migrate example configs to canonical attn_implementation

2026-04-23 22:15:07 +00:00
parent 2d64d009d8
commit 39226623d2
222 changed files with 209 additions and 243 deletions
--- a/examples/qwen3.5/122b-a10b-moe-qlora-fsdp.yaml
+++ b/examples/qwen3.5/122b-a10b-moe-qlora-fsdp.yaml
@@ -65,7 +65,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/122b-a10b-moe-qlora.yaml
+++ b/examples/qwen3.5/122b-a10b-moe-qlora.yaml
@@ -65,7 +65,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/27b-fft.yaml
+++ b/examples/qwen3.5/27b-fft.yaml
@@ -50,7 +50,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/27b-qlora-fsdp.yaml
+++ b/examples/qwen3.5/27b-qlora-fsdp.yaml
@@ -61,7 +61,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/27b-qlora.yaml
+++ b/examples/qwen3.5/27b-qlora.yaml
@@ -61,7 +61,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/35b-a3b-moe-qlora-fsdp.yaml
+++ b/examples/qwen3.5/35b-a3b-moe-qlora-fsdp.yaml
@@ -65,7 +65,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/35b-a3b-moe-qlora.yaml
+++ b/examples/qwen3.5/35b-a3b-moe-qlora.yaml
@@ -75,7 +75,7 @@ gradient_checkpointing: true
 activation_offloading: true
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 4
--- a/examples/qwen3.5/35b-a3b-moe-vision-lora.yaml
+++ b/examples/qwen3.5/35b-a3b-moe-vision-lora.yaml
@@ -50,7 +50,7 @@ gradient_checkpointing: true
 gradient_checkpointing_kwargs:
  use_reentrant: false
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 weight_decay: 0.0
--- a/examples/qwen3.5/9b-fft-vision.yaml
+++ b/examples/qwen3.5/9b-fft-vision.yaml
@@ -40,7 +40,7 @@ gradient_checkpointing_kwargs:
  use_reentrant: false
 resume_from_checkpoint:
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 1
--- a/examples/qwen3.5/9b-lora-vision.yaml
+++ b/examples/qwen3.5/9b-lora-vision.yaml
@@ -58,7 +58,7 @@ gradient_checkpointing: true
 gradient_checkpointing_kwargs:
  use_reentrant: false
 logging_steps: 1
-flash_attention: true
+attn_implementation: flash_attention_2

 warmup_ratio: 0.1
 evals_per_epoch: 1