b69b04d3a9
### What this PR does / why we need it?
This PR aim to implement model runner v2 basic framework in vllm-ascend,
the e2e function is not guaranteed by this pr.
### Does this PR introduce _any_ user-facing change?
use envs.VLLM_USE_V2_MODEL_RUNNER to decide if choose model_runenr_v2.
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
38 lines
1.0 KiB
Python
38 lines
1.0 KiB
Python
import numpy as np
|
|
import torch
|
|
from vllm.v1.worker.gpu.input_batch import InputBuffers
|
|
|
|
|
|
class AscendInputBuffers(InputBuffers):
|
|
"""Input buffers for Ascend NPUs."""
|
|
|
|
def __init__(
|
|
self,
|
|
max_num_reqs: int,
|
|
max_num_tokens: int,
|
|
inputs_embeds_size: int,
|
|
vocab_size: int,
|
|
dtype: torch.dtype,
|
|
device: torch.device,
|
|
pin_memory: bool,
|
|
):
|
|
super().__init__(
|
|
max_num_reqs,
|
|
max_num_tokens,
|
|
inputs_embeds_size,
|
|
vocab_size,
|
|
dtype,
|
|
device,
|
|
pin_memory,
|
|
)
|
|
# Create seq_lens_cpu and seq_lens_np.
|
|
# npu's attention backend still needs seq_lens on CPU side.
|
|
self.seq_lens_cpu: torch.Tensor = torch.zeros(
|
|
max_num_reqs,
|
|
dtype=torch.int32,
|
|
device="cpu",
|
|
)
|
|
# seq_len_np and seq_lens_cpu share the same memory.
|
|
# define seq_lens_np for easier calculation with numpy.
|
|
self.seq_lens_np: np.ndarray = self.seq_lens_cpu.numpy()
|