[CI] Revert PRs 34818 and 33600 (#34979)

tests/models/multimodal/processing/test_tensor_schema.py

@@ -13,7 +13,6 @@ import torch.nn as nn
 from PIL import Image
 
 from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
-from vllm.config.cache import CacheConfig
 from vllm.config.multimodal import (
     AudioDummyOptions,
     BaseDummyOptions,
@@ -132,9 +131,7 @@ def initialize_dummy_model(
 ):
     temp_file = tempfile.mkstemp()[1]
     current_device = torch.get_default_device()
-    vllm_config = VllmConfig(
-        model_config=model_config, cache_config=CacheConfig(block_size=16)
-    )
+    vllm_config = VllmConfig(model_config=model_config)
     with set_current_vllm_config(vllm_config=vllm_config):
         init_distributed_environment(
             world_size=1,

tests/models/utils.py

@@ -457,9 +457,6 @@ def dummy_hf_overrides(
     # Kimi uses `num_expert_group` instead of `n_group`.
     if n_group is None:
         n_group = getattr(text_config, "num_expert_group", None)
-    # InternS1Pro uses `router_n_groups` instead of `n_group`.
-    if n_group is None:
-        n_group = getattr(text_config, "router_n_groups", None)
     num_experts = n_group * 2 if n_group is not None else 2
 
     # we use three layers for Gemma-3n to check
@@ -489,14 +486,12 @@ def dummy_hf_overrides(
     # Only set MoE related config when the model has MoE layers.
     # Otherwise all models detected as MoE by _get_transformers_backend_cls.
     if model_arch_config.num_experts > 0:
-        orig_topk = getattr(text_config, "num_experts_per_tok", 2)
-        topk = min(orig_topk, 2)
         update_dict.update(
             {
                 "num_experts": num_experts,
-                "num_experts_per_tok": topk,
+                "num_experts_per_tok": 2,
                 # Kimi uses `num_experts_per_token`.
-                "num_experts_per_token": topk,
+                "num_experts_per_token": 2,
                 "num_local_experts": num_experts,
                 # Otherwise there will not be any expert layers
                 "first_k_dense_replace": 0,

tests/v1/spec_decode/test_eagle.py

@@ -78,7 +78,7 @@ def _create_proposer(
     device = current_platform.device_type
     vllm_config = VllmConfig(
         model_config=model_config,
-        cache_config=CacheConfig(block_size=16),
+        cache_config=CacheConfig(),
         speculative_config=speculative_config,
         device_config=DeviceConfig(device=device),
         parallel_config=ParallelConfig(),