f8d03d21f1
### What this PR does / why we need it?
`vllm_ascend` already supports several speculative decoding strategies
such as MTP, EAGLE, N-gram, and suffix decoding. However, Medusa is not
yet supported. Medusa is an efficient speculative decoding framework
that leverages a lightweight draft model to propose multiple tokens in a
single step, which can significantly improve decoding throughput and
reduce latency.
To enable Medusa-based speculative decoding on Ascend hardware and
provide more decoding options for users, this PR adds Medusa support
into the `vllm_ascend` speculative decoding pipeline.
### Does this PR introduce _any_ user-facing change?
This PR introduces Medusa speculative decoding as an additional
speculative decoding method:
✔ Adds `MedusaProposer` and integrates it into the speculative decoding
registry
✔ Extends `SpecDcodeType` with a `MEDUSA` enum entry
✔ Updates `NPUModelRunner` to recognize and invoke Medusa during
decoding
✔ Adds Medusa-specific handling in the draft token generation logic
✔ Ensures backward compatibility — Medusa is only used when explicitly
enabled
Key code changes include:
* New file: `vllm_ascend/spec_decode/medusa_proposer.py`
* Register Medusa in `get_spec_decode_method`
* Extend proposer type hints to include `MedusaProposer`
* Add a Medusa-specific branch in `generate_draft_token_ids`
* Pass `sample_hidden_states` required by Medusa
### How was this patch tested?
Medusa is implemented as a new proposer class (`MedusaProposer`)
following the existing speculative decoding interface. The integration
works as follows:
1. Users enable Medusa via the speculative decoding configuration.
2. `get_spec_decode_method()` returns a `MedusaProposer` instance when
`method="medusa"`.
3. During decoding, `NPUModelRunner` detects that the active drafter is
a `MedusaProposer`.
4. Instead of the generic speculative decoding path, the Medusa-specific
`generate_token_ids()` method is invoked, which consumes:
* `valid_sampled_token_ids`
* `sampling_metadata`
* `spec_decode_metadata`
* `sample_hidden_states`
5. The proposed tokens are validated by the target model as usual.
When Medusa is not enabled, the decoding pipeline behaves exactly as
before, ensuring full backward compatibility.
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: simplzyu <191163281@qq.com>
Signed-off-by: simplzyu <zhenyuguo@cmbchina.com>
54 lines
1.8 KiB
Python
54 lines
1.8 KiB
Python
import enum
|
|
from typing import Optional
|
|
|
|
import torch
|
|
from vllm.config import CUDAGraphMode, VllmConfig
|
|
from vllm.v1.core.sched.output import SchedulerOutput
|
|
from vllm.v1.sample.metadata import SamplingMetadata
|
|
from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
|
|
|
|
|
|
class SpecDcodeType(enum.Enum):
|
|
NGRAM = 0
|
|
EAGLE = 1
|
|
EAGLE3 = 2
|
|
MTP = 4
|
|
SUFFIX = 5
|
|
MEDUSA = 6
|
|
|
|
|
|
class Proposer:
|
|
|
|
def __init__(self,
|
|
vllm_config: VllmConfig,
|
|
device: torch.device = None,
|
|
runner=None):
|
|
pass
|
|
|
|
def load_model(self, model):
|
|
"""Called by load_model in model_runner"""
|
|
raise NotImplementedError
|
|
|
|
@torch.inference_mode()
|
|
def dummy_run(self,
|
|
num_tokens: int,
|
|
with_prefill: bool = False,
|
|
in_graph_capturing: bool = False,
|
|
num_reqs: int = 0,
|
|
num_tokens_across_dp: Optional[torch.Tensor] = None,
|
|
aclgraph_runtime_mode: CUDAGraphMode = CUDAGraphMode.NONE,
|
|
batch_descriptor=None):
|
|
"""Called by dummy_run in modle_runner"""
|
|
raise NotImplementedError
|
|
|
|
def generate_token_ids(self,
|
|
valid_sampled_token_ids: list[list[int]],
|
|
sampling_metadata: SamplingMetadata = None,
|
|
scheduler_output: SchedulerOutput = None,
|
|
spec_decode_metadata: SpecDecodeMetadata = None,
|
|
positions: torch.Tensor = None,
|
|
num_scheduled_tokens: int = 0,
|
|
hidden_states: torch.Tensor = None,
|
|
aux_hidden_states: torch.Tensor = None):
|
|
"""Called by execute_model in model_runner"""
|
|
raise NotImplementedError |