wip

justfile

@@ -2,6 +2,7 @@
 
 
 sft:
+    . ./.venv/bin/activate
     ACCELERATE_LOG_LEVEL=info accelerate launch --config_file accelerate_configs/deepspeed_zero3.yaml scripts/run_simpo.py training_configs/llama-3-2-1b-base-sft.yaml
 
 

training_configs/llama-3-2-1b-base-sft.yaml

@@ -18,11 +18,11 @@ bf16: true
 do_eval: true
 evaluation_strategy: steps
 eval_steps: 200
-gradient_accumulation_steps: 4
+gradient_accumulation_steps: 1
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: False
-hub_model_id: zephyr-7b-sft-full
+hub_model_id: llama-3-2-1b-sft
 hub_strategy: every_save
 learning_rate: 2.0e-05
 log_level: info
@@ -32,11 +32,11 @@ lr_scheduler_type: cosine
 max_seq_length: 2048
 max_steps: -1
 num_train_epochs: 1
-output_dir: /scratch/gpfs/DANQIC/ym0081/checkpoints_new/llama-3-8b-sft
-run_name: llama-3-8b-sft
+output_dir: /workspace/checkpoints_new/llama-3-2-1b-sft
+run_name: llama-3-2-1b-sft
 overwrite_output_dir: true
-per_device_eval_batch_size: 8
-per_device_train_batch_size: 8
+per_device_eval_batch_size: 32
+per_device_train_batch_size: 32
 push_to_hub: false
 remove_unused_columns: true
 report_to: