@@ -391,25 +391,25 @@ def parse_args(args=None):
391391 rft_parser .add_argument ("--base-model" , help = "Base model resource id" )
392392 rft_parser .add_argument ("--warm-start-from" , help = "Addon model to warm start from" )
393393 rft_parser .add_argument ("--output-model" , help = "Output model id (defaults from evaluator)" )
394- rft_parser .add_argument ("--epochs" , type = int )
395- rft_parser .add_argument ("--batch-size" , type = int )
396- rft_parser .add_argument ("--learning-rate" , type = float )
397- rft_parser .add_argument ("--max-context-length" , type = int )
398- rft_parser .add_argument ("--lora-rank" , type = int )
399- rft_parser .add_argument ("--accelerator-count" , type = int )
394+ rft_parser .add_argument ("--epochs" , type = int , default = 8 )
395+ rft_parser .add_argument ("--batch-size" , type = int , default = 128000 )
396+ rft_parser .add_argument ("--learning-rate" , type = float , default = 3e-5 )
397+ rft_parser .add_argument ("--max-context-length" , type = int , default = 65536 )
398+ rft_parser .add_argument ("--lora-rank" , type = int , default = 16 )
399+ rft_parser .add_argument ("--accelerator-count" , type = int , default = 1 )
400400 rft_parser .add_argument ("--region" , help = "Fireworks region enum value" )
401401 rft_parser .add_argument ("--display-name" , help = "RFT job display name" )
402402 rft_parser .add_argument ("--evaluation-dataset" , help = "Optional separate eval dataset id" )
403403 rft_parser .add_argument ("--eval-auto-carveout" , dest = "eval_auto_carveout" , action = "store_true" , default = True )
404404 rft_parser .add_argument ("--no-eval-auto-carveout" , dest = "eval_auto_carveout" , action = "store_false" )
405405 # Rollout chunking
406- rft_parser .add_argument ("--chunk-size" , type = int , help = "Data chunk size for rollout batching" )
406+ rft_parser .add_argument ("--chunk-size" , type = int , default = 10 , help = "Data chunk size for rollout batching" )
407407 # Inference params
408408 rft_parser .add_argument ("--temperature" , type = float )
409409 rft_parser .add_argument ("--top-p" , type = float )
410410 rft_parser .add_argument ("--top-k" , type = int )
411- rft_parser .add_argument ("--max-tokens" , type = int )
412- rft_parser .add_argument ("--n" , type = int )
411+ rft_parser .add_argument ("--max-tokens" , type = int , default = 32768 )
412+ rft_parser .add_argument ("--n" , type = int , default = 8 )
413413 rft_parser .add_argument ("--inference-extra-body" , help = "JSON string for extra inference params" )
414414 # Wandb
415415 rft_parser .add_argument ("--wandb-enabled" , action = "store_true" )
0 commit comments