update examples

Former-commit-id: 2f78b5d62a34ea4d157bbe91a253859d25c8a7fe
2025-08-22 13:42:51 +08:00 · 2024-06-28 01:17:07 +08:00 · 2024-06-28 01:17:07 +08:00 · 2105cf6000
commit 2105cf6000
parent 835f0578c2
18 changed files with 19 additions and 19 deletions
--- a/examples/extras/fsdp_qlora/llama3_lora_sft.yaml
+++ b/examples/extras/fsdp_qlora/llama3_lora_sft.yaml
@ -30,7 +30,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/extras/llama_pro/llama3_freeze_sft.yaml
+++ b/examples/extras/llama_pro/llama3_freeze_sft.yaml
@ -31,7 +31,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/extras/loraplus/llama3_lora_sft.yaml
+++ b/examples/extras/loraplus/llama3_lora_sft.yaml
@ -30,7 +30,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/extras/pissa/llama3_lora_sft.yaml
+++ b/examples/extras/pissa/llama3_lora_sft.yaml
@ -32,7 +32,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_full/llama3_full_sft_ds3.yaml
+++ b/examples/train_full/llama3_full_sft_ds3.yaml
@ -29,7 +29,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_dpo.yaml
+++ b/examples/train_lora/llama3_lora_dpo.yaml
@ -31,7 +31,7 @@ learning_rate: 5.0e-6
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_kto.yaml
+++ b/examples/train_lora/llama3_lora_kto.yaml
@ -30,7 +30,7 @@ learning_rate: 5.0e-6
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_ppo.yaml
+++ b/examples/train_lora/llama3_lora_ppo.yaml
@ -30,7 +30,7 @@ learning_rate: 1.0e-5
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### generate
--- a/examples/train_lora/llama3_lora_pretrain.yaml
+++ b/examples/train_lora/llama3_lora_pretrain.yaml
@ -28,7 +28,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_reward.yaml
+++ b/examples/train_lora/llama3_lora_reward.yaml
@ -25,11 +25,11 @@ overwrite_output_dir: true
 ### train
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 8
-learning_rate: 1.0e-5
+learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_sft.yaml
+++ b/examples/train_lora/llama3_lora_sft.yaml
@ -29,7 +29,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_sft_ds0.yaml
+++ b/examples/train_lora/llama3_lora_sft_ds0.yaml
@ -30,7 +30,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llama3_lora_sft_ds3.yaml
+++ b/examples/train_lora/llama3_lora_sft_ds3.yaml
@ -30,7 +30,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_lora/llava1_5_lora_sft.yaml
+++ b/examples/train_lora/llava1_5_lora_sft.yaml
@ -30,7 +30,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_qlora/llama3_lora_sft_aqlm.yaml
+++ b/examples/train_qlora/llama3_lora_sft_aqlm.yaml
@ -29,7 +29,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_qlora/llama3_lora_sft_awq.yaml
+++ b/examples/train_qlora/llama3_lora_sft_awq.yaml
@ -29,7 +29,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_qlora/llama3_lora_sft_gptq.yaml
+++ b/examples/train_qlora/llama3_lora_sft_gptq.yaml
@ -29,7 +29,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval
--- a/examples/train_qlora/llama3_lora_sft_otfq.yaml
+++ b/examples/train_qlora/llama3_lora_sft_otfq.yaml
@ -31,7 +31,7 @@ learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
-fp16: true
+bf16: true
 ddp_timeout: 180000000

 ### eval