use `get_convert_fx_fn` for eager mode convert

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/564

As title, as we need `ai_factory.quantization.convert.convert_eager` for Stinson models. This diff renames ``get_convert_fx_fn` to `get_convert_fn` and includes eager mode convert functions as well

Reviewed By: wat3rBro

Differential Revision: D46368438

fbshipit-source-id: 5ebea1f05b43b476a14ab1091f6ce39bffe614d3

d2go/quantization/fx.py

@@ -10,8 +10,10 @@ from mobile_cv.common.misc.oss_utils import fb_overwritable
 
 TORCH_VERSION: Tuple[int, ...] = tuple(int(x) for x in torch.__version__.split(".")[:2])
 if TORCH_VERSION > (1, 10):
+    from torch.ao.quantization.quantize import convert
     from torch.ao.quantization.quantize_fx import convert_fx, prepare_fx, prepare_qat_fx
 else:
+    from torch.quantization.quantize import convert
     from torch.quantization.quantize_fx import convert_fx, prepare_fx, prepare_qat_fx
 
 
@@ -21,5 +23,8 @@ def get_prepare_fx_fn(cfg, is_qat):
 
 
 @fb_overwritable()
-def get_convert_fx_fn(cfg, example_inputs, qconfig_mapping=None, backend_config=None):
-    return convert_fx
+def get_convert_fn(cfg, example_inputs=None, qconfig_mapping=None, backend_config=None):
+    if cfg.QUANTIZATION.EAGER_MODE:
+        return convert
+    else:
+        return convert_fx

d2go/quantization/modeling.py

@@ -10,7 +10,7 @@ from typing import Any, Dict, Tuple
 import detectron2.utils.comm as comm
 import torch
 from d2go.quantization import learnable_qat
-from d2go.quantization.fx import get_convert_fx_fn, get_prepare_fx_fn
+from d2go.quantization.fx import get_convert_fn, get_prepare_fx_fn
 from d2go.quantization.qconfig import (
     set_backend_and_create_qconfig,
     smart_decode_backend,
@@ -329,7 +329,7 @@ def default_custom_prepare_fx(cfg, model, is_qat, example_input=None):
         qconfig_mapping=qconfig_dict,
         example_inputs=(example_input,),
     )
-    convert_fn = get_convert_fx_fn(cfg, (example_input,))
+    convert_fn = get_convert_fn(cfg, (example_input,))
     return model, convert_fn
 
 
@@ -396,7 +396,8 @@ def convert_to_quantized_model(cfg, fp32_model):
     quantized model (int8 operators).
     """
     if cfg.QUANTIZATION.EAGER_MODE:
-        int8_model = convert(fp32_model, inplace=False)
+        convert_fn = get_convert_fn(cfg)
+        int8_model = convert_fn(fp32_model, inplace=False)
     else:
         # FX graph mode quantization
         if not hasattr(fp32_model, _CONVERT_FX_CALLBACK_ATTRIBUTE):