This commit is contained in:
hiyouga 2024-06-06 00:50:32 +08:00
parent 83a005e3d4
commit 67fe822324
3 changed files with 13 additions and 20 deletions

View File

@ -2,7 +2,7 @@ transformers>=4.41.2
datasets>=2.16.0 datasets>=2.16.0
accelerate>=0.30.1 accelerate>=0.30.1
peft>=0.11.1 peft>=0.11.1
trl>=0.8.6 trl>=0.9.3
gradio>=4.0.0 gradio>=4.0.0
scipy scipy
einops einops

View File

@ -65,7 +65,7 @@ def check_dependencies() -> None:
require_version("datasets>=2.16.0", "To fix: pip install datasets>=2.16.0") require_version("datasets>=2.16.0", "To fix: pip install datasets>=2.16.0")
require_version("accelerate>=0.30.1", "To fix: pip install accelerate>=0.30.1") require_version("accelerate>=0.30.1", "To fix: pip install accelerate>=0.30.1")
require_version("peft>=0.11.1", "To fix: pip install peft>=0.11.1") require_version("peft>=0.11.1", "To fix: pip install peft>=0.11.1")
require_version("trl>=0.8.6", "To fix: pip install trl>=0.8.6") require_version("trl>=0.9.3", "To fix: pip install trl>=0.9.3")
def count_parameters(model: torch.nn.Module) -> Tuple[int, int]: def count_parameters(model: torch.nn.Module) -> Tuple[int, int]:

View File

@ -93,18 +93,6 @@ class CustomDPOTrainer(DPOTrainer):
output_dir = output_dir if output_dir is not None else self.args.output_dir output_dir = output_dir if output_dir is not None else self.args.output_dir
getattr(self.processor, "image_processor").save_pretrained(output_dir) getattr(self.processor, "image_processor").save_pretrained(output_dir)
def sft_loss(self, batch: Dict[str, "torch.Tensor"], chosen_logits: "torch.FloatTensor") -> "torch.Tensor":
r"""
Computes supervised cross-entropy loss of given labels under the given logits.
Returns:
A tensor of shape (batch_size,) containing the cross-entropy loss of each samples.
"""
batch_size = batch["input_ids"].size(0) // 2
chosen_labels, _ = batch["labels"].split(batch_size, dim=0)
chosen_logps = self.get_batch_logps(chosen_logits, chosen_labels, average_log_prob=True)
return -chosen_logps
def odds_ratio_loss(self, chosen_logps: "torch.Tensor", rejected_logps: "torch.Tensor") -> "torch.Tensor": def odds_ratio_loss(self, chosen_logps: "torch.Tensor", rejected_logps: "torch.Tensor") -> "torch.Tensor":
r""" r"""
Computes ORPO's odds ratio (OR) loss for batched log probabilities of the policy model. Computes ORPO's odds ratio (OR) loss for batched log probabilities of the policy model.
@ -156,9 +144,9 @@ class CustomDPOTrainer(DPOTrainer):
def concatenated_forward( def concatenated_forward(
self, model: "PreTrainedModel", batch: Dict[str, "torch.Tensor"] self, model: "PreTrainedModel", batch: Dict[str, "torch.Tensor"]
) -> Tuple["torch.Tensor", "torch.Tensor", "torch.Tensor", "torch.Tensor"]: ) -> Tuple["torch.Tensor", "torch.Tensor", "torch.Tensor", "torch.Tensor", "torch.Tensor"]:
r""" r"""
Computes the sum log probabilities of the labels under the given logits if loss_type != IPO. Computes the sum log probabilities of the labels under given logits if loss_type is not IPO, ORPO or SimPO.
Otherwise the average log probabilities. Otherwise the average log probabilities.
""" """
@ -167,17 +155,20 @@ class CustomDPOTrainer(DPOTrainer):
all_logits: "torch.Tensor" = model(**batch, return_dict=True, use_cache=False).logits.to(torch.float32) all_logits: "torch.Tensor" = model(**batch, return_dict=True, use_cache=False).logits.to(torch.float32)
all_logps = self.get_batch_logps( all_logps, valid_length = self.get_batch_logps(
logits=all_logits, logits=all_logits,
labels=batch["labels"], labels=batch["labels"],
average_log_prob=(self.loss_type in ["ipo", "orpo", "simpo"]),
is_encoder_decoder=self.is_encoder_decoder, is_encoder_decoder=self.is_encoder_decoder,
label_pad_token_id=self.label_pad_token_id, label_pad_token_id=self.label_pad_token_id,
) )
if self.loss_type in ["ipo", "orpo", "simpo"]:
all_logps = all_logps / valid_length
batch_size = batch["input_ids"].size(0) // 2 batch_size = batch["input_ids"].size(0) // 2
chosen_logps, rejected_logps = all_logps.split(batch_size, dim=0) chosen_logps, rejected_logps = all_logps.split(batch_size, dim=0)
chosen_logits, rejected_logits = all_logits.split(batch_size, dim=0) chosen_logits, rejected_logits = all_logits.split(batch_size, dim=0)
return chosen_logps, rejected_logps, chosen_logits, rejected_logits chosen_length, _ = valid_length.split(batch_size, dim=0)
return chosen_logps, rejected_logps, chosen_logits, rejected_logits, chosen_logps / chosen_length
def compute_reference_log_probs( def compute_reference_log_probs(
self, model: "PreTrainedModel", batch: Dict[str, "torch.Tensor"] self, model: "PreTrainedModel", batch: Dict[str, "torch.Tensor"]
@ -201,6 +192,7 @@ class CustomDPOTrainer(DPOTrainer):
reference_rejected_logps, reference_rejected_logps,
_, _,
_, _,
_,
) = self.concatenated_forward(ref_model, batch) ) = self.concatenated_forward(ref_model, batch)
return reference_chosen_logps, reference_rejected_logps return reference_chosen_logps, reference_rejected_logps
@ -220,6 +212,7 @@ class CustomDPOTrainer(DPOTrainer):
policy_rejected_logps, policy_rejected_logps,
policy_chosen_logits, policy_chosen_logits,
policy_rejected_logits, policy_rejected_logits,
policy_chosen_logps_avg,
) = self.concatenated_forward(model, batch) ) = self.concatenated_forward(model, batch)
reference_chosen_logps, reference_rejected_logps = self.compute_reference_log_probs(model, batch) reference_chosen_logps, reference_rejected_logps = self.compute_reference_log_probs(model, batch)
@ -229,7 +222,7 @@ class CustomDPOTrainer(DPOTrainer):
reference_chosen_logps, reference_chosen_logps,
reference_rejected_logps, reference_rejected_logps,
) )
sft_loss = self.sft_loss(batch, policy_chosen_logits) # compute chosen_logps with masks sft_loss = -policy_chosen_logps_avg
if self.ftx_gamma > 1e-6: if self.ftx_gamma > 1e-6:
losses += self.ftx_gamma * sft_loss losses += self.ftx_gamma * sft_loss