[Chore] Clean up deepseek v2/v3 config copy (#28055)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
@@ -292,6 +292,7 @@ class DeepseekDecoderLayer(nn.Module):
|
||||
rope_theta = getattr(config, "rope_theta", 10000)
|
||||
rope_scaling = getattr(config, "rope_scaling", None)
|
||||
max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
|
||||
moe_layer_freq = getattr(config, "moe_layer_freq", 1)
|
||||
self.self_attn = DeepseekAttention(
|
||||
hidden_size=self.hidden_size,
|
||||
num_heads=config.num_attention_heads,
|
||||
@@ -306,7 +307,7 @@ class DeepseekDecoderLayer(nn.Module):
|
||||
if (
|
||||
config.n_routed_experts is not None
|
||||
and layer_idx >= config.first_k_dense_replace
|
||||
and layer_idx % config.moe_layer_freq == 0
|
||||
and layer_idx % moe_layer_freq == 0
|
||||
):
|
||||
self.mlp = DeepseekMoE(
|
||||
config=config, quant_config=quant_config, prefix=f"{prefix}.mlp"
|
||||
|
||||
@@ -994,6 +994,7 @@ class DeepseekV2DecoderLayer(nn.Module):
|
||||
rope_theta = getattr(config, "rope_theta", 10000)
|
||||
rope_scaling = getattr(config, "rope_scaling", None)
|
||||
max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
|
||||
moe_layer_freq = getattr(config, "moe_layer_freq", 1)
|
||||
# DecoderLayers are created with `make_layers` which passes the prefix
|
||||
# with the layer's index.
|
||||
layer_idx = int(prefix.split(sep=".")[-1])
|
||||
@@ -1024,7 +1025,7 @@ class DeepseekV2DecoderLayer(nn.Module):
|
||||
if (
|
||||
config.n_routed_experts is not None
|
||||
and layer_idx >= config.first_k_dense_replace
|
||||
and layer_idx % config.moe_layer_freq == 0
|
||||
and layer_idx % moe_layer_freq == 0
|
||||
):
|
||||
self.mlp = DeepseekV2MoE(
|
||||
config=config,
|
||||
|
||||
@@ -50,7 +50,7 @@ from typing import Annotated, Any, Literal
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
from transformers import BatchFeature
|
||||
from transformers import BatchFeature, DeepseekV2Config
|
||||
from transformers.activations import GELUActivation
|
||||
|
||||
from vllm.config import VllmConfig
|
||||
@@ -91,7 +91,6 @@ from vllm.multimodal.processing import (
|
||||
from vllm.multimodal.profiling import BaseDummyInputsBuilder
|
||||
from vllm.sequence import IntermediateTensors
|
||||
from vllm.transformers_utils.configs import KimiVLConfig, MoonViTConfig
|
||||
from vllm.transformers_utils.configs.deepseek_vl2 import DeepseekV2Config
|
||||
from vllm.utils.tensor_schema import TensorSchema, TensorShape
|
||||
|
||||
from .utils import PPMissingLayer, is_pp_missing_parameter, maybe_prefix
|
||||
|
||||
Reference in New Issue
Block a user