ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix #4005 #4013

Browse Source 
Former-commit-id: 8608fa268cde5cddf8d0c6c2eb2cb5fa246c1831
This commit is contained in:
hiyouga
2024-06-03 19:12:29 +08:00
parent 1a3764ab8f
commit 82d744716a
24 changed files with 52 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
examples/extras/badam/llama3_lora_sft.yaml
View File

@@ -28,10 +28,10 @@ overwrite_output_dir: true
### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 0.0001
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_steps: 0.1
warmup_ratio: 0.1
pure_bf16: true
### eval

4
examples/extras/fsdp_qlora/llama3_lora_sft.yaml
View File

@@ -29,10 +29,10 @@ overwrite_output_dir: true
### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 0.0001
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_steps: 0.1
warmup_ratio: 0.1
fp16: true
### eval

4
examples/extras/galore/llama3_full_sft.yaml
View File

@@ -29,10 +29,10 @@ overwrite_output_dir: true
### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 1
learning_rate: 0.0001
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_steps: 0.1
warmup_ratio: 0.1
pure_bf16: true
### eval

4
examples/extras/llama_pro/llama3_freeze_sft.yaml
View File

@@ -27,10 +27,10 @@ overwrite_output_dir: true
### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 0.0001
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_steps: 0.1
warmup_ratio: 0.1
fp16: true
### eval

4
examples/extras/loraplus/llama3_lora_sft.yaml
View File

@@ -26,10 +26,10 @@ overwrite_output_dir: true
### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 0.0001
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_steps: 0.1
warmup_ratio: 0.1
fp16: true
### eval

4
examples/extras/mod/llama3_full_sft.yaml
View File

@@ -26,10 +26,10 @@ overwrite_output_dir: true
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
optim: paged_adamw_8bit
learning_rate: 0.0001
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_steps: 0.1
warmup_ratio: 0.1
pure_bf16: true
### eval