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

[Refactor][EAGLE] 3/N delete redundant methods in mtp_proposer (#5420)

Browse Source 
### What this PR does / why we need it?
This PR aims to delete redundant methods in mtp_proposer. All the
deleted methods now can be found in eagle_proposer. We also remove some
methods in eagle_proposer since they are identical to those in
vllm-eagle.

### Does this PR introduce _any_ user-facing change?
N/A

### How was this patch tested?
by ci

- vLLM version: release/v0.13.0
- vLLM main:
81786c8774

---------

Signed-off-by: Zetong Li <slippersss@126.com>
This commit is contained in:
Zetong Li
2026-01-06 16:47:39 +08:00
committed by GitHub
parent b94d589769
commit fe3f2c7702
3 changed files with 97 additions and 483 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
tests/ut/spec_decode/test_mtp_proposer.py
View File

@@ -262,6 +262,7 @@ class TestMtpProposer:
device=torch.device("cpu")) device=torch.device("cpu"))
assert torch.equal(next_token_ids, expected_next_tokens) assert torch.equal(next_token_ids, expected_next_tokens)
@patch("vllm_ascend.spec_decode.eagle_proposer.HAS_TRITON", False)
@patch("vllm.v1.spec_decode.eagle.CpuGpuBuffer") @patch("vllm.v1.spec_decode.eagle.CpuGpuBuffer")
def test_prepare_inputs_padded(self, mock_cpu_gpu_buffer): def test_prepare_inputs_padded(self, mock_cpu_gpu_buffer):
mock_buffer_instance = MagicMock() mock_buffer_instance = MagicMock()

163
vllm_ascend/spec_decode/eagle_proposer.py
View File

@@ -4,6 +4,7 @@ from typing import Optional
import numpy as np import numpy as np
import torch import torch
import torch.nn as nn import torch.nn as nn
import torch.nn.functional as F
from vllm.config import (CompilationMode, CUDAGraphMode, VllmConfig, from vllm.config import (CompilationMode, CUDAGraphMode, VllmConfig,
get_layers_from_vllm_config) get_layers_from_vllm_config)
from vllm.distributed.parallel_state import get_pp_group from vllm.distributed.parallel_state import get_pp_group
@@ -15,6 +16,7 @@ from vllm.model_executor.models import supports_multimodal
from vllm.model_executor.models.deepseek_v2 import DeepseekV32IndexerCache from vllm.model_executor.models.deepseek_v2 import DeepseekV32IndexerCache
from vllm.model_executor.models.llama_eagle3 import Eagle3LlamaForCausalLM from vllm.model_executor.models.llama_eagle3 import Eagle3LlamaForCausalLM
from vllm.triton_utils import HAS_TRITON, triton from vllm.triton_utils import HAS_TRITON, triton
from vllm.utils.math_utils import cdiv
from vllm.utils.platform_utils import is_pin_memory_available from vllm.utils.platform_utils import is_pin_memory_available
from vllm.v1.attention.backends.utils import CommonAttentionMetadata from vllm.v1.attention.backends.utils import CommonAttentionMetadata
from vllm.v1.core.sched.output import SchedulerOutput from vllm.v1.core.sched.output import SchedulerOutput
@@ -28,7 +30,10 @@ from vllm_ascend.attention.attention_mask import AttentionMaskBuilder
from vllm_ascend.attention.attention_v1 import AscendAttentionState from vllm_ascend.attention.attention_v1 import AscendAttentionState
from vllm_ascend.attention.utils import AscendCommonAttentionMetadata from vllm_ascend.attention.utils import AscendCommonAttentionMetadata
from vllm_ascend.compilation.acl_graph import (ACLGraphWrapper, from vllm_ascend.compilation.acl_graph import (ACLGraphWrapper,
update_attn_params) update_attn_dcp_pcp_params,
update_attn_params,
update_mla_attn_dcp_pcp_params,
update_mla_attn_params)
from vllm_ascend.ops.rotary_embedding import update_cos_sin from vllm_ascend.ops.rotary_embedding import update_cos_sin
from vllm_ascend.ops.triton.spec_decode.utils import \ from vllm_ascend.ops.triton.spec_decode.utils import \
prepare_inputs_padded_kernel prepare_inputs_padded_kernel
@@ -37,10 +42,6 @@ from vllm_ascend.utils import shared_expert_dp_enabled
PADDING_SLOT_ID = -1 PADDING_SLOT_ID = -1
_DEFAULT_FIRST_LAYER = 'model.layers.0.self_attn.attn'
_FIRST_LAYERS = {"Qwen3NextForCausalLM": 'model.layers.3.self_attn.attn'}
# Currently we will fix block size to a small one since `num_reqs` can't be too large # Currently we will fix block size to a small one since `num_reqs` can't be too large
_PREPARE_INPUTS_BLOCK_SIZE = 4 _PREPARE_INPUTS_BLOCK_SIZE = 4
@@ -93,27 +94,6 @@ class EagleProposer(VllmEagleProposer):
self.use_sparse = hasattr(vllm_config.model_config.hf_text_config, self.use_sparse = hasattr(vllm_config.model_config.hf_text_config,
"index_topk") "index_topk")
def _get_eagle3_use_aux_hidden_state_from_config(self) -> bool:
"""
NOTE(2025-12-18): This is an explicit copy from vLLM EagleProposer, only added
to align with its logics.
Some eagle3 heads (e.g., nvidia/gpt-oss-120b-Eagle3-v2) do not use auxiliary
hidden states and directly uses the last layer output just like eagle1.
They might indicate this by setting "use_aux_hidden_state" to False
inside the "eagle_config" dict of their hf_config.
"""
if self.method != "eagle3":
return False
# Assume that eagle3 heads use aux hidden states by default
use_aux_hidden_state = True
eagle_config = getattr(self.draft_model_config.hf_config,
"eagle_config", None)
if eagle_config is not None:
use_aux_hidden_state = eagle_config.get("use_aux_hidden_state",
True)
return use_aux_hidden_state
def load_model(self, model: nn.Module) -> None: def load_model(self, model: nn.Module) -> None:
target_attn_layer_names = set( target_attn_layer_names = set(
get_layers_from_vllm_config(self.vllm_config, get_layers_from_vllm_config(self.vllm_config,
@@ -512,48 +492,6 @@ class EagleProposer(VllmEagleProposer):
draft_token_ids = draft_token_ids_tensor.swapaxes(0, 1) draft_token_ids = draft_token_ids_tensor.swapaxes(0, 1)
return draft_token_ids return draft_token_ids
def _get_attn_metadata(self, attn_metadata):
if attn_metadata is not None and isinstance(attn_metadata, dict):
architecture = self.vllm_config.model_config.architecture
layer_name = _FIRST_LAYERS.get(architecture, _DEFAULT_FIRST_LAYER)
attn_metadata = attn_metadata[layer_name]
return attn_metadata
def prepare_next_token_ids_cpu(
self,
sampled_token_ids: list[list[int]],
requests: dict[str, CachedRequestState],
gpu_input_batch: InputBatch,
num_scheduled_tokens: dict[str, int],
) -> torch.Tensor:
"""
This function is used to prepare the inputs for speculative decoding.
It calculates the next token ids for each request based on the sampled
token ids from the CPU. If a request has no sampled token ids (e.g.,
during the initial decoding steps), it falls back to using the request
state to get the next token id.
"""
req_ids = gpu_input_batch.req_ids
next_token_ids: list[int] = []
for i, token_ids in enumerate(sampled_token_ids):
if token_ids:
# Common case.
next_token_id = token_ids[-1]
else:
# Partial prefill (rare case).
# Get the next token id from the request state.
req_id = req_ids[i]
req_state = requests[req_id]
seq_len = req_state.num_computed_tokens + num_scheduled_tokens[
req_id]
next_token_id = req_state.get_token_id(seq_len)
next_token_ids.append(next_token_id)
next_token_ids = torch.tensor(next_token_ids,
dtype=torch.int32,
device=self.input_ids.device)
return next_token_ids
def prepare_next_token_ids_padded( def prepare_next_token_ids_padded(
self, self,
common_attn_metadata: CommonAttentionMetadata, common_attn_metadata: CommonAttentionMetadata,
@@ -829,3 +767,92 @@ class EagleProposer(VllmEagleProposer):
max_seq_len=0) max_seq_len=0)
return spec_common_attn_metadata, token_indices, token_indices_to_sample return spec_common_attn_metadata, token_indices, token_indices_to_sample
def _split_pcp_input(self, req_scheduled_tokens, input_ids,
target_hidden_states):
"""
Split prefill input_ids and target_hidden_states in pcp group.
1. input_ids padding: [t0, t1, t2, t3, t4, t5] -> [t0, t1, t2, t3, t4, t5, pad, pad]
2. split input_ids: pcp0 [t0, t1, pad, pad], pcp1 [t2, t3, t4, t5]
3. split target_hidden_states (already include pcp padding):
[h0, h1, h2, h3, h4, h5, pad, pad] -> pcp0 [h0, h1, pad, pad], pcp1 [h2, h3, h4, h5]
4. also update max_query_len, seq_lens, cu_num_tokens according to pcp split.
"""
if len(req_scheduled_tokens) == 0:
# no prefill inputs to split, return empty result
return (
0,
torch.zeros([0], device='npu'),
torch.zeros([0, target_hidden_states.size(1)], device='npu'),
0,
torch.zeros([0]),
torch.tensor([0], dtype=torch.int32),
)
def _pcp_pad_and_split(num_tokens):
num_pcp_padded_scheduled_tokens = cdiv(
num_tokens, 2 * self.pcp_size) * 2 * self.pcp_size
pcp_pad = num_pcp_padded_scheduled_tokens - num_tokens
chunk_size = num_pcp_padded_scheduled_tokens // (2 * self.pcp_size)
# split position_ids (and use split position_ids to split input_ids afterwards)
req_position_cp: list[int] = []
req_position_cp.extend(
self.full_indices[self.pcp_rank *
chunk_size:(self.pcp_rank + 1) * chunk_size])
req_position_cp.extend(
self.full_indices[num_pcp_padded_scheduled_tokens -
(self.pcp_rank + 1) *
chunk_size:num_pcp_padded_scheduled_tokens -
self.pcp_rank * chunk_size])
return req_position_cp, num_pcp_padded_scheduled_tokens, pcp_pad
num_pcp_scheduled_tokens = []
ori_start_index = 0
pad_start_index = 0
pcp_split_input_ids_list = []
pcp_split_hidden_states_list = []
for ori_num_tokens in req_scheduled_tokens.values():
req_position_pcp, num_pcp_padded_scheduled_tokens, num_pcp_pad = \
_pcp_pad_and_split(ori_num_tokens)
actual_num_tokens = len(req_position_pcp)
num_pcp_scheduled_tokens.append(actual_num_tokens)
pad_input_ids = F.pad(
input_ids[ori_start_index:ori_start_index + ori_num_tokens],
(0, num_pcp_pad))
ori_start_index += ori_num_tokens
pcp_chunk_indices = [
pad_start_index + pos for pos in req_position_pcp
]
pcp_split_input_ids = pad_input_ids[req_position_pcp]
pcp_split_hidden_states = target_hidden_states[pcp_chunk_indices]
pcp_split_input_ids_list.append(pcp_split_input_ids)
pcp_split_hidden_states_list.append(pcp_split_hidden_states)
pad_start_index += num_pcp_padded_scheduled_tokens
num_tokens = sum(num_pcp_scheduled_tokens)
input_ids = torch.cat(pcp_split_input_ids_list)
target_hidden_states = torch.cat(pcp_split_hidden_states_list, dim=0)
max_query_len = max(num_pcp_scheduled_tokens)
seq_lens = torch.tensor(num_pcp_scheduled_tokens, dtype=torch.int32)
cu_num_tokens = torch.tensor(
np.insert(np.cumsum(np.array(num_pcp_scheduled_tokens)), 0, 0))
return num_tokens, input_ids, target_hidden_states, max_query_len, seq_lens, cu_num_tokens
# update full-graph params for one spec token
def _update_full_graph_params(self, forward_context, num_tokens):
if self.vllm_config.model_config.use_mla:
if self.pcp_size * self.dcp_size > 1:
update_mla_attn_dcp_pcp_params(self.update_stream,
forward_context, num_tokens)
else:
update_mla_attn_params(self.update_stream, forward_context,
num_tokens,
self.vllm_config.speculative_config)
else:
if self.pcp_size * self.dcp_size > 1:
update_attn_dcp_pcp_params(self.update_stream, forward_context,
num_tokens)
else:
update_attn_params(self.update_stream, forward_context,
num_tokens, self.vllm_config)

416
vllm_ascend/spec_decode/mtp_proposer.py
View File

@@ -1,73 +1,31 @@
from typing import Optional, Union from typing import Optional, Union
import numpy as np
import torch import torch
import torch.nn as nn import torch.nn as nn
import torch.nn.functional as F
from vllm.config import CUDAGraphMode from vllm.config import CUDAGraphMode
from vllm.distributed import get_pcp_group from vllm.distributed import get_pcp_group
from vllm.forward_context import get_forward_context from vllm.forward_context import get_forward_context
from vllm.logger import init_logger
from vllm.model_executor.models.llama_eagle3 import Eagle3LlamaForCausalLM from vllm.model_executor.models.llama_eagle3 import Eagle3LlamaForCausalLM
from vllm.utils.math_utils import cdiv
from vllm.utils.platform_utils import is_pin_memory_available
from vllm.v1.attention.backends.utils import CommonAttentionMetadata from vllm.v1.attention.backends.utils import CommonAttentionMetadata
from vllm.v1.core.sched.output import SchedulerOutput from vllm.v1.core.sched.output import SchedulerOutput
from vllm.v1.sample.metadata import SamplingMetadata from vllm.v1.sample.metadata import SamplingMetadata
from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch
from vllm_ascend.ascend_forward_context import set_ascend_forward_context from vllm_ascend.ascend_forward_context import set_ascend_forward_context
from vllm_ascend.attention.attention_v1 import AscendAttentionState from vllm_ascend.attention.attention_v1 import AscendAttentionState
from vllm_ascend.attention.utils import AscendCommonAttentionMetadata from vllm_ascend.attention.utils import AscendCommonAttentionMetadata
from vllm_ascend.compilation.acl_graph import (ACLGraphWrapper, from vllm_ascend.compilation.acl_graph import ACLGraphWrapper
update_attn_dcp_pcp_params,
update_attn_params,
update_mla_attn_dcp_pcp_params,
update_mla_attn_params)
from vllm_ascend.ops.rotary_embedding import get_cos_and_sin_mla from vllm_ascend.ops.rotary_embedding import get_cos_and_sin_mla
from vllm_ascend.spec_decode.eagle_proposer import EagleProposer from vllm_ascend.spec_decode.eagle_proposer import EagleProposer
from vllm_ascend.utils import ProfileExecuteDuration, lmhead_tp_enable from vllm_ascend.utils import ProfileExecuteDuration, lmhead_tp_enable
logger = init_logger(__name__)
PADDING_SLOT_ID = -1 PADDING_SLOT_ID = -1
_MTP_MODELS = {
"DeepseekV3ForCausalLM":
("vllm.model_executor.models.deepseek_mtp", "DeepSeekMTP"),
"PanguUltraMoEForCausalLM":
("vllm.model_executor.models.openpangu_mtp", "OpenPanguMTP"),
"DeepseekV32ForCausalLM":
("vllm.model_executor.models.deepseek_mtp", "DeepSeekMTP"),
"Qwen3NextForCausalLM":
("vllm.model_executor.models.qwen3_next_mtp", "Qwen3NextMTP")
}
class MtpProposer(EagleProposer): class MtpProposer(EagleProposer):
# TODO: Find out why ModelRunner does not this explicit typing? # TODO: Find out why ModelRunner does not this explicit typing?
model: Union[nn.Module, ACLGraphWrapper] model: Union[nn.Module, ACLGraphWrapper]
# update full-graph params for one spec token
def _update_full_graph_params(self, forward_context, num_tokens):
if self.vllm_config.model_config.use_mla:
if self.pcp_size * self.dcp_size > 1:
update_mla_attn_dcp_pcp_params(self.update_stream,
forward_context, num_tokens)
else:
update_mla_attn_params(self.update_stream, forward_context,
num_tokens,
self.vllm_config.speculative_config)
else:
if self.pcp_size * self.dcp_size > 1:
update_attn_dcp_pcp_params(self.update_stream, forward_context,
num_tokens)
else:
update_attn_params(self.update_stream, forward_context,
num_tokens, self.vllm_config)
@torch.inference_mode() @torch.inference_mode()
def dummy_run(self, def dummy_run(self,
num_tokens: int, num_tokens: int,
@@ -180,124 +138,6 @@ class MtpProposer(EagleProposer):
if with_prefill: if with_prefill:
break break
def _prepare_inputs(
self,
common_attn_metadata: CommonAttentionMetadata,
sampled_token_ids: list[list[int]],
num_draft_tokens: list[int],
) -> tuple[CommonAttentionMetadata, torch.Tensor]:
"""
This function is used to prepare the inputs for speculative decoding.
It updates to the common_attn_metadata to account for the rejected
tokens (and newly sampled tokens). It also returns the token indices
of the tokens that should be fed to the speculator.
"""
# E.g.
# common_attn_metadata.query_start_loc{_cpu}:
# [0, q1, q1 + q2, q1 + q2 + q3]
# common_attn_metadata.seq_lens{_cpu}: [s1, s2, s3]
# num_rejected_tokens: [n1, n2, n3]
# This function computes the intermediate values:
# num_tokens_per_req: [q1 - n1, q2 - n2, q3 - n3]
# And returns:
# common_attn_metadata.query_start_loc{_cpu}:
# [0, q1 - n1, q1 + q2 - n1 - n2, q1 + q2 + q3 - n1 - n2 - n3]
# common_attn_metadata.seq_lens{_cpu}:
# [s1 - n1 + 1, s2 - n2 + 1, s3 - n3 + 1]
# token_indices: [0, 1, ..., q1 - n1 - 1,
# q1, q1 + 1, ..., q1 + q2 - n2 - 1,
# q1 + q2, q1 + q2 + 1, ..., q1 + q2 + q3 - n3 - 1]
num_actual_reqs = len(num_draft_tokens)
num_rejected_tokens = [
n + 1 - len(sampled_token_ids[i]) if n > 0 else 0
for i, n in enumerate(num_draft_tokens)
]
num_rejected_tokens = torch.tensor(num_rejected_tokens,
dtype=torch.int32)
device = common_attn_metadata.query_start_loc.device
query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu[:
num_actual_reqs
+ 1]
seq_lens_cpu = common_attn_metadata.seq_lens_cpu[:num_actual_reqs]
new_seq_lens_cpu = seq_lens_cpu - num_rejected_tokens
# [0, q1, q1 + q2, q1 + q2 + q3] -> [q1, q2, q3]
new_query_len_per_req = query_start_loc_cpu[
1:] - query_start_loc_cpu[:-1]
# [q1, q2, q3] -> [q1 - n1, q2 - n2, q3 - n3]
new_num_tokens_per_req = new_query_len_per_req - num_rejected_tokens
new_num_tokens_per_req_np = new_num_tokens_per_req.numpy()
# [q1 - n1, q2 - n2, q3 - n3] ->
# [0, q1 - n1, q1 + q2 - n1 - n2, q1 + q2 + q3 - n1 - n2 - n3]
new_query_start_loc_cpu = torch.zeros(
query_start_loc_cpu.shape,
dtype=torch.int32,
pin_memory=is_pin_memory_available(),
)
new_query_start_loc_np = new_query_start_loc_cpu.numpy()
np.cumsum(new_num_tokens_per_req_np, out=new_query_start_loc_np[1:])
total_num_tokens = new_query_start_loc_np[-1]
# Example assuming num_tokens_per_req_np = [2, 4, 3]
# this implies that `new_query_start_locs` is:
# [0, 2, 6, 9] ->
# [0, 0, 2, 2, 2, 2, 6, 6, 6]
# _r1_ ____r2____ ___r3__
new_query_start_locs_expanded = np.repeat(new_query_start_loc_np[:-1],
new_num_tokens_per_req_np)
# [0, 1, 2, 3, 4, 5, 6, 7, 8] ->
# [0, 1, 0, 1, 2, 3, 0, 1, 2]
# _r1_ ____r2____ ___r3__
token_offests = (self.token_arange_np[:total_num_tokens] -
new_query_start_locs_expanded)
# Expand starting positions to match token pattern
# [0, q1, q1 + q2] ->
# [0, 0, q1, q1, q1, q1, q1 + q2, q1 + q2, q1 + q2]
# _r1_ _____r2_______ ___________r3____________
old_query_start_locs_expanded = np.repeat(
query_start_loc_cpu[:-1].numpy(), new_num_tokens_per_req_np)
# Final token indices are:
# [0, 1, // req 1
# q1 + 0, q1 + 1, q1 + 2, q1 + 3, // req 2
# q1 + q2 + 0, q1 + q2 + 1, q1 + q2 + 2] // req 3
token_indices_np = token_offests + old_query_start_locs_expanded
token_indices = torch.from_numpy(token_indices_np).to(
device, non_blocking=True)
common_attn_metadata.slot_mapping[:token_indices.shape[0]].copy_(
common_attn_metadata.slot_mapping[token_indices])
common_attn_metadata.slot_mapping[token_indices.shape[0]:].fill_(-1)
# NOTE: Currently positions and seq_lens are not used in mla_v1 forward
# so we do not need to fixed them. But if they are used in the future,
# we should fixed them.
spec_common_attn_metadata = AscendCommonAttentionMetadata(
query_start_loc=new_query_start_loc_cpu.to(device,
non_blocking=True),
query_start_loc_cpu=new_query_start_loc_cpu,
seq_lens=new_seq_lens_cpu.to(device, non_blocking=True),
seq_lens_cpu=new_seq_lens_cpu,
num_computed_tokens_cpu=common_attn_metadata.
num_computed_tokens_cpu,
num_reqs=common_attn_metadata.num_reqs,
num_actual_tokens=total_num_tokens,
num_input_tokens=common_attn_metadata.num_input_tokens,
max_query_len=new_query_len_per_req.max().item(),
block_table_tensor=common_attn_metadata.block_table_tensor,
slot_mapping=common_attn_metadata.slot_mapping,
actual_seq_lengths_q=self.runner.actual_seq_lengths_q,
positions=common_attn_metadata.positions[token_indices],
attn_mask=self.runner.attn_mask,
spec_attn_mask=self.runner.spec_attn_mask,
attn_state=self.runner.attn_state,
decode_token_per_req=self.runner.decode_token_per_req,
max_seq_len=0)
return spec_common_attn_metadata, token_indices
def _propose( def _propose(
self, self,
# [num_tokens] # [num_tokens]
@@ -731,257 +571,3 @@ class MtpProposer(EagleProposer):
# mtp>1: [batch_size, k] # mtp>1: [batch_size, k]
draft_token_ids = torch.stack(draft_token_ids_list, dim=1) draft_token_ids = torch.stack(draft_token_ids_list, dim=1)
return draft_token_ids return draft_token_ids
# TODO Using torch instead of triton may result in poor performance
def _prepare_input_kernel(self, out_ptr: torch.Tensor,
cu_query_lens: torch.Tensor,
cu_num_tokens: torch.Tensor, block_size: int):
device = cu_query_lens.device
dtype = out_ptr.dtype
offsets = torch.arange(block_size, device=device, dtype=dtype)
start_pos = cu_num_tokens[:-1]
end_pos = cu_num_tokens[1:]
num_tokens = end_pos - start_pos
global_indices = (start_pos.view(-1, 1) + offsets.view(1, -1))
values = (cu_query_lens[:-1].view(-1, 1) + offsets.view(1, -1))
mask = (offsets.view(1, -1) < num_tokens.view(-1, 1))
global_indices_flat = global_indices[mask]
values_flat = values[mask]
out_ptr[global_indices_flat] = values_flat
def prepare_next_token_ids_cpu(
self,
sampled_token_ids: list[list[int]],
requests: dict[str, CachedRequestState],
gpu_input_batch: InputBatch,
num_scheduled_tokens: dict[str, int],
) -> torch.Tensor:
"""
This function is used to prepare the inputs for speculative decoding.
It calculates the next token ids for each request based on the sampled
token ids from the CPU. If a request has no sampled token ids (e.g.,
during the initial decoding steps), it falls back to using the request
state to get the next token id.
"""
req_ids = gpu_input_batch.req_ids
next_token_ids: list[int] = []
for i, token_ids in enumerate(sampled_token_ids):
if token_ids:
# Common case.
next_token_id = token_ids[-1]
else:
# Partial prefill (rare case).
# Get the next token id from the request state.
req_id = req_ids[i]
req_state = requests[req_id]
seq_len = req_state.num_computed_tokens + num_scheduled_tokens[
req_id]
next_token_id = req_state.get_token_id(seq_len)
next_token_ids.append(next_token_id)
next_token_ids = torch.tensor(next_token_ids,
dtype=torch.int32,
device=self.input_ids.device)
return next_token_ids
def prepare_next_token_ids_padded(
self,
common_attn_metadata: CommonAttentionMetadata,
sampled_token_ids: torch.Tensor,
requests: dict[str, CachedRequestState],
gpu_input_batch: InputBatch,
discard_request_indices: torch.Tensor,
num_discarded_requests: int,
) -> tuple[torch.Tensor, torch.Tensor]:
"""
This function is used to prepare the inputs for speculative decoding.
It calculates the next token ids and the number of valid sampled tokens
for each request, considering the "discarded" requests whose next token
is not sampled and comes from `request.get_token_id()` instead.
It also accounts for the rejected tokens in `sampled_token_ids`.
This function must use device functions to operate on the inputs, and
should not introduce any blocking CPU-GPU synchronization.
"""
# TODO(Ben): Combine this into a custom fused kernel
# Precompute get_token_id for when there is no valid next token
num_reqs = gpu_input_batch.num_reqs
self.backup_next_token_ids.np[:num_reqs] = np.array([
requests[gpu_input_batch.req_ids[i]].get_token_id(
common_attn_metadata.seq_lens_cpu[i].item())
for i in range(num_reqs)
])
self.backup_next_token_ids.copy_to_gpu(num_reqs)
# Mask out the sampled tokens indices that should not be sampled.
discard_sampled_tokens_req_indices = discard_request_indices[:
num_discarded_requests]
valid_sampled_token_ids_gpu = sampled_token_ids.clone()
valid_sampled_token_ids_gpu.index_fill_(
0, discard_sampled_tokens_req_indices, -1)
# Generate a mask for all valid tokens within those requests
valid_mask = (valid_sampled_token_ids_gpu != -1) & (
valid_sampled_token_ids_gpu < gpu_input_batch.vocab_size)
# Count the number of valid tokens in each request
valid_sampled_tokens_count = valid_mask.sum(dim=1)
# Get the rightmost valid index per row
last_valid_indices = valid_sampled_tokens_count - 1
last_valid_indices_safe = torch.clamp(last_valid_indices, min=0)
# Get last valid token from each row
# (assume undefined state where there is no valid token)
selected_tokens = torch.gather(
valid_sampled_token_ids_gpu, 1,
last_valid_indices_safe.unsqueeze(1)).squeeze(1)
# Use last token if valid, pre-computed backup if not
batch_size = valid_sampled_token_ids_gpu.shape[0]
next_token_ids = torch.where(
last_valid_indices != -1,
selected_tokens,
self.backup_next_token_ids.gpu[:batch_size],
)
return next_token_ids, valid_sampled_tokens_count
def prepare_inputs_padded(
self,
common_attn_metadata: CommonAttentionMetadata,
spec_decode_metadata: SpecDecodeMetadata,
valid_sampled_tokens_count: torch.Tensor,
) -> tuple[CommonAttentionMetadata, torch.Tensor, torch.Tensor]:
"""
This function is used to prepare the inputs for speculative decoding
It updates the common_attn_metadata for speculative decoding,
but does not consider the rejected tokens. Instead, all tokens
are included as inputs to the speculator, with the rejected tokens
used as padding and filtered out later by `token_indices_to_sample`.
No blocking CPU operations should be introduced in this function.
"""
num_draft_tokens_gpu = torch.cat([
spec_decode_metadata.cu_num_draft_tokens[0:1],
spec_decode_metadata.cu_num_draft_tokens[1:] -
spec_decode_metadata.cu_num_draft_tokens[:-1],
])
num_rejected_tokens_gpu = torch.where(
num_draft_tokens_gpu > 0,
num_draft_tokens_gpu + 1 - valid_sampled_tokens_count,
torch.zeros_like(num_draft_tokens_gpu),
)
query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu
new_query_len_per_req = query_start_loc_cpu[
1:] - query_start_loc_cpu[:-1]
total_num_tokens = query_start_loc_cpu[-1].item()
token_indices = self.arange[:total_num_tokens]
# NOTE: Currently positions and seq_lens are not used in mla_v1 forward
# so we do not need to fixed them. But if they are used in the future,
# we should fixed them.
spec_common_attn_metadata = AscendCommonAttentionMetadata(
query_start_loc=common_attn_metadata.query_start_loc,
query_start_loc_cpu=query_start_loc_cpu,
seq_lens_cpu=common_attn_metadata.seq_lens_cpu,
num_reqs=common_attn_metadata.num_reqs,
num_actual_tokens=total_num_tokens,
num_input_tokens=common_attn_metadata.num_input_tokens,
max_query_len=new_query_len_per_req.max().item(),
actual_seq_lengths_q=self.runner.actual_seq_lengths_q,
block_table_tensor=common_attn_metadata.block_table_tensor,
slot_mapping=common_attn_metadata.slot_mapping,
positions=common_attn_metadata.positions,
attn_mask=self.runner.attn_mask,
spec_attn_mask=self.runner.spec_attn_mask,
attn_state=self.runner.attn_state,
decode_token_per_req=self.runner.decode_token_per_req,
num_computed_tokens_cpu=common_attn_metadata.
num_computed_tokens_cpu,
seq_lens=common_attn_metadata.seq_lens,
max_seq_len=0)
query_start_loc = common_attn_metadata.query_start_loc[
1:1 + num_rejected_tokens_gpu.shape[0]]
token_indices_to_sample = query_start_loc - 1 - num_rejected_tokens_gpu
return spec_common_attn_metadata, token_indices, token_indices_to_sample
def _split_pcp_input(self, req_scheduled_tokens, input_ids,
target_hidden_states):
"""
Split prefill input_ids and target_hidden_states in pcp group.
1. input_ids padding: [t0, t1, t2, t3, t4, t5] -> [t0, t1, t2, t3, t4, t5, pad, pad]
2. split input_ids: pcp0 [t0, t1, pad, pad], pcp1 [t2, t3, t4, t5]
3. split target_hidden_states (already include pcp padding):
[h0, h1, h2, h3, h4, h5, pad, pad] -> pcp0 [h0, h1, pad, pad], pcp1 [h2, h3, h4, h5]
4. also update max_query_len, seq_lens, cu_num_tokens according to pcp split.
"""
if len(req_scheduled_tokens) == 0:
# no prefill inputs to split, return empty result
return (
0,
torch.zeros([0], device='npu'),
torch.zeros([0, target_hidden_states.size(1)], device='npu'),
0,
torch.zeros([0]),
torch.tensor([0], dtype=torch.int32),
)
def _pcp_pad_and_split(num_tokens):
num_pcp_padded_scheduled_tokens = cdiv(
num_tokens, 2 * self.pcp_size) * 2 * self.pcp_size
pcp_pad = num_pcp_padded_scheduled_tokens - num_tokens
chunk_size = num_pcp_padded_scheduled_tokens // (2 * self.pcp_size)
# split position_ids (and use split position_ids to split input_ids afterwards)
req_position_cp: list[int] = []
req_position_cp.extend(
self.full_indices[self.pcp_rank *
chunk_size:(self.pcp_rank + 1) * chunk_size])
req_position_cp.extend(
self.full_indices[num_pcp_padded_scheduled_tokens -
(self.pcp_rank + 1) *
chunk_size:num_pcp_padded_scheduled_tokens -
self.pcp_rank * chunk_size])
return req_position_cp, num_pcp_padded_scheduled_tokens, pcp_pad
num_pcp_scheduled_tokens = []
ori_start_index = 0
pad_start_index = 0
pcp_split_input_ids_list = []
pcp_split_hidden_states_list = []
for ori_num_tokens in req_scheduled_tokens.values():
req_position_pcp, num_pcp_padded_scheduled_tokens, num_pcp_pad = \
_pcp_pad_and_split(ori_num_tokens)
actual_num_tokens = len(req_position_pcp)
num_pcp_scheduled_tokens.append(actual_num_tokens)
pad_input_ids = F.pad(
input_ids[ori_start_index:ori_start_index + ori_num_tokens],
(0, num_pcp_pad))
ori_start_index += ori_num_tokens
pcp_chunk_indices = [
pad_start_index + pos for pos in req_position_pcp
]
pcp_split_input_ids = pad_input_ids[req_position_pcp]
pcp_split_hidden_states = target_hidden_states[pcp_chunk_indices]
pcp_split_input_ids_list.append(pcp_split_input_ids)
pcp_split_hidden_states_list.append(pcp_split_hidden_states)
pad_start_index += num_pcp_padded_scheduled_tokens
num_tokens = sum(num_pcp_scheduled_tokens)
input_ids = torch.cat(pcp_split_input_ids_list)
target_hidden_states = torch.cat(pcp_split_hidden_states_list, dim=0)
max_query_len = max(num_pcp_scheduled_tokens)
seq_lens = torch.tensor(num_pcp_scheduled_tokens, dtype=torch.int32)
cu_num_tokens = torch.tensor(
np.insert(np.cumsum(np.array(num_pcp_scheduled_tokens)), 0, 0))
return num_tokens, input_ids, target_hidden_states, max_query_len, seq_lens, cu_num_tokens