output,
softmax_lse=attn_lse) softmax_lse=attn_lse)
return attn_output return attn_output, attn_lse
def _forward_decode_pcp_dcp( def _forward_decode_pcp_dcp(
self, self,
@@ -1788,3 +2145,33 @@ class AscendMLAImpl(MLAAttentionImpl):
attn_out_lse_list = attn_out_lse_list_pcp_dcp attn_out_lse_list = attn_out_lse_list_pcp_dcp
return attn_out_lse_list return attn_out_lse_list
# TODO use update op to replace this
def _update_out_and_lse(
self,
out: torch.Tensor,
lse: torch.Tensor,
block_out: torch.Tensor,
block_lse: torch.Tensor,
mask: torch.Tensor = None,
):
if out is None:
out = block_out.to(torch.float32)
lse = block_lse
else:
if mask is None:
mask = torch.ones([block_out.size(0)],
dtype=torch.uint8,
device=block_out.device)
out_mask = mask[:, None, None].expand_as(block_out)
lse_mask = mask[:, None, None].expand_as(block_lse)
block_out = block_out.to(torch.float32)
out_without_update = out.clone()
lse_without_update = lse.clone()
out = out - F.sigmoid(block_lse - lse) * (out - block_out)
lse = lse - F.logsigmoid(lse - block_lse)
# mask
out = torch.where(out_mask, out, out_without_update)
lse = torch.where(lse_mask, lse, lse_without_update)
return out, lse

50
vllm_ascend/attention/utils.py
View File

@@ -14,10 +14,19 @@ from vllm.forward_context import ForwardContext, get_forward_context
class AscendPrefillContextParallelMetadata: class AscendPrefillContextParallelMetadata:
pcp_allgather_restore_idx: torch.Tensor = None pcp_allgather_restore_idx: torch.Tensor = None
cp_kv_recover_idx_for_chunk: torch.Tensor = None
num_actual_tokens_pcp_padded: Optional[int] = None num_actual_tokens_pcp_padded: Optional[int] = None
num_computed_tokens_of_pcp_dcp: Optional[list[list[list[int]]]] = None num_computed_tokens_of_pcp_dcp: Optional[list[list[list[int]]]] = None
local_chunked_kv_lens: Optional[list[Optional[list[Optional[list[Optional[
list[int]]]]]]]] = None
mask_for_non_zero_chunk: Optional[List[bool]] = None
max_chunk_num: int = 0
q_head_idx_tensor: torch.Tensor = None q_head_idx_tensor: torch.Tensor = None
q_tail_idx_tensor: torch.Tensor = None q_tail_idx_tensor: torch.Tensor = None
@@ -106,6 +115,47 @@ class AscendCommonAttentionMetadata:
AscendPrefillContextParallelMetadata] = None AscendPrefillContextParallelMetadata] = None
def extract_req_dcp_by_chunk_pcp(lst,
chunk_idx,
dcp_size,
pcp_rank,
fill_value=0):
num_reqs = len(lst)
results: List[List[int]] = []
for i in range(num_reqs):
if len(lst[i]) == 0 or chunk_idx >= len(lst[i]):
# empty req or this req has no corresponding chunk, fill 0
results.append([fill_value] * dcp_size)
continue
dcp_values = lst[i][chunk_idx][pcp_rank]
results.append(dcp_values)
return results
def filter_chunked_req_indices(
seq_len: torch.Tensor,
mask_for_non_zero_chunk: Optional[List[bool]],
) -> torch.Tensor:
"""
filter the reqs which are doing real chunk_prefill.
Args:
seq_len: contains multi-req length: [req0_len, req1_len, ...]
mask_for_non_zero_chunk: [True, False, True, False, ...]
Returns:
filtered_indices: the real chunked req's indices
"""
assert mask_for_non_zero_chunk is not None and len(seq_len) == len(
mask_for_non_zero_chunk)
offsets = torch.cumsum(torch.cat([torch.tensor([0]), seq_len[:-1]]), dim=0)
filtered_indices = torch.cat([
torch.arange(offsets[i], offsets[i] + seq_len[i])
for i in range(len(mask_for_non_zero_chunk))
if mask_for_non_zero_chunk[i]
])
return filtered_indices
def split_decodes_and_prefills( def split_decodes_and_prefills(
common_attn_metadata: AscendCommonAttentionMetadata, common_attn_metadata: AscendCommonAttentionMetadata,
decode_threshold: int = 1, decode_threshold: int = 1,

30
vllm_ascend/worker/block_table.py
View File

@@ -77,14 +77,6 @@ class BlockTable:
self.block_table_np = self.block_table_cpu.numpy() self.block_table_np = self.block_table_cpu.numpy()
self.num_blocks_per_row = np.zeros(max_num_reqs, dtype=np.int32) self.num_blocks_per_row = np.zeros(max_num_reqs, dtype=np.int32)
self.slot_mapping_cpu = torch.zeros(self.max_num_batched_tokens,
dtype=torch.int64,
device="cpu",
pin_memory=self.pin_memory)
self.slot_mapping_np = self.slot_mapping_cpu.numpy()
self.slot_mapping = torch.zeros(self.max_num_batched_tokens,
dtype=torch.int32,
device=self.device)
try: try:
self.pcp_world_size = get_pcp_group( self.pcp_world_size = get_pcp_group(
).world_size if prefill_context_parallel_enable() else 1 ).world_size if prefill_context_parallel_enable() else 1
@@ -98,6 +90,20 @@ class BlockTable:
self.dcp_rank = 0 self.dcp_rank = 0
self.pcp_world_size = 1 self.pcp_world_size = 1
self.pcp_rank = 0 self.pcp_rank = 0
self.slot_mapping_cpu = torch.zeros(
self.max_num_batched_tokens +
2 * self.pcp_world_size * self.max_num_reqs,
dtype=torch.int64,
device="cpu",
pin_memory=self.pin_memory)
self.slot_mapping_np = self.slot_mapping_cpu.numpy()
self.slot_mapping = torch.zeros(
self.max_num_batched_tokens +
2 * self.pcp_world_size * self.max_num_reqs,
dtype=torch.int32,
device=self.device)
self.kernel_sizes = kernel_sizes self.kernel_sizes = kernel_sizes
self.cp_kv_cache_interleave_size = cp_kv_cache_interleave_size self.cp_kv_cache_interleave_size = cp_kv_cache_interleave_size
@@ -148,7 +154,7 @@ class BlockTable:
if self.dcp_world_size * self.pcp_world_size > 1: if self.dcp_world_size * self.pcp_world_size > 1:
# Note(hc): The DCP implement store kvcache with an interleave # Note(hc): The DCP implement store kvcache with an interleave
# style, the kvcache for the token whose token_idx is i is # style, the kvcache for the token whose token_idx is i is
# always stored on the GPU whose dcp_rank equals i % cp_world_size: # always stored on the GPU whose dcp_rank equals i % pcp_world_size:
# Use a "virtual block" which equals to world_size * block_size # Use a "virtual block" which equals to world_size * block_size
# for block_table_indices calculation. # for block_table_indices calculation.
@@ -268,12 +274,12 @@ class MultiGroupBlockTable:
# must be multiplied by dcp_world_size. # must be multiplied by dcp_world_size.
try: try:
dcp_world_size = get_dcp_group().world_size dcp_world_size = get_dcp_group().world_size
cp_world_size = get_pcp_group( pcp_world_size = get_pcp_group(
).world_size if prefill_context_parallel_enable() else 1 ).world_size if prefill_context_parallel_enable() else 1
except AssertionError: except AssertionError:
# DCP might not be initialized in testing # DCP might not be initialized in testing
dcp_world_size = 1 dcp_world_size = 1
cp_world_size = 1 pcp_world_size = 1
if kernel_sizes is None: if kernel_sizes is None:
kernel_sizes = [[0]] * len(block_sizes) kernel_sizes = [[0]] * len(block_sizes)
@@ -291,7 +297,7 @@ class MultiGroupBlockTable:
block_size, max_num_reqs, block_size, max_num_reqs,
max( max(
cdiv(max_model_len, cdiv(max_model_len,
block_size * dcp_world_size * cp_world_size), block_size * dcp_world_size * pcp_world_size),
1 + num_speculative_tokens), max_num_batched_tokens, 1 + num_speculative_tokens), max_num_batched_tokens,
pin_memory, device, kernel_size_list, pin_memory, device, kernel_size_list,
cp_kv_cache_interleave_size) cp_kv_cache_interleave_size)

301
vllm_ascend/worker/model_runner_v1.py
View File

@@ -29,7 +29,7 @@ from copy import deepcopy
from dataclasses import dataclass from dataclasses import dataclass
from multiprocessing import Manager from multiprocessing import Manager
from typing import (TYPE_CHECKING, Any, Dict, List, NamedTuple, Optional, from typing import (TYPE_CHECKING, Any, Dict, List, NamedTuple, Optional,
Union, cast) Tuple, Union, cast)
import numpy as np import numpy as np
import numpy.typing as npt import numpy.typing as npt
@@ -471,13 +471,19 @@ class NPUModelRunner(LoRAModelRunnerMixin):
device="cpu", device="cpu",
pin_memory=True) pin_memory=True)
self.seq_lens_np = self.seq_lens_cpu.numpy() self.seq_lens_np = self.seq_lens_cpu.numpy()
self.pcp_allgather_restore_idx = torch.zeros(self.max_num_tokens, self.pcp_allgather_restore_idx = torch.zeros(
dtype=torch.int32, self.max_num_tokens + 2 * self.pcp_size * self.max_num_reqs,
device=self.device) dtype=torch.int32,
device=self.device)
self.cp_kv_recover_idx_for_chunk: List[List[int]] = [
[] for _ in range(self.pcp_size)
]
self.num_pcp_pads = torch.zeros(self.max_num_reqs, dtype=torch.int32) self.num_pcp_pads = torch.zeros(self.max_num_reqs, dtype=torch.int32)
self.pcp_padded_slot_mapping = torch.zeros(self.max_num_tokens, self.pcp_padded_slot_mapping = torch.zeros(
dtype=torch.int32, self.max_num_tokens + 2 * self.pcp_size * self.max_num_reqs,
device=self.device) dtype=torch.int32,
device=self.device)
self.num_actual_tokens_pcp_padded = 0 self.num_actual_tokens_pcp_padded = 0
if self.speculative_config and self.pcp_size > 1: if self.speculative_config and self.pcp_size > 1:
self.input_ids_pcp_full = torch.zeros(self.max_num_tokens, self.input_ids_pcp_full = torch.zeros(self.max_num_tokens,
@@ -739,7 +745,8 @@ class NPUModelRunner(LoRAModelRunnerMixin):
backward_kwargs = {} backward_kwargs = {}
backward_kwargs["mm_features"] = new_req_data.mm_features backward_kwargs["mm_features"] = new_req_data.mm_features
self.requests[req_id] = CachedRequestState( # Create request state - PCP/DCP tracking will be computed below
req_state = CachedRequestState(
req_id=req_id, req_id=req_id,
prompt_token_ids=new_req_data.prompt_token_ids, prompt_token_ids=new_req_data.prompt_token_ids,
prompt_embeds=new_req_data.prompt_embeds, prompt_embeds=new_req_data.prompt_embeds,
@@ -750,9 +757,42 @@ class NPUModelRunner(LoRAModelRunnerMixin):
num_computed_tokens=new_req_data.num_computed_tokens, num_computed_tokens=new_req_data.num_computed_tokens,
output_token_ids=[], output_token_ids=[],
lora_request=new_req_data.lora_request, lora_request=new_req_data.lora_request,
local_chunked_kv_lens=None,
**backward_kwargs, **backward_kwargs,
) )
# Compute PCP/DCP tracking fields for chunked prefill
self.input_batch.local_chunked_kv_lens = [None] * self.max_num_reqs
if self.pcp_size * self.dcp_size > 1:
num_computed_tokens = new_req_data.num_computed_tokens
if num_computed_tokens > 0:
# Initialize with starting rank 0
temp_start_rank_dict = {req_id: (0, 0)}
# Compute token distribution for initial tokens
current_distribution = self.get_split_computed_tokens(
np.array([num_computed_tokens]),
request_ids=[req_id],
request_start_rank_dict=temp_start_rank_dict,
cp_kv_cache_interleave_size=self.parallel_config.
cp_kv_cache_interleave_size,
)[0]
# Update next_pcp_dcp_start_rank
req_state.next_pcp_dcp_start_rank = temp_start_rank_dict[
req_id][0]
req_state.token_blank_in_last_blk = temp_start_rank_dict[
req_id][1]
req_state.local_chunked_kv_lens = [
copy.deepcopy(current_distribution)
]
else:
# No computed tokens yet
req_state.local_chunked_kv_lens = []
self.requests[req_id] = req_state
# Only relevant for models using M-RoPE (e.g, Qwen2-VL) # Only relevant for models using M-RoPE (e.g, Qwen2-VL)
if self.uses_mrope: if self.uses_mrope:
self._init_mrope_positions(self.requests[req_id]) self._init_mrope_positions(self.requests[req_id])
@@ -769,8 +809,44 @@ class NPUModelRunner(LoRAModelRunnerMixin):
resumed_from_preemption = req_data.resumed_from_preemption[i] resumed_from_preemption = req_data.resumed_from_preemption[i]
# Update the cached states. # Update the cached states.
prev_num_computed_tokens = req_state.num_computed_tokens
req_state.num_computed_tokens = num_computed_tokens req_state.num_computed_tokens = num_computed_tokens
# Compute PCP/DCP tracking fields for chunked prefill
if self.pcp_size * self.dcp_size > 1:
# If this is the first chunk, initialize tracking fields
if req_state.local_chunked_kv_lens is None:
req_state.local_chunked_kv_lens = []
# Compute tokens added in this chunk (not cumulative)
chunk_tokens = num_computed_tokens - prev_num_computed_tokens
if chunk_tokens > 0:
# Create a temporary dict with this request's starting rank
temp_start_rank_dict = {
req_id: (req_state.next_pcp_dcp_start_rank,
req_state.token_blank_in_last_blk)
}
# Compute distribution for this chunk only
chunk_distribution = self.get_split_computed_tokens(
np.array([chunk_tokens]),
request_ids=[req_id],
request_start_rank_dict=temp_start_rank_dict,
cp_kv_cache_interleave_size=self.parallel_config.
cp_kv_cache_interleave_size,
)[0]
# Update next_pcp_dcp_start_rank for this request
req_state.next_pcp_dcp_start_rank = temp_start_rank_dict[
req_id][0]
req_state.token_blank_in_last_blk = temp_start_rank_dict[
req_id][1]
# Append this chunk's distribution to accumulation list
req_state.local_chunked_kv_lens.append(
copy.deepcopy(chunk_distribution))
if not is_last_rank: if not is_last_rank:
# When using PP, the scheduler sends the sampled tokens back, # When using PP, the scheduler sends the sampled tokens back,
# because there's no direct communication between the first- # because there's no direct communication between the first-
@@ -815,6 +891,10 @@ class NPUModelRunner(LoRAModelRunnerMixin):
self.input_batch.block_table.append_row( self.input_batch.block_table.append_row(
new_block_ids, req_index) new_block_ids, req_index)
# Update PCP/DCP tracking fields in input_batch
self.input_batch.local_chunked_kv_lens[
req_index] = req_state.local_chunked_kv_lens
# For the last rank, we don't need to update the token_ids_cpu # For the last rank, we don't need to update the token_ids_cpu
# because the sampled tokens are already cached. # because the sampled tokens are already cached.
if not is_last_rank: if not is_last_rank:
@@ -979,6 +1059,8 @@ class NPUModelRunner(LoRAModelRunnerMixin):
return None return None
if self.attn_mask_builder is None: if self.attn_mask_builder is None:
raise ValueError("Attn mask builder is None") raise ValueError("Attn mask builder is None")
if self.dcp_size > 1:
return self.attn_mask_builder.get_splitfuse_attn_mask()
# Pooling situation. # Pooling situation.
if self.model_config.runner_type == "pooling" and self.model_config.pooler_config.pooling_type == "CLS": if self.model_config.runner_type == "pooling" and self.model_config.pooler_config.pooling_type == "CLS":
return self.attn_mask_builder.get_pooling_mask(self.device) return self.attn_mask_builder.get_pooling_mask(self.device)
@@ -1378,6 +1460,49 @@ class NPUModelRunner(LoRAModelRunnerMixin):
scheduler_output, scheduler_output,
decode_threshold=self.reorder_batch_threshold) decode_threshold=self.reorder_batch_threshold)
def generate_kv_idx(self, tokens, scheduler_output):
if not self.pcp_size > 1:
return
self.cp_kv_recover_idx_for_chunk = [[] for _ in range(self.pcp_size)]
for i, req_id in enumerate(self.input_batch.req_ids):
num_scheduled_tokens = scheduler_output.num_scheduled_tokens[
req_id]
is_prefill = self.input_batch.num_computed_tokens_cpu[
i] < self.input_batch.num_prompt_tokens[i]
if is_prefill:
num_cp_padded_scheduled_tokens = cdiv(
num_scheduled_tokens,
2 * self.pcp_size) * (2 * self.pcp_size)
full_indices = list(
range(self.max_num_tokens * self.pcp_size * self.dcp_size +
self.pcp_size * self.dcp_size * self.max_num_reqs))
chunk_size = num_cp_padded_scheduled_tokens // (2 *
self.pcp_size)
num_added_recover_tokens = len(
self.cp_kv_recover_idx_for_chunk[0]) * self.pcp_size
for rank in range(self.pcp_size):
self.cp_kv_recover_idx_for_chunk[rank].extend(
full_indices[rank * chunk_size +
num_added_recover_tokens:(rank + 1) *
chunk_size + num_added_recover_tokens])
self.cp_kv_recover_idx_for_chunk[rank].extend(
full_indices[num_cp_padded_scheduled_tokens -
(rank + 1) * chunk_size +
num_added_recover_tokens:
num_cp_padded_scheduled_tokens -
rank * chunk_size +
num_added_recover_tokens])
cp_kv_recover_idx_for_chunk = torch.from_numpy(
np.concatenate(
self.cp_kv_recover_idx_for_chunk)).to(device=self.device)
cp_kv_recover_idx_for_chunk.copy_(torch.tensor(
np.array(self.cp_kv_recover_idx_for_chunk).flatten().tolist()),
non_blocking=True)
self.cp_kv_recover_idx_for_chunk = cp_kv_recover_idx_for_chunk.to(
torch.float32).argsort().to(torch.int32)
def _prepare_inputs( def _prepare_inputs(
self, self,
scheduler_output: "SchedulerOutput", scheduler_output: "SchedulerOutput",
@@ -1406,7 +1531,7 @@ class NPUModelRunner(LoRAModelRunnerMixin):
self.input_batch.num_computed_tokens_cpu[req_indices], self.input_batch.num_computed_tokens_cpu[req_indices],
arange, arange,
) )
self.generate_kv_idx(tokens, scheduler_output)
self.input_batch.block_table.compute_slot_mapping( self.input_batch.block_table.compute_slot_mapping(
req_indices, positions_np) req_indices, positions_np)
self.input_batch.block_table.commit_slot_mapping( self.input_batch.block_table.commit_slot_mapping(
@@ -1610,15 +1735,19 @@ class NPUModelRunner(LoRAModelRunnerMixin):
] ]
num_tokens_np = np.array(num_tokens, dtype=np.int32) num_tokens_np = np.array(num_tokens, dtype=np.int32)
num_reqs = self.input_batch.num_reqs num_reqs = self.input_batch.num_reqs
if self.pcp_size == 1: if self.pcp_size > 1:
discard_requests_mask = self.seq_lens_np[:num_reqs] < num_tokens_np
else:
# while pcp > 1, we need the original num_scheduled_tokens before split # while pcp > 1, we need the original num_scheduled_tokens before split
# to calculate discard_requests_mask # to calculate discard_requests_mask
tokens_original = [
scheduler_output.num_scheduled_tokens[i] for i in req_ids
]
original_seq_lens_np = ( original_seq_lens_np = (
self.input_batch.num_computed_tokens_cpu[:num_reqs] + self.input_batch.num_computed_tokens_cpu[:num_reqs] +
np.array(list(scheduler_output.num_scheduled_tokens.values()))) np.array(tokens_original, dtype=np.int32))
discard_requests_mask = original_seq_lens_np < num_tokens_np discard_requests_mask = original_seq_lens_np < num_tokens_np
else:
discard_requests_mask = self.seq_lens_np[:num_reqs] < num_tokens_np
discard_request_indices = np.nonzero(discard_requests_mask)[0] discard_request_indices = np.nonzero(discard_requests_mask)[0]
self.num_discarded_requests = len(discard_request_indices) self.num_discarded_requests = len(discard_request_indices)
self.discard_request_indices.np[:self.num_discarded_requests] = ( self.discard_request_indices.np[:self.num_discarded_requests] = (
@@ -1762,8 +1891,17 @@ class NPUModelRunner(LoRAModelRunnerMixin):
scheduler_output.num_scheduled_tokens) scheduler_output.num_scheduled_tokens)
# prepare pcp meta data # prepare pcp meta data
# For chunked prefill, use num_scheduled_tokens instead of cumulative seq_lens
# to correctly calculate chunk_len in _generate_pcp_metadata
if self.vllm_config.scheduler_config.chunked_prefill_enabled and self.pcp_size > 1:
# In chunked prefill, seq_lens_for_chunk should be the current chunk size
seq_lens_for_chunk = torch.from_numpy(
num_scheduled_tokens[:num_reqs])
else:
# Normal mode: use cumulative sequence lengths
seq_lens_for_chunk = seq_lens_cpu
long_seq_metadata = self._generate_pcp_metadata( long_seq_metadata = self._generate_pcp_metadata(
total_num_scheduled_tokens, seq_lens_cpu) total_num_scheduled_tokens, seq_lens_for_chunk, seq_lens_cpu)
# Prepare the attention metadata for each KV cache group and make layers # Prepare the attention metadata for each KV cache group and make layers
# in the same group share the same metadata. # in the same group share the same metadata.
for kv_cache_group_id, kv_cache_group_spec in enumerate( for kv_cache_group_id, kv_cache_group_spec in enumerate(
@@ -2690,7 +2828,7 @@ class NPUModelRunner(LoRAModelRunnerMixin):
dtype=torch.int32, dtype=torch.int32,
device=self.device) device=self.device)
long_seq_metadata = self._generate_pcp_metadata( long_seq_metadata = self._generate_pcp_metadata(
num_tokens, self.seq_lens_cpu) num_tokens, self.seq_lens_cpu, self.seq_lens_cpu)
if long_seq_metadata is not None: if long_seq_metadata is not None:
pcp_world_size = get_pcp_group( pcp_world_size = get_pcp_group(
).world_size if prefill_context_parallel_enable() else 1 ).world_size if prefill_context_parallel_enable() else 1
@@ -4266,23 +4404,149 @@ class NPUModelRunner(LoRAModelRunnerMixin):
[-1, pcp_world_size, dcp_world_size]) [-1, pcp_world_size, dcp_world_size])
return dcp_local_seq_lens return dcp_local_seq_lens
def _generate_pcp_metadata(self, total_num_scheduled_tokens, seq_lens): def get_split_computed_tokens(
self,
num_computed_tokens: np.ndarray,
request_ids: Optional[List[str]] = None,
request_start_rank_dict: Dict[str, tuple[
int, int]] = {}, # tuple: start_rank, tokens_blank_in_this_block
cp_kv_cache_interleave_size: int = 1
) -> list[Optional[list[Optional[list[int]]]]]:
"""Splits computed token counts across dcp and sp dimensions for distributed allocation.
Args:
num_computed_tokens: Number of tokens for each request (current chunk, not cumulative)
request_ids: Request IDs to track state
request_start_rank_dict: Dict mapping req_id to the starting rank for this chunk.
Will be updated with next starting rank after distribution.
Returns:
List of [pcp_size][dcp_size] distribution for each request
"""
self.pcp_world_size = get_pcp_group(
).world_size if prefill_context_parallel_enable() else 1
self.dcp_world_size = get_dcp_group().world_size
num_requests = len(num_computed_tokens)
assert request_start_rank_dict is not None and request_ids is not None and len(
request_ids) == num_requests
local_chunked_kv_lens = [[[0] * self.dcp_world_size
for _ in range(self.pcp_world_size)]
for _ in range(num_requests)]
total_ranks = self.pcp_world_size * self.dcp_world_size
for req_idx, (req_id, total_tokens) in enumerate(
zip(request_ids, num_computed_tokens)):
if total_tokens <= 0:
continue
# Get starting rank for this chunk
start_rank = 0
tokens_blank = 0
if request_start_rank_dict is not None:
start_rank, tokens_blank = request_start_rank_dict.get(
req_id, (0, 0))
if tokens_blank > 0: # need to continue writing in the last block of previous chunk
consumed_tokens = min(tokens_blank, total_tokens)
total_tokens -= consumed_tokens
tokens_blank -= consumed_tokens
pcp_idx = start_rank // self.dcp_world_size
dcp_idx = start_rank % self.dcp_world_size
local_chunked_kv_lens[req_idx][pcp_idx][
dcp_idx] += consumed_tokens
if tokens_blank == 0:
start_rank = (start_rank + 1) % total_ranks
if total_tokens == 0:
request_start_rank_dict[req_id] = (start_rank,
tokens_blank)
continue
virtual_size = total_ranks * cp_kv_cache_interleave_size
base = int(total_tokens) // virtual_size
# Distribute base tokens to all ranks
for rank_idx in range(total_ranks):
pcp_idx = rank_idx // self.dcp_world_size
dcp_idx = rank_idx % self.dcp_world_size
local_chunked_kv_lens[req_idx][pcp_idx][
dcp_idx] += base * cp_kv_cache_interleave_size
remainder = int(total_tokens) % virtual_size
if remainder == 0:
request_start_rank_dict[req_id] = (start_rank, tokens_blank)
continue
remain_blocks = cdiv(remainder, cp_kv_cache_interleave_size)
assert remain_blocks > 0
# Distribute remainder tokens starting from start_rank
for i in range(remain_blocks):
rank = (start_rank + i) % total_ranks
pcp_idx = rank // self.dcp_world_size
dcp_idx = rank % self.dcp_world_size
if i < remain_blocks - 1 or remainder % cp_kv_cache_interleave_size == 0: # not last block or divisible
local_chunked_kv_lens[req_idx][pcp_idx][
dcp_idx] += 1 * cp_kv_cache_interleave_size
tokens_blank = 0
else: # if last block and undivisible
local_chunked_kv_lens[req_idx][pcp_idx][
dcp_idx] += remainder % cp_kv_cache_interleave_size
tokens_blank = cp_kv_cache_interleave_size - (
remainder % cp_kv_cache_interleave_size)
start_rank = (start_rank + remain_blocks - 1) % total_ranks
if tokens_blank == 0:
start_rank = (start_rank + 1) % total_ranks
# Update next starting rank for this request
request_start_rank_dict[req_id] = (start_rank, tokens_blank)
return cast(List[Optional[List[Optional[List[int]]]]],
local_chunked_kv_lens)
def _get_chunked_req_mask_and_max_chunk(
self,
local_chunked_kv_lens: Optional[list[Optional[list[Optional[list[
Optional[list[int]]]]]]]] = None
) -> Tuple[List[bool], int]:
"""
given 4-d list [req][chunk][pcp][dcp], return:
1. if each req has any chunk (list[bool])
2. max chunk num along all reqs (int)
"""
assert local_chunked_kv_lens is not None
if len(local_chunked_kv_lens) == 0:
return ([], 0)
mask_for_non_zero_chunk = [
len(req) > 0 for req in local_chunked_kv_lens if req is not None
]
max_chunk_num = max(
(len(req) for req in local_chunked_kv_lens if req is not None),
default=0)
return mask_for_non_zero_chunk, max_chunk_num
def _generate_pcp_metadata(self, total_num_scheduled_tokens, seq_lens,
seq_lens_origin):
num_reqs = self.input_batch.num_reqs num_reqs = self.input_batch.num_reqs
num_decodes = sum(self.input_batch.num_computed_tokens_cpu[:num_reqs] num_decodes = sum(self.input_batch.num_computed_tokens_cpu[:num_reqs]
>= self.input_batch.num_prompt_tokens[:num_reqs]) >= self.input_batch.num_prompt_tokens[:num_reqs])
num_actual_tokens_pcp_padded = total_num_scheduled_tokens * self.pcp_size num_actual_tokens_pcp_padded = total_num_scheduled_tokens * self.pcp_size
self.num_actual_tokens_pcp_padded = num_actual_tokens_pcp_padded self.num_actual_tokens_pcp_padded = num_actual_tokens_pcp_padded
local_chunked_kv_lens = self.input_batch.local_chunked_kv_lens[
num_decodes:num_reqs]
mask_for_non_zero_chunk, max_chunk_num = self._get_chunked_req_mask_and_max_chunk(
local_chunked_kv_lens)
long_seq_metadata = None long_seq_metadata = None
if self.pcp_size * self.dcp_size > 1: if self.pcp_size * self.dcp_size > 1:
long_seq_metadata = AscendPrefillContextParallelMetadata( long_seq_metadata = AscendPrefillContextParallelMetadata(
num_actual_tokens_pcp_padded=num_actual_tokens_pcp_padded, num_actual_tokens_pcp_padded=num_actual_tokens_pcp_padded,
num_computed_tokens_of_pcp_dcp=self._get_pcp_local_seq_lens( num_computed_tokens_of_pcp_dcp=self._get_pcp_local_seq_lens(
seq_lens, seq_lens_origin,
self.pcp_size, self.pcp_size,
self.dcp_size, self.dcp_size,
self.parallel_config.cp_kv_cache_interleave_size, self.parallel_config.cp_kv_cache_interleave_size,
).numpy(), ).numpy(),
) local_chunked_kv_lens=local_chunked_kv_lens,
mask_for_non_zero_chunk=mask_for_non_zero_chunk,
max_chunk_num=max_chunk_num)
if self.pcp_size > 1: if self.pcp_size > 1:
q_head_idx, q_tail_idx = [], [] q_head_idx, q_tail_idx = [], []
kv_with_q_head_nomask_idx, kv_with_q_head_mask_idx = [], [] kv_with_q_head_nomask_idx, kv_with_q_head_mask_idx = [], []
@@ -4393,6 +4657,7 @@ class NPUModelRunner(LoRAModelRunnerMixin):
} }
long_seq_metadata.pcp_allgather_restore_idx = self.pcp_allgather_restore_idx[: long_seq_metadata.pcp_allgather_restore_idx = self.pcp_allgather_restore_idx[:
num_actual_tokens_pcp_padded] num_actual_tokens_pcp_padded]
long_seq_metadata.cp_kv_recover_idx_for_chunk = self.cp_kv_recover_idx_for_chunk
long_seq_metadata.q_head_idx_tensor = self.q_head_idx_tensor long_seq_metadata.q_head_idx_tensor = self.q_head_idx_tensor
long_seq_metadata.q_tail_idx_tensor = self.q_tail_idx_tensor long_seq_metadata.q_tail_idx_tensor = self.q_tail_idx_tensor
long_seq_metadata.q_full_idx = self.q_full_idx long_seq_metadata.q_full_idx = self.q_full_idx

15
vllm_ascend/worker/npu_input_batch.py
View File

@@ -73,6 +73,12 @@ class CachedRequestState:
lora_request: Optional[LoRARequest] = None lora_request: Optional[LoRARequest] = None
prompt_embeds: Optional[torch.Tensor] = None prompt_embeds: Optional[torch.Tensor] = None
# pcp/dcp param
local_chunked_kv_lens: Optional[list[Optional[list[Optional[
list[int]]]]]] = None # Records computed tokens for each chunk
next_pcp_dcp_start_rank: int = 0 # Tracks next starting rank for round-robin distribution
token_blank_in_last_blk: int = 0 # if the last block is not full, how many future tokens can be stored
def __post_init__(self): def __post_init__(self):
self.num_prompt_tokens = length_from_prompt_token_ids_or_embeds( self.num_prompt_tokens = length_from_prompt_token_ids_or_embeds(
self.prompt_token_ids, self.prompt_embeds) self.prompt_token_ids, self.prompt_embeds)
@@ -313,6 +319,10 @@ class InputBatch:
self.prev_sampled_token_ids_invalid_indices: Optional[set[int]] = None self.prev_sampled_token_ids_invalid_indices: Optional[set[int]] = None
self.prev_req_id_to_index: Optional[dict[str, int]] = None self.prev_req_id_to_index: Optional[dict[str, int]] = None
# pcp/dcp parameters
self.local_chunked_kv_lens: list[Optional[list[Optional[list[Optional[
list[int]]]]]]] = [None] * max_num_reqs
@property @property
def req_ids(self) -> list[str]: def req_ids(self) -> list[str]:
# None elements should only be present transiently # None elements should only be present transiently
@@ -385,6 +395,9 @@ class InputBatch:
self.num_computed_tokens_cpu[req_index] = request.num_computed_tokens self.num_computed_tokens_cpu[req_index] = request.num_computed_tokens
self.block_table.add_row(request.block_ids, req_index) self.block_table.add_row(request.block_ids, req_index)
# Add PCP/DCP tracking fields
self.local_chunked_kv_lens[req_index] = request.local_chunked_kv_lens
if sampling_params := request.sampling_params: if sampling_params := request.sampling_params:
if (self.is_spec_decode if (self.is_spec_decode
and is_spec_decode_unsupported(sampling_params)): and is_spec_decode_unsupported(sampling_params)):
@@ -680,6 +693,8 @@ class InputBatch:
last_req_index] last_req_index]
self.num_computed_tokens_cpu[ self.num_computed_tokens_cpu[
empty_index] = self.num_computed_tokens_cpu[last_req_index] empty_index] = self.num_computed_tokens_cpu[last_req_index]
self.local_chunked_kv_lens[
empty_index] = self.local_chunked_kv_lens[last_req_index]
self.block_table.move_row(last_req_index, empty_index) self.block_table.move_row(last_req_index, empty_index)
self.temperature_cpu[empty_index] = self.temperature_cpu[ self.temperature_cpu[empty_index] = self.temperature_cpu[
last_req_index] last_req_index]