Fix .reshape(...) in GaudiCLIPAttention (#2338)

Author: pbielak

optimum/habana/transformers/models/clip/modeling_clip.py

@@ -132,10 +132,8 @@ def forward(
 
             attn_output = self.bmm2(attn_weights, values)
 
-        # attn_output = attn_output.view(bsz, self.num_heads, tgt_len, self.head_dim)
         attn_output = attn_output.transpose(1, 2).contiguous()
-
-        attn_output = attn_output.reshape(batch_size, seq_length, embed_dim).contiguous()
+        attn_output = attn_output.reshape(batch_size, seq_length, -1).contiguous()
         attn_output = self.out_proj(attn_output)
 
         if not output_attentions: