update training sh files

Author: yilin-wu98

run_exp_bag.sh

@@ -32,7 +32,7 @@ for start_index in "${start_indices[@]}"; do
     --peft_method lora \
     --use_wm True \
     --seed 128 \
-    --output_dir /data/finetuned_models/run_01_16_custom_wm_gt_seed_128_vlm_finetuning_0.2%_gt_imagined_step63_1_history_16sample_size_bag_new_answer_set_task_open-word_test2_18epoch_print_eval_metrics_3class_bag_aug_to_same_save_every_shuffle_key_no_demo_data_hist_no_start_from_$start_index \
+    --output_dir /data/finetuned_models/run_bag_exp \
     --target_modules ['down_proj','multi_modal_linear','o_proj','k_proj','q_proj','gate_proj','up_proj','v_proj'] \
     --use_wandb \
     --custom_dataset.start_index  $start_index

run_exp_cup.sh

@@ -0,0 +1,39 @@
+#!/bin/bash
+
+# Array of start indices
+start_indices=(35)
+
+# Loop through each start index and run the command
+for start_index in "${start_indices[@]}"; do
+    echo "Running command with start index: $start_index"
+    CUDA_VISIBLE_DEVICES=1 FSDP_CPU_RAM_EFFICIENT_LOADING=1 torchrun --nnodes 1 --nproc_per_node 1 \
+    recipes/quickstart/finetuning/finetuning_wm.py \
+    --enable_fsdp \
+    --lr 1e-4 \
+    --num_epochs 10 \
+    --batch_size_training 10 \
+    --model_name mllama/Llama-3.2-11B-Vision-Instruct/custom \
+    --dist_checkpoint_root_folder ./finetuned_model \
+    --dist_checkpoint_folder fine-tuned \
+    --use_fast_kernels \
+    --dataset "custom_dataset" \
+    --custom_dataset.test_split "test" \
+    --custom_dataset.file "recipes/quickstart/finetuning/datasets/realcup_dataset_latent.py" \
+    --custom_dataset.data_path "realcup_data" \
+    --custom_dataset.answer_type "open-word" \
+    --custom_dataset.num_images 16 \
+    --custom_dataset.sample_size 16 \
+    --custom_dataset.num_history_images 1 \
+    --custom_dataset.imagined_steps 63 \
+    --custom_dataset.latent_mode "all" \
+    --run_validation True \
+    --batching_strategy padding \
+    --use_peft \
+    --peft_method lora \
+    --use_wm True \
+    --output_dir /data/finetuned_models/run_cup_exp \
+    --target_modules ['down_proj','multi_modal_linear','o_proj','k_proj','q_proj','gate_proj','up_proj','v_proj'] \
+    --use_wandb \
+    --custom_dataset.start_index  $start_index
+done
+

run_exp_fork_all.sh

@@ -6,7 +6,7 @@ start_indices=(75)
 # Loop through each start index and run the command
 for start_index in "${start_indices[@]}"; do
     echo "Running command with start index: $start_index"
-    CUDA_VISIBLE_DEVICES=1 FSDP_CPU_RAM_EFFICIENT_LOADING=1 torchrun --nnodes 1 --nproc_per_node 1 \
+    CUDA_VISIBLE_DEVICES=0 FSDP_CPU_RAM_EFFICIENT_LOADING=1 torchrun --nnodes 1 --nproc_per_node 1 \
     recipes/quickstart/finetuning/finetuning_wm.py \
     --enable_fsdp \
     --lr 1e-4 \
@@ -18,9 +18,8 @@ for start_index in "${start_indices[@]}"; do
     --use_fast_kernels \
     --dataset "custom_dataset" \
     --custom_dataset.test_split "test" \
-    --custom_dataset.file "recipes/quickstart/finetuning/datasets/realfork_dataset_latent_all.py" \
+    --custom_dataset.file "recipes/quickstart/finetuning/datasets/realfork_dataset_latent.py" \
     --custom_dataset.data_path "realfork_data" \
-    --custom_dataset.answer_type "open-word-fork" \
     --custom_dataset.num_images 16 \
     --custom_dataset.sample_size 16 \
     --custom_dataset.num_history_images 1 \
@@ -31,7 +30,7 @@ for start_index in "${start_indices[@]}"; do
     --use_peft \
     --peft_method lora \
     --use_wm True \
-    --output_dir /data/finetuned_models/run_02_21_custom_wm_150k_vlm_finetuning_0.2%_imagined_step63_1_history_16sample_size_fork_task_open-word-fork-all_18epoch_print_eval_metrics_3class_aug_failure_by2_shuffle_key_correct_prompt_hist_no_start_from_$start_index \
+    --output_dir /data/finetuned_models/run_fork_exp \
     --target_modules ['down_proj','multi_modal_linear','o_proj','k_proj','q_proj','gate_proj','up_proj','v_proj'] \
     --use_wandb \
     --custom_dataset.start_index  $start_index