fix nf4 memory issue by init op_context in forward (#1349)

* fix nf4 memory issue by init op_context in forward

* disable repack in init

* fix code style

bitsandbytes/backends/cpu_xpu_common.py

@@ -370,25 +370,6 @@ def quantize_4bit_impl(
             quant_type=quant_type,
         )
 
-    if ipex_cpu and _ipex_cpu_version_prereq(2, 3) and input_shape[1] % blocksize == 0 and quant_type == "nf4":
-        # lowp_mode: lowest precision for computation
-        lowp_mode = ipex_cpu.quantization.WoqLowpMode.BF16
-        state.op_context = torch.ops.ipex_prepack.weight_only_qlinear_prepack(
-            out.reshape([input_shape[0], input_shape[1] // 2]),
-            ipex_cpu.quantization.WoqWeightDtype.NF4,
-            input_shape,  # weight shape
-            absmax.view(input_shape[0], input_shape[1] // blocksize),  # scales
-            None,  # zero_points
-            None,  # bias
-            None,  # g_idx
-            None,  # batch_size
-            blocksize,
-            int(lowp_mode),
-            -1,  # act_quant_mode. -1 means don't quant activation
-        )
-        state.absmax = torch.Tensor()
-        return torch.empty([1, 0], dtype=torch.uint8), state
-
     return out.unsqueeze(0), state
 
 

bitsandbytes/nn/modules.py

@@ -19,6 +19,7 @@
     INVERSE_LINEAR_8BIT_WEIGHTS_FORMAT_MAPPING,
     LINEAR_8BIT_WEIGHTS_FORMAT_MAPPING,
     OutlierTracer,
+    enable_ipex_fusion,
 )
 
 T = TypeVar("T", bound="torch.nn.Module")
@@ -444,17 +445,35 @@ def _save_to_state_dict(self, destination, prefix, keep_vars):
         save weight and bias,
         then fill state_dict with components of quant_state
         """
+        if (
+            getattr(self.weight, "quant_state", None) is not None
+            and getattr(self.weight.quant_state, "op_context", None) is not None
+        ):
+            context = self.weight.quant_state.op_context
+            self.weight.data = context.to_public(context.get_weight()).reshape([1, -1])
+
         super()._save_to_state_dict(destination, prefix, keep_vars)  # saving weight and bias
 
         if getattr(self.weight, "quant_state", None) is not None:
+            if (
+                self.weight.quant_state.absmax.shape.numel() == 0
+                and getattr(self.weight.quant_state, "op_context", None) is not None
+            ):
+                self.weight.quant_state.absmax = context.get_scales().reshape(-1)
+                delattr(self.weight.quant_state, "op_context")
             for k, v in self.weight.quant_state.as_dict(packed=True).items():
                 destination[prefix + "weight." + k] = v if keep_vars else v.detach()
-            if getattr(self.weight.quant_state, "op_context", None) is not None:
-                context = self.weight.quant_state.op_context
-                destination[prefix + "weight." + "absmax"] = context.get_scales().reshape(-1)
-                self.weight.data = context.to_public(context.get_weight()).reshape([1, -1])
 
     def forward(self, x: torch.Tensor):
+        # Check if ipex fusion can be used
+        if (
+            x.device.type == "cpu"
+            and not hasattr(self.weight.quant_state, "op_context")
+            and self.weight.quant_state.shape[1] % self.weight.quant_state.blocksize == 0
+            and self.weight.quant_state.quant_type == "nf4"
+        ):
+            enable_ipex_fusion(self.weight, self.weight.quant_state)
+
         # weights are cast automatically as Int8Params, but the bias has to be cast manually
         if self.bias is not None and self.bias.dtype != x.dtype:
             self.bias.data = self.bias.data.to(x.dtype)

bitsandbytes/utils.py

@@ -200,6 +200,30 @@ def unpack_tensor_to_dict(tensor_data):
     return unpacked_dict
 
 
+def enable_ipex_fusion(weight, quant_state):
+    from bitsandbytes.backends.cpu_xpu_common import _ipex_cpu_version_prereq
+
+    if _ipex_cpu_version_prereq(2, 3):
+        import intel_extension_for_pytorch as ipex
+
+        lowp_mode = ipex.quantization.WoqLowpMode.BF16
+        quant_state.op_context = torch.ops.ipex_prepack.weight_only_qlinear_prepack(
+            weight.data.reshape([quant_state.shape[0], quant_state.shape[1] // 2]),
+            ipex.quantization.WoqWeightDtype.NF4,
+            quant_state.shape,  # weight shape
+            quant_state.absmax.view(quant_state.shape[0], quant_state.shape[1] // quant_state.blocksize),  # scales
+            None,  # zero_points
+            None,  # bias
+            None,  # g_idx
+            None,  # batch_size
+            quant_state.blocksize,
+            int(lowp_mode),
+            -1,  # act_quant_mode. -1 means don't quant activation
+        )
+        quant_state.absmax = torch.Tensor()
+        weight.data = torch.empty([1, 0], dtype=torch.uint8)
+
+
 class QuantState:
     """container for quantization state components to work with Params4bit and similar classes"""