LLM: support speecht5_tts (#10077)

* support speecht5_tts

* fix

Author: rnwang04

python/llm/src/bigdl/llm/transformers/convert.py

@@ -238,6 +238,9 @@ def _replace_with_low_bit_linear(model, qtype, modules_to_not_convert=None,
                             new_linear._parameters['bias'] = nn.Parameter(module.bias.data)\
                                 .to(device)
                     elif qtype not in [ggml_tensor_qtype["fp16"], ggml_tensor_qtype["bf16"]]:
+                        if in_features % 64 != 0:
+                            # now our kernel requires in_features is a multiple of 64
+                            continue
                         new_linear = LowBitLinear(
                             in_features,
                             out_features,

python/llm/src/bigdl/llm/transformers/low_bit_linear.py

@@ -478,7 +478,11 @@ def forward(self, x: torch.Tensor):
             if x_2d.is_contiguous() is False:
                 x_2d = x_2d.contiguous()
 
-            input_seq_size = x_shape[1]
+            if len(x_shape) == 3:
+                input_seq_size = x_shape[1]
+            elif len(x_shape) < 3:
+                input_seq_size = 1
+
             if is_training:
                 # training path
                 if x_2d.requires_grad: