Revert "Embedding quantization per backend (#402)" (#411)

This reverts commit 052fb1a to  fix [test-tinystories-executorch](https://github.com/pytorch/torchchat/actions/runs/8802811768/job/24159454773?pr=411#logs)

Author: malfet

build/utils.py

@@ -23,7 +23,7 @@
 def set_backend(dso, pte):
     global active_builder_args_dso
     global active_builder_args_pte
-    active_builder_args_dso = dso
+    active_builder_args_dso = dso 
     active_builder_args_pte = pte
 
 

quantize.py

@@ -15,7 +15,7 @@
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from build.utils import find_multiple, get_precision, use_et_backend
+from build.utils import find_multiple, get_precision
 
 
 #########################################################################
@@ -92,6 +92,30 @@ def quantized_model(self) -> nn.Module:
         return self.quantizer.quantize(self.model_)
 
 
+#########################################################################
+###                QuantHandler API definition                        ###
+###               (unify with torchao in future)                      ###
+
+
+class QuantHandler:
+    def __init__(self, model: nn.Module, device="cpu", tokenizer=None):
+        self.model_ = model
+        self.device = device
+        self.tokenizer = tokenizer
+
+    def create_quantized_state_dict(self) -> Dict:  # "StateDict"
+        pass
+
+    def convert_for_runtime(self) -> nn.Module:
+        pass
+
+    def quantized_model(self) -> nn.Module:
+        model_updated_state_dict = self.create_quantized_state_dict()
+        self.convert_for_runtime()
+        self.model_.load_state_dict(model_updated_state_dict)
+        return self.model_
+
+
 #########################################################################
 ###           wrapper for setting precision as a QuantHandler         ###
 
@@ -623,12 +647,6 @@ def __init__(
         self.groupsize = groupsize
         self.dtype = dtype
         self.packed = packed
-
-        if use_et_backend():
-            self.forward = self.et_forward
-        else:
-            self.forward = self.aoti_forward
-
         if not packed:
             self.register_buffer(
                 "weight",
@@ -657,18 +675,12 @@ def __init__(
             )
 
     @torch.no_grad()
-    def et_forward(self, indices: torch.Tensor) -> torch.Tensor:
-        if self.packed:
-            return torch.ops.quantized_decomposed.embedding_byte.dtype(
-                self.weight, self.scales, None, 0, 0, indices, dtype=self.dtype
-            )
-        else:
-            return torch.ops.quantized_decomposed.embedding_4bit.dtype(
+    def forward(self, indices: torch.Tensor) -> torch.Tensor:
+        if False:  # Used for Executorch
+            return torch.ops.llama_quantized.embedding_byte.dtype(
                 self.weight, self.scales, None, 0, 0, indices, dtype=self.dtype
             )
 
-    @torch.no_grad()
-    def aoti_forward(self, indices: torch.Tensor) -> torch.Tensor:
         # result_weights = self.weight.index_select(0, indices.view(-1))
         # result_scales = self.scales.index_select(0, indices.view(-1))