[Mypy] Fix typing for R/S models (#54157)

Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
This commit is contained in:
Taneem Ibrahim
2026-09-10 09:43:08 -04:00
committed by GitHub
parent 73fb19151f
commit f9083cb83a
15 changed files with 146 additions and 85 deletions
+7 -7
View File
@@ -10,7 +10,7 @@ import torch
from torch import nn
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, ModelConfig, VllmConfig
from vllm.config import CacheConfig, VllmConfig
from vllm.distributed import (
get_pp_group,
get_tensor_model_parallel_world_size,
@@ -55,7 +55,7 @@ logger = init_logger(__name__)
class FusedMoEBlock(nn.Module):
def __init__(
self,
config: ModelConfig,
config: Step3TextConfig,
quant_config: QuantizationConfig | None = None,
prefix: str = "",
):
@@ -384,7 +384,7 @@ class Step3TextModel(nn.Module):
return hidden_states
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
qkv_params_mapping = [
qkv_params_mapping: list[tuple[str, str, float, float]] = [
# (param_name, shard_name, relative_start_idx, relative_end_idx)
(
".qkv_proj",
@@ -410,7 +410,7 @@ class Step3TextModel(nn.Module):
/ (self.config.share_q_dim + self.config.head_dim * 2),
),
]
stacked_params_mapping = [
stacked_params_mapping: list[tuple[str, str, int]] = [
# (param_name, shard_name, shard_id)
(".gate_up_proj", ".gate_proj", 0),
(".gate_up_proj", ".up_proj", 1),
@@ -421,7 +421,7 @@ class Step3TextModel(nn.Module):
"base_layer." if any(".base_layer." in name for name in params_dict) else ""
)
expert_params_mapping = [
expert_params_mapping: list[tuple[str, str, str]] = [
(
f".moe.experts.routed_experts.{base_layer}w13_weight",
".moe.gate_proj.weight",
@@ -460,7 +460,7 @@ class Step3TextModel(nn.Module):
break
else:
for mapping in expert_params_mapping:
param_name, weight_name, shard_id = mapping
param_name, weight_name, expert_shard_id = mapping
if weight_name not in name:
continue
name = name.replace(weight_name, param_name)
@@ -480,7 +480,7 @@ class Step3TextModel(nn.Module):
param,
loaded_weight_expert,
name,
shard_id=shard_id,
shard_id=expert_shard_id,
expert_id=expert_id,
)
loaded_params.add(name)