Revert model package exports.

Restore tensorrt_llm/_torch/models/__init__.py to its state before the last nine file commits while keeping formatting lint-clean.

Signed-off-by: Athena Cai <athenac@nvidia.com>

Author: athena-nv

tensorrt_llm/_torch/models/__init__.py

@@ -5,16 +5,13 @@
 # under transformers >= 5.5; see _torch/configs/__init__.py.
 import tensorrt_llm._torch.configs  # noqa: F401
 
-from .modeling_afmoe import AfmoeForCausalLM
 from .modeling_auto import AutoModelForCausalLM
-from .modeling_bart import (BartForConditionalGeneration,
-                            MBartForConditionalGeneration)
 from .modeling_bert import BertForSequenceClassification
 from .modeling_clip import CLIPVisionModel
 from .modeling_cohere2 import Cohere2ForCausalLM
 from .modeling_deepseekv3 import DeepseekV3ForCausalLM
 from .modeling_exaone4 import Exaone4ForCausalLM
-from .modeling_exaone4_5 import Exaone4_5_ForConditionalGeneration
+from .modeling_exaone_moe import ExaoneMoeForCausalLM
 from .modeling_gemma3 import Gemma3ForCausalLM
 from .modeling_gemma3vl import Gemma3VLM
 from .modeling_glm import Glm4MoeForCausalLM
@@ -23,12 +20,9 @@
 from .modeling_hunyuan_moe import HunYuanMoEV1ForCausalLM
 from .modeling_hyperclovax import HCXVisionForCausalLM
 from .modeling_kimi_k25 import KimiK25ForConditionalGeneration
-from .modeling_laguna import LagunaForCausalLM
 from .modeling_llama import LlamaForCausalLM
 from .modeling_llava_next import LlavaNextModel
 from .modeling_minimaxm2 import MiniMaxM2ForCausalLM
-from .modeling_minimaxm3 import (MiniMaxM3ForCausalLM,
-                                 MiniMaxM3VLForConditionalGeneration)
 from .modeling_mistral import Mistral3VLM, MistralForCausalLM
 from .modeling_mixtral import MixtralForCausalLM
 from .modeling_nemotron import NemotronForCausalLM
@@ -50,31 +44,20 @@
 from .modeling_seedoss import SeedOssForCausalLM
 from .modeling_siglip import SiglipVisionModel
 from .modeling_starcoder2 import Starcoder2ForCausalLM
-from .modeling_step3p7 import Step3p7ForCausalLM
-from .modeling_step3p7vl import Step3p7VLForConditionalGeneration
-from .modeling_t5 import T5ForConditionalGeneration
 from .modeling_utils import get_model_architecture
 from .modeling_vila import VilaModel
 
-try:
-    from .modeling_exaone_moe import ExaoneMoeForCausalLM
-except ImportError:
-    ExaoneMoeForCausalLM = None
-
 # Note: for better readiblity, this should have same order as imports above
 __all__ = [
-    "AfmoeForCausalLM",
     "AutoModelForCausalLM",
-    "BartForConditionalGeneration",
     "BertForSequenceClassification",
     "CLIPVisionModel",
     "DeepseekV3ForCausalLM",
     "Exaone4ForCausalLM",
-    "Exaone4_5_ForConditionalGeneration",
+    "ExaoneMoeForCausalLM",
     "Gemma3ForCausalLM",
     "Gemma3VLM",
     "HCXVisionForCausalLM",
-    "LagunaForCausalLM",
     "HunYuanDenseV1ForCausalLM",
     "HunYuanMoEV1ForCausalLM",
     "KimiK25ForConditionalGeneration",
@@ -95,8 +78,6 @@
     "Qwen2MoeForCausalLM",
     "SiglipVisionModel",
     "Starcoder2ForCausalLM",
-    "T5ForConditionalGeneration",
-    "MBartForConditionalGeneration",
     "get_model_architecture",
     "VilaModel",
     "Qwen2VLModel",
@@ -112,24 +93,16 @@
     "Glm4MoeForCausalLM",
     "Qwen3VLModel",
     "MiniMaxM2ForCausalLM",
-    "MiniMaxM3ForCausalLM",
-    "MiniMaxM3VLForConditionalGeneration",
     "Cohere2ForCausalLM",
-    "Step3p7ForCausalLM",
-    "Step3p7VLForConditionalGeneration",
 ]
 
-if ExaoneMoeForCausalLM is not None:
-    __all__.append("ExaoneMoeForCausalLM")
-
 if transformers.__version__ >= "4.45.1":
     from .modeling_mllama import MllamaForConditionalGeneration  # noqa
 
     __all__.append("MllamaForConditionalGeneration")
 else:
-    print(
-        f"Failed to import MllamaForConditionalGeneration as transformers.__version__ {transformers.__version__} < 4.45.1"
-    )
+    print("Failed to import MllamaForConditionalGeneration as "
+          f"transformers.__version__ {transformers.__version__} < 4.45.1")
 
 # Gemma4 requires transformers>=5.5.0 (native Gemma4 config/model classes).
 # Import silently on failure -- `get_model_architecture` in modeling_utils.py

tensorrt_llm/_torch/pyexecutor/kv_cache_transceiver.py

@@ -10,7 +10,8 @@
 from tensorrt_llm.mapping import Mapping
 
 from .llm_request import LlmRequest
-from .mamba_cache_manager import BaseMambaCacheManager, CppMambaHybridCacheManager
+from .mamba_cache_manager import (BaseMambaCacheManager,
+                                  CppMambaHybridCacheManager)
 from .resource_manager import KVCacheManager
 
 CacheTransceiverCpp = tensorrt_llm.bindings.internal.batch_manager.CacheTransceiver
@@ -97,10 +98,9 @@ def create_kv_cache_transceiver(
                 f"enable chunked transfer.")
     elif (runtime == "CPP"
           and cache_transceiver_config.chunk_size_blocks is not None):
-        logger.warning(
-            "chunk_size_blocks is set but transceiver_runtime='CPP' "
-            "explicitly disables Python auto-selection; "
-            "chunk_size_blocks will be ignored.")
+        logger.warning("chunk_size_blocks is set but transceiver_runtime='CPP' "
+                       "explicitly disables Python auto-selection; "
+                       "chunk_size_blocks will be ignored.")
 
     # Warn when chunk_size_blocks is below the recommended floor.  The Pydantic
     # field is PositiveInt (>=1), but values below ~16 push the per-chunk RDMA
@@ -127,7 +127,8 @@ def create_kv_cache_transceiver(
                 f"got {cache_transceiver_config.backend}. "
                 f"Please use transceiver_runtime='CPP' for MPI, UCX, or MOONCAKE backends."
             )
-        from tensorrt_llm._torch.disaggregation.transceiver import KvCacheTransceiverV2
+        from tensorrt_llm._torch.disaggregation.transceiver import \
+            KvCacheTransceiverV2
         logger.info("Using KvCacheTransceiverV2")
         return KvCacheTransceiverV2(mapping, dist, kv_cache_manager,
                                     cache_transceiver_config)