Removed CB architecture support check

Signed-off-by: Rishin Raj <[email protected]>

Author: quic-rishinr

QEfficient/exporter/export_hf_to_cloud_ai_100.py

@@ -16,7 +16,6 @@
 from QEfficient.base.common import AUTO_MODEL_MAP_TO_MODEL_TYPE_MAP, QEFF_MODEL_TYPE, QEFFCommonLoader
 from QEfficient.base.modeling_qeff import QEFFBaseModel
 from QEfficient.exporter.export_utils import export_onnx, fix_onnx_fp16, generate_input_files, run_model_on_ort
-from QEfficient.transformers.modeling_utils import get_lists_of_cb_qeff_models
 from QEfficient.transformers.models.modeling_auto import QEFFAutoModelForCausalLM
 from QEfficient.utils import load_hf_tokenizer
 from QEfficient.utils.constants import QEFF_MODELS_DIR, Constants
@@ -316,13 +315,6 @@ def export_for_cloud(
     seq_length: int = Constants.SEQ_LEN,
     full_batch_size: Optional[int] = None,
 ) -> str:
-    # Check if model architecture is supported for continuous batching.
-    if full_batch_size and qeff_model.model.config.architectures[0].lower() not in {
-        x.lower() for x in get_lists_of_cb_qeff_models.architectures
-    }:
-        raise NotImplementedError(
-            f"Continuous batching is not supported for {qeff_model.model.config.architectures[0]}"
-        )
 
     # FIXME: move all this to class instead of here, and just call qeff_model.export here.
     if AUTO_MODEL_MAP_TO_MODEL_TYPE_MAP.get(qeff_model.__class__, None) == QEFF_MODEL_TYPE.CAUSALLM:  # type: ignore

QEfficient/transformers/modeling_utils.py

@@ -137,25 +137,6 @@
 # Required for the Automation tool
 ModelArchitectures = namedtuple("ModelArchitectures", ["architectures"])
 
-get_lists_of_cb_qeff_models = ModelArchitectures(
-    [
-        LlamaForCausalLM.__name__,
-        GemmaForCausalLM.__name__,
-        Gemma2ForCausalLM.__name__,
-        MistralForCausalLM.__name__,
-        MixtralForCausalLM.__name__,
-        Starcoder2ForCausalLM.__name__,
-        Qwen2ForCausalLM.__name__,
-        Phi3ForCausalLM.__name__,
-        PhiForCausalLM.__name__,
-        CodeGenForCausalLM.__name__,
-        GPT2LMHeadModel.__name__,
-        GPTJForCausalLM.__name__,
-        MptForCausalLM.__name__,
-        FalconForCausalLM.__name__,
-        GPTBigCodeForCausalLM.__name__,
-    ]
-)
 # Create an instance of the named tuple
 qeff_supported_architectures = ModelArchitectures(
     [

tests/cloud/conftest.py

@@ -11,7 +11,6 @@
 
 import pytest
 
-from QEfficient.transformers.modeling_utils import get_lists_of_cb_qeff_models
 from QEfficient.utils import get_onnx_dir_name
 from QEfficient.utils.constants import QEFF_MODELS_DIR
 from QEfficient.utils.logging_utils import logger
@@ -271,12 +270,6 @@ def pytest_collection_modifyitems(config, items):
             if item.module.__name__ in ["test_export", "test_compile", "test_execute", "test_infer"]:
                 if hasattr(item, "callspec"):
                     params = item.callspec.params
-                    model_class = model_class_dict[params["model_name"]]
-                    if (
-                        params["full_batch_size"] is not None
-                        and model_class not in get_lists_of_cb_qeff_models.architectures
-                    ):
-                        item.add_marker(pytest.mark.skip(reason="Skipping because FULL BATCH SIZE does not support..."))
 
             if item.module.__name__ in ["test_export", "test_compile", "test_execute"]:
                 if hasattr(item, "callspec"):