remyxai

/

SpaceThinker-Qwen2.5VL-3B

like 9

Follow

Remyx AI 24

Certainly, @Abdorifaat
We've added train.py to the files and you can run:

python train.py \
  --model_id YOUR_ORG/QWEN_2.5_MODEL_ID_VARIANT \
  --dataset_id remyxai/SpaceThinker \
  --lora_r 64 \
  --lora_alpha 128 \
  --lora_dropout 0.1 \
  --target_modules q_proj,v_proj \
  --num_train_epochs 5 \
  --train_batch_size 2 \
  --eval_batch_size 2 \
  --gradient_accumulation_steps 4 \
  --learning_rate 1e-5 \
  --warmup_ratio 0.1 \
  --output_dir my-experiment-output \
  --wandb_project my-wandb-project \
  --wandb_run_name my-wandb-run

Will follow up by adding this to the documentation.
Hope your training run goes well!