WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 0 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 8 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8
WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 20 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 8 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8
# WANDB_MODE=online CUDA_VISIBLE_DEVICES=7  python src/lgmodeling/finetune.py \
#     --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
#     --seed 40 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 8 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
#     --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
#     --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
#     --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8
NVIDIA_TF32_OVERRIDE=0 JAX_DEFAULT_MATMUL_PRECISION=highest CUBLAS_WORKSPACE_CONFIG=:4096:8 WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 0 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 16 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8

NVIDIA_TF32_OVERRIDE=0 JAX_DEFAULT_MATMUL_PRECISION=highest CUBLAS_WORKSPACE_CONFIG=:4096:8 WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 20 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 16 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8
NVIDIA_TF32_OVERRIDE=0 JAX_DEFAULT_MATMUL_PRECISION=highest CUBLAS_WORKSPACE_CONFIG=:4096:8 WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 40 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 16 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8

NVIDIA_TF32_OVERRIDE=0 JAX_DEFAULT_MATMUL_PRECISION=highest CUBLAS_WORKSPACE_CONFIG=:4096:8 WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 0 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 4 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8

NVIDIA_TF32_OVERRIDE=0 JAX_DEFAULT_MATMUL_PRECISION=highest CUBLAS_WORKSPACE_CONFIG=:4096:8 WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 20 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 4 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8
NVIDIA_TF32_OVERRIDE=0 JAX_DEFAULT_MATMUL_PRECISION=highest CUBLAS_WORKSPACE_CONFIG=:4096:8 WANDB_MODE=online CUDA_VISIBLE_DEVICES=5  python src/lgmodeling/finetune.py \
    --model-path /mnt/data/vinhbk/weights/enwik8/lr0.00025-rope-step60000-warm0-size24-layer12-embd512-heads8-shared1-routed0-topk0/best_56000 \
    --seed 40 --tgt_len 512 --mem_len 512 --eval_tgt_len 128  --n_head 4 --finetune-mlp --lmc-layer-indices 0 1 2 3 4 5 6 7 8 9 10 11 \
    --learning-rate 0.00025 --batch-size 24 --max_step 60000 --warmup_step 0 --dataset enwik8 \
    --wandb-project LMC-Attention --wandb-group "GPT2-Enwik8-FFN" --wandb-entity "vinh-bui0512-hcmut"\
    --model-save-dir /mnt/data/vinhbk/weights/enwik8/gpt2-finetune --data-path /mnt/data/vinhbk/datasets/enwik8