ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update rl_replacements.py

Browse source
This commit is contained in:
Daniel Han 2025-02-14 04:53:06 -08:00
commit 407011c7ba
1 changed files with 5 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

15
unsloth/models/rl_replacements.py
View file

@ -196,7 +196,7 @@ RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer__get_per_token_logps)
# Custom compiled GRPO loss - creates 3 Triton kernels
@torch.compile(dynamic = True, fullgraph = True, options = torch_compile_options,)
def _grpo_compute_loss(old_logits, new_logits, input_ids, mask, beta):
def grpo_compute_loss(old_logits, new_logits, input_ids, mask, beta, advantages):
old_logits = old_logits.to(torch.float32)
new_logits = new_logits.to(torch.float32)
input_ids = input_ids.unsqueeze(-1)
@ -224,11 +224,6 @@ def _grpo_compute_loss(old_logits, new_logits, input_ids, mask, beta):
pass
return loss, completion_length, mean_kl
pass
def grpo_compute_loss(old_logits, new_logits, input_ids, mask, beta):
loss, completion_length, mean_kl = _grpo_compute_loss(old_logits, new_logits, input_ids, mask, beta)
return loss, completion_length.item(), mean_kl.item()
pass
RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource((_grpo_compute_loss)))
RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource((grpo_compute_loss)))
@ -247,7 +242,7 @@ def grpo_trainer_compute_loss(function_name, function):
# attention_mask = torch.cat([prompt_mask, completion_mask], dim=1)
attention_mask = None
logits_to_keep = completion_ids.size(1) # we only need to compute the logits for the completion tokens
per_token_logps = self._get_per_token_logps(model, input_ids, attention_mask, logits_to_keep)
# Compute the KL divergence between the model and the reference model
@ -261,15 +256,15 @@ def grpo_trainer_compute_loss(function_name, function):
# loss = ((per_token_loss * completion_mask).sum(dim=1) / completion_mask.sum(dim=1)).mean()
input_ids = input_ids[:, -logits_to_keep:]
loss, completion_length, mean_kl = grpo_compute_loss(
ref_per_token_logps, per_token_logps, input_ids, completion_mask, self.beta,
ref_per_token_logps, per_token_logps, input_ids, completion_mask, self.beta, advantages,
)
# Log the metrics
# completion_length = self.accelerator.gather_for_metrics(completion_mask.sum(1)).float().mean().item()
self._metrics["completion_length"].append(completion_length)
self._metrics["completion_length"].append(completion_length.item())
# mean_kl = ((per_token_kl * completion_mask).sum(dim=1) / completion_mask.sum(dim=1)).mean()
# self._metrics["kl"].append(self.accelerator.gather_for_metrics(mean_kl).mean().item())
self._metrics["kl"].append(mean_kl)
self._metrics["kl"].append(mean_kl.item())
return loss
pass