37 lines
6.0 KiB
Plaintext
37 lines
6.0 KiB
Plaintext
|
PK training_args/data.pklFB ZZZZZZZZ€ctrl.trainer.sft_config
|
|||
|
|
SFTConfig
|
|||
|
|
q )�q}q(X
|
|||
|
|
output_dirqX ./outputsqX overwrite_output_dirq‰X do_trainq‰X do_evalq‰X
|
|||
|
|
do_predictq‰X
|
|||
|
|
eval_strategyq ctransformers.trainer_utils
|
|||
|
|
IntervalStrategy
|
|||
|
|
q
|
|||
|
|
X noq …q Rq
|
|||
|
|
X prediction_loss_onlyq‰X per_device_train_batch_sizeqKX per_device_eval_batch_sizeqKX per_gpu_train_batch_sizeqNX per_gpu_eval_batch_sizeqNX gradient_accumulation_stepsqKX eval_accumulation_stepsqNX
|
|||
|
|
eval_delayqK X torch_empty_cache_stepsqNX
|
|||
|
|
learning_rateqG?*6âë C-X weight_decayqG?„záG®{X
|
|||
|
|
adam_beta1qG?ìÌÌÌÌÌÍX
|
|||
|
|
adam_beta2qG?ï÷ÎÙ‡+X adam_epsilonqG>EyŽâ0Œ:X
|
|||
|
|
max_grad_normq G?ð X num_train_epochsq KX max_stepsq K<X lr_scheduler_typeq ctransformers.trainer_utils
|
|||
|
|
SchedulerType
|
|||
|
|
q X linearq!…q"Rq#X lr_scheduler_kwargsq$}q%X warmup_ratioq&G X warmup_stepsq'KX log_levelq(X passiveq)X log_level_replicaq*X warningq+X log_on_each_nodeq,ˆX logging_dirq-X* ./outputs/runs/Aug13_23-30-02_df7db15e9335q.X logging_strategyq/h
|
|||
|
|
X stepsq0…q1Rq2X logging_first_stepq3‰X
|
|||
|
|
logging_stepsq4KX logging_nan_inf_filterq5ˆX
|
|||
|
|
save_strategyq6ctransformers.trainer_utils
|
|||
|
|
SaveStrategy
|
|||
|
|
q7h0…q8Rq9X
|
|||
|
|
save_stepsq:MôX save_total_limitq;NX save_safetensorsq<ˆX save_on_each_nodeq=‰X save_only_modelq>‰X' restore_callback_states_from_checkpointq?‰X no_cudaq@‰X use_cpuqA‰X use_mps_deviceqB‰X seedqCMO
|
|||
|
|
X data_seedqDNX
|
|||
|
|
jit_mode_evalqE‰X use_ipexqF‰X bf16qGˆX fp16qH‰X fp16_opt_levelqIX O1qJX half_precision_backendqKX autoqLX bf16_full_evalqMˆX fp16_full_evalqN‰X tf32qONX
|
|||
|
|
local_rankqPK X ddp_backendqQNX
|
|||
|
|
tpu_num_coresqRNX tpu_metrics_debugqS‰X debugqT]qUX dataloader_drop_lastqV‰X
|
|||
|
|
eval_stepsqWNX dataloader_num_workersqXK X dataloader_prefetch_factorqYNX
|
|||
|
|
past_indexqZJÿÿÿÿX run_nameq[NX disable_tqdmq\‰X remove_unused_columnsq]ˆX label_namesq^NX load_best_model_at_endq_‰X metric_for_best_modelq`NX greater_is_betterqaNX ignore_data_skipqb‰X fsdpqc]qdX fsdp_min_num_paramsqeK X fsdp_configqf}qg(X min_num_paramsqhK X xlaqi‰X xla_fsdp_v2qj‰X xla_fsdp_grad_ckptqk‰uX" fsdp_transformer_layer_cls_to_wrapqlNX accelerator_configqmctransformers.trainer_pt_utils
|
|||
|
|
AcceleratorConfig
|
|||
|
|
qn)�qo}qp(X
|
|||
|
|
split_batchesqq‰X dispatch_batchesqrNX even_batchesqsˆX use_seedable_samplerqtˆX non_blockingqu‰X gradient_accumulation_kwargsqvNubX deepspeedqwNX label_smoothing_factorqxG X optimqyctransformers.training_args
|
|||
|
|
OptimizerNames
|
|||
|
|
qzX
|
|||
|
|
adamw_8bitq{…q|Rq}X
|
|||
|
|
optim_argsq~NX adafactorq‰X group_by_lengthq€‰X length_column_nameq�X lengthq‚X report_toqƒ]q„X ddp_find_unused_parametersq…NX ddp_bucket_cap_mbq†NX ddp_broadcast_buffersq‡NX dataloader_pin_memoryqˆˆX dataloader_persistent_workersq‰‰X skip_memory_metricsqŠˆX use_legacy_prediction_loopq‹‰X push_to_hubqŒ‰X resume_from_checkpointq�NX hub_model_idqŽNX hub_strategyq�ctransformers.trainer_utils
|
|||
|
|
HubStrategy
|