fix(qlinear): size GPTQ buffers by bits/pack_dtype_bits instead of pack_factor or 32

The number of packed words is ceil(dim * bits / pack_dtype_bits), not ceil(dim / pack_factor).

- In _register_gptq_buffers, use in_features*bits/pack_dtype_bits and out_features*bits/pack_dtype_bits for qweight/qzeros rows/cols.

- In pack_original, use dim*bits/pack_dtype_bits for both qweight and qzeros.

This is equivalent to the previous pack_factor formula for 2/4/8-bit, and correct for 3-bit and non-32-bit pack dtypes (int8/int16).

Author: Qubitium

gptqmodel/nn_modules/qlinear/__init__.py

@@ -750,14 +750,14 @@ def _register_gptq_buffers(
 
         self.register_buffer(
             "qweight",
-            t.zeros((math.ceil(in_features / self.pack_factor), out_features), dtype=self.pack_dtype),
+            t.zeros((math.ceil(in_features * self.bits / self.pack_dtype_bits), out_features), dtype=self.pack_dtype),
         )
         self.register_buffer(
             "qzeros",
             t.zeros(
                 (
                     math.ceil(in_features / self.group_size),
-                    math.ceil(out_features / self.pack_factor),
+                    math.ceil(out_features * self.bits / self.pack_dtype_bits),
                 ),
                 dtype=self.pack_dtype,
             ),
@@ -1395,7 +1395,7 @@ def pack_original(self, linear: nn.Module, scales: t.Tensor, zeros: t.Tensor, g_
             int_weight = int_weight.to(t.int32).T.contiguous()
             int_weight = int_weight.numpy().astype(self.pack_np_math_dtype)
 
-            qweight = np.zeros((math.ceil(int_weight.shape[0] * self.bits / 32), int_weight.shape[1]),
+            qweight = np.zeros((math.ceil(int_weight.shape[0] * self.bits / self.pack_dtype_bits), int_weight.shape[1]),
                                dtype=self.pack_np_math_dtype)
             if self.bits in [2, 4, 8]:
                 for row in range(qweight.shape[0]):
@@ -1428,7 +1428,7 @@ def pack_original(self, linear: nn.Module, scales: t.Tensor, zeros: t.Tensor, g_
             self.register_buffer("qweight", t.from_numpy(qweight.astype(self.pack_np_dtype)))
 
             zeros = zeros.numpy().astype(self.pack_np_math_dtype)
-            qzeros = np.zeros((zeros.shape[0], math.ceil(zeros.shape[1] * self.bits / 32)), dtype=self.pack_np_math_dtype)
+            qzeros = np.zeros((zeros.shape[0], math.ceil(zeros.shape[1] * self.bits / self.pack_dtype_bits)), dtype=self.pack_np_math_dtype)
             if self.bits in [2, 4, 8]:
                 for col in range(qzeros.shape[1]):
                     for j in range(self.pack_factor):