zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-06-27 11:21:32 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

niits

Browse Source
This commit is contained in:
Goekdeniz-Guelmez 2025-01-22 21:39:29 +01:00
parent 4098c3bd2f
commit 0bb001121e
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llms/mlx_lm/lora.py
View File

@ -331,7 +331,7 @@ def evaluate_model(args, model: nn.Module, tokenizer: TokenizerWrapper, test_set
delta=args.delta, delta=args.delta,
loss_type=args.dpo_loss_type, loss_type=args.dpo_loss_type,
) )
print(f"Test loss {test_loss:.3f}, Rewards: {test_rewards[0]:.3f}, {test_rewards[1]:.3f}") print(f"Test loss {test_loss:.8f}, Rewards: {test_rewards[0]:.3f}, {test_rewards[1]:.3f}")
elif args.training_mode == "orpo": elif args.training_mode == "orpo":
test_loss, test_rewards = evaluate_orpo( test_loss, test_rewards = evaluate_orpo(
model=model, model=model,
@ -343,7 +343,7 @@ def evaluate_model(args, model: nn.Module, tokenizer: TokenizerWrapper, test_set
beta=args.beta, beta=args.beta,
reward_scaling=args.reward_scaling, reward_scaling=args.reward_scaling,
) )
print(f"Test loss {test_loss:.3f}, Rewards: {test_rewards[0]:.3f}, {test_rewards[1]:.3f}") print(f"Test loss {test_loss:.8f}, Rewards: {test_rewards[0]:.3f}, {test_rewards[1]:.3f}")
else: else:
test_loss = evaluate( test_loss = evaluate(
model=model, model=model,

4
llms/mlx_lm/tuner/orpo_trainer.py
View File

@ -312,7 +312,7 @@ def train_orpo(
if rank == 0: if rank == 0:
print( print(
f"Iter {it}: " f"Iter {it}: "
f"Val loss {val_loss:.3f}, " f"Val loss {val_loss:.8f}, "
f"Val chosen reward {val_rewards[0]:.3f}, " f"Val chosen reward {val_rewards[0]:.3f}, "
f"Val rejected reward {val_rewards[1]:.3f}, " f"Val rejected reward {val_rewards[1]:.3f}, "
f"Val took {val_time:.3f}s", f"Val took {val_time:.3f}s",
@ -353,7 +353,7 @@ def train_orpo(
if rank == 0: if rank == 0:
print( print(
f"Iter {it}: Train loss {train_loss:.3f}, " f"Iter {it}: Train loss {train_loss:.8f}, "
f"Chosen reward {train_rewards[0]:.3f}, " f"Chosen reward {train_rewards[0]:.3f}, "
f"Rejected reward {train_rewards[1]:.3f}, " f"Rejected reward {train_rewards[1]:.3f}, "
f"Learning Rate {learning_rate:.3e}, " f"Learning Rate {learning_rate:.3e}, "