[Feature][habana-main]: HPUAttentionImpl support for 'Encoder_Decoder'

[xuechendi]

🚀 The feature, motivation and pitch

Llama3.2 vision (Mllama) models requires model runner as "Enocoder_Decoder_Model_Runner"
which includes:

1. prepare "encoder_seq_lens" and "encoder_seq_lens_tensor" when preparing input data
   necessary fix for "HPUModelRunner - prepare_input_tensors" xuechendi@1f5a702
2. enable "Encoder self-attention" and "encoder/decoder cross-attention" in HPUAttentionImpl

test cmd:

python offline_inference_vision_language.py --model_type mllama

Error msg:

File "/workspace/vllm/vllm/attention/backends/hpu_attn.py", line 159, in forward
[rank0]:     raise NotImplementedError("Encoder self-attention and "
[rank0]: NotImplementedError: Encoder self-attention and encoder/decoder cross-attention are not implemented for HPUAttentionImpl

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.