Skip to content

Commit

Permalink
tweak recipe
Browse files Browse the repository at this point in the history
  • Loading branch information
yuiseki committed May 2, 2024
1 parent a125b92 commit a59e427
Show file tree
Hide file tree
Showing 4 changed files with 42 additions and 6 deletions.
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@ dataset_train_split_test_size: 0.2
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
train_claim_gpu_num: 4
train_per_device_train_batch_size: 8
train_gradient_accumulation_steps: 2
train_claim_gpu_num: 8
train_per_device_train_batch_size: 1
train_gradient_accumulation_steps: 16
train_num_train_epochs: 4
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,8 @@ base_model_id: yuiseki/Mistral-7B-v0.1-ja-wikipedia-v0.1
model_name: Mistral-7B-v0.1-ja-wikipedia-amenokaku-v0.1
output_base_dir: /data/output
dataset_id: kunishou/amenokaku-code-instruct
dataset_filter_field_name: liscence
dataset_filter_field_value: MIT
dataset_input_field_name: instruction
dataset_context_field_name: input
dataset_output_field_name: output
Expand All @@ -11,7 +13,7 @@ dataset_train_split_test_size: 0.2
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
train_claim_gpu_num: 4
train_per_device_train_batch_size: 8
train_gradient_accumulation_steps: 4
train_claim_gpu_num: 8
train_per_device_train_batch_size: 1
train_gradient_accumulation_steps: 16
train_num_train_epochs: 4
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
target_task: tasks/i18n/ja.md
base_model_id: yuiseki/Mistral-7B-v0.1-ja-wikipedia-v0.1
model_name: Mistral-7B-v0.1-ja-wikipedia-databricks-dolly-v0.1
output_base_dir: /data/output
dataset_id: llm-jp/databricks-dolly-15k-ja
dataset_context_field_name: context
dataset_input_field_name: instruction
dataset_output_field_name: response
dataset_filter_field_value: 0
dataset_train_split_seed: 42
dataset_train_split_test_size: 0.2
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
train_claim_gpu_num: 8
train_per_device_train_batch_size: 1
train_gradient_accumulation_steps: 16
train_num_train_epochs: 4
16 changes: 16 additions & 0 deletions recipes/A5000_24GB_x8/Mistral-7B-v0.1/i18n-ja-wikipedia-jimba.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
target_task: tasks/i18n/ja.md
base_model_id: yuiseki/Mistral-7B-v0.1-ja-wikipedia-v0.1
model_name: yuiseki/Mistral-7B-v0.1-ja-wikipedia-jimba-v0.1
output_base_dir: /data/output
dataset_id: Kendamarron/jimba-instuction-1k-beta
dataset_input_field_name: instruction
dataset_output_field_name: output
dataset_train_split_seed: 42
dataset_train_split_test_size: 0.2
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
train_claim_gpu_num: 4
train_per_device_train_batch_size: 8
train_gradient_accumulation_steps: 4
train_num_train_epochs: 4

0 comments on commit a59e427

Please sign in to comment.