[Mypy] Fix mypy typing for model interfaces and H/I models (#54079)

Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Taneem Ibrahim
2026-08-28 16:49:19 -04:00
committed by GitHub
co-authored by Wentao Ye
parent 74850f9f66
commit 21fa2c5a2a
23 changed files with 222 additions and 137 deletions
+18 -12
View File
@@ -26,7 +26,7 @@
import functools
from collections.abc import Iterable
from typing import Any
from typing import Any, ClassVar
import torch
from torch import nn
@@ -321,6 +321,14 @@ class InternS1ProMoeAttention(nn.Module):
dual_chunk_attention_config=dual_chunk_attention_config,
)
attn_kwargs: dict[str, Any] = (
{
"layer_idx": extract_layer_index(prefix),
"dual_chunk_attention_config": dual_chunk_attention_config,
}
if dual_chunk_attention_config
else {}
)
self.attn = Attention(
self.num_heads,
self.head_dim,
@@ -329,12 +337,7 @@ class InternS1ProMoeAttention(nn.Module):
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
**{
"layer_idx": extract_layer_index(prefix),
"dual_chunk_attention_config": dual_chunk_attention_config,
}
if dual_chunk_attention_config
else {},
**attn_kwargs,
)
self.q_norm = RMSNorm(self.head_dim, eps=rms_norm_eps)
@@ -489,6 +492,9 @@ class InternS1ProMoeLLMForCausalLM(Qwen3MoeForCausalLM):
class InternS1ProMoeMixtureOfExperts(MixtureOfExperts):
language_model: InternS1ProMoeLLMForCausalLM
num_local_physical_experts: int
def update_physical_experts_metadata(
self,
num_physical_experts: int,
@@ -538,8 +544,8 @@ class InternS1ProMoeMixtureOfExperts(MixtureOfExperts):
class InternS1ProForConditionalGeneration(
Qwen3VLForConditionalGeneration, InternS1ProMoeMixtureOfExperts
):
is_3d_moe_weight: bool = True
packed_modules_mapping = {
is_3d_moe_weight: ClassVar[bool] = True
packed_modules_mapping: dict[str, list[str]] = {
"qkv_proj": [
"q_proj",
"k_proj",
@@ -559,7 +565,7 @@ class InternS1ProForConditionalGeneration(
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super(Qwen3VLForConditionalGeneration, self).__init__()
config: PretrainedConfig = vllm_config.model_config.hf_config
multimodal_config = vllm_config.model_config.multimodal_config
multimodal_config = vllm_config.model_config.get_multimodal_config()
self.config = config
self.multimodal_config = multimodal_config
@@ -601,8 +607,8 @@ class InternS1ProForConditionalGeneration(
# Set MoE hyperparameters
self.set_moe_parameters()
def get_frope_params_map(self) -> str:
mapper = {}
def get_frope_params_map(self) -> dict[str, str]:
mapper: dict[str, str] = {}
for name, params in self.language_model.model.named_parameters():
if "rotary_emb.sin_coef" in name:
mapper["language_model.model.rotary_emb.sin_coef"] = (