Remove flash-attn from requirements and GPU inference example

Remove flash-attn as a required dependency and remove attn_implementation="flash_attention_2" from the GPU inference example.

The model works with PyTorch's built-in SDPA attention which is auto-selected by transformers when flash-attn is not installed.

README.md

@@ -535,16 +535,6 @@ pip install transformers==4.50.1
 pip install librosa
 ```
 
-To run in GPU, MERaLiON-2 requires `flash-attn`.
-
-```
-pip install flash-attn --no-build-isolation
-```
-
-> [!TIP]
-> Should you face any difficulties installing the above packages, you can try installing within this Docker container instead:
-> `pytorch/pytorch:2.5.1-cuda12.1-cudnn9-devel`, whose cuda and torch environments have been tested working.
-
 ### Audio Input
 
 - For ASR tasks, the maximum audio length is suggested to be 30 seconds at 16,000 Hz.
@@ -664,7 +654,6 @@ model = AutoModelForSpeechSeq2Seq.from_pretrained(
     repo_id,
     use_safetensors=True,
     trust_remote_code=True,
-    attn_implementation="flash_attention_2",
     torch_dtype=torch.bfloat16
 ).to(device)