Fix llama model o_proj lora_ids passing for finite lorax (#575)

This is regarding the issue reported in
#572

The finite lorax feature failed to execute when testing on a llama
adapter
([jumip/llama-lora-adapter](https://huggingface.co/jumip/llama-lora-adapter))
that contains o_proj as target module.

Work in progress -- test to be added to avoid future regression

Signed-off-by: Jou-An Chen <[email protected]>

Author: quic-jouachen

QEfficient/transformers/models/llama/modeling_llama.py

@@ -174,7 +174,7 @@ def forward(
         )
 
         attn_output = attn_output.reshape(*input_shape, -1).contiguous()
-        attn_output = self.o_proj(attn_output)
+        attn_output = self.o_proj(attn_output, **kwargs)
         return attn_output, attn_weights, past_key_value