migrate example configs to canonical attn_implementation

examples/llama-4/do-no-use-fa2/maverick-qlora-fsdp1.yaml

@@ -60,7 +60,7 @@ bf16: true
 tf32: true
 
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2
 
 gradient_checkpointing: offload
 gradient_checkpointing_kwargs:

examples/llama-4/do-no-use-fa2/scout-qlora-fsdp1.yaml

@@ -67,7 +67,7 @@ bf16: true
 tf32: true
 
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2
 
 warmup_ratio: 0.1
 evals_per_epoch: 1

examples/llama-4/do-no-use-fa2/scout-qlora-single-h100.yaml

@@ -70,7 +70,7 @@ bf16: true
 tf32: true
 
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2
 
 gradient_checkpointing: offload
 gradient_checkpointing_kwargs:

examples/llama-4/do-no-use-fa2/scout-vision-qlora-fsdp.yaml

@@ -62,7 +62,7 @@ bf16: true
 tf32: true
 
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2
 
 warmup_ratio: 0.1
 evals_per_epoch: 1

examples/llama-4/scout-qlora-flexattn-fsdp2.yaml

@@ -59,7 +59,7 @@ bf16: true
 tf32: true
 
 logging_steps: 1
-flex_attention: true
+attn_implementation: flex_attention
 flex_attn_compile_kwargs:
   dynamic: false
   mode: max-autotune-no-cudagraphs

examples/llama-4/scout-qlora-single-h100-flex.yaml

@@ -64,7 +64,7 @@ bf16: true
 tf32: true
 
 torch_compile: true
-flex_attention: true
+attn_implementation: flex_attention
 flex_attn_compile_kwargs:
   dynamic: false
   mode: max-autotune-no-cudagraphs

examples/llama-4/scout-vision-qlora-fsdp2-flex.yaml

@@ -61,7 +61,7 @@ bf16: true
 tf32: true
 
 logging_steps: 1
-flex_attention: true
+attn_implementation: flex_attention
 flex_attn_compile_kwargs:
   dynamic: false
   mode: max-autotune-no-cudagraphs