diff --git a/data/config.yaml b/data/config.yaml new file mode 100755 index 0000000..86ee1bb --- /dev/null +++ b/data/config.yaml @@ -0,0 +1,85 @@ +dataset: + training: [ + "./data/captchas/" + ] + + validation: [] + + use_hdf5: False + + workers: 0 + cache: True + +models: + _models: + - name: "captcha" + tokens: 0 + len: 6 + +hyperparameters: + batch_size: 256 + gradient_accumulation_steps: 5 + gradient_clipping: 100 + + optimizer: Adamw + learning_rate: 5.0e-5 + + scheduler_type: "" + #scheduler_type: OneCycle + #scheduler_params: + # cycle_first_step_size: 10_000 + # cycle_first_stair_count: 10_000 + + # cycle_second_step_size: 15_000 + # cycle_second_stair_count: 15_000 + + # decay_step_size: 5_000 + + # cycle_min_lr: 2.5e-4 # 1.0e-5 + # cycle_max_lr: 2.5e-4 # 1.0e-4 + # decay_lr_rate: 0.0 + + # cycle_min_mom: 0.90 + # cycle_max_mom: 0.99 + # decay_mom_rate: 0.0 + +evaluation: + batch_size: 32 + frequency: 250 + size: 32 + + steps: 300 + temperature: 1.0 + +trainer: + iterations: 100_000 + + save_tag: step + save_on_oom: True + save_on_quit: True + save_frequency: 100 + + aggressive_optimizations: False + + check_for_oom: False + + #load_tag: "9500" + #load_state_dict: True + #load_states: False + #strict_loading: False + #restart_step_count: True + + gc_mode: None # "global_step" + + weight_dtype: float32 + + backend: local + deepspeed: + zero_optimization_level: 0 + use_compression_training: True + +inference: + use_vocos: True + +bitsandbytes: + enabled: false \ No newline at end of file