Skip to content

retain origin dtype in compute log prob for megatron backend #3651

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

retain origin dtype in compute log prob for megatron backend #3651

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 6 additions & 6 deletions verl/workers/actor/megatron_actor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -229,7 +229,7 @@ def compute_logprobs_fn(output, data, use_dynamic_bsz=False, indices=None):
log_probs = [o[0]["log_probs"] for o in output["output"]] # (bs, seq_size)
else:
log_probs = [o["log_probs"] for o in output["output"]] # (bs, seq_size)
log_probs = torch.cat(log_probs, dim=0).to(torch.float32)
log_probs = torch.cat(log_probs, dim=0)
if use_dynamic_bsz:
indices = output["indices"]
indices = list(itertools.chain.from_iterable(indices))
Expand All @@ -238,7 +238,7 @@ def compute_logprobs_fn(output, data, use_dynamic_bsz=False, indices=None):
log_probs = log_probs[revert_indices]
else:
log_probs = torch.empty(
size=(batch_size, response_length), dtype=torch.float32, device=input_ids.device
size=(batch_size, response_length), device=input_ids.device
)
log_probs = log_probs.to(get_device_id())
# broadcast across pp ranks
Expand All @@ -253,7 +253,7 @@ def compute_logprobs_fn(output, data, use_dynamic_bsz=False, indices=None):
# Note that o[0] is metrics, o[1] is entropy
if mpu.is_pipeline_last_stage(ignore_virtual=True):
entropys = torch.cat([o[1] for o in output["output"]], dim=0)
entropys = entropys.to(torch.float32)
entropys = entropys
if use_dynamic_bsz:
indices = output["indices"]
indices = list(itertools.chain.from_iterable(indices))
Expand All @@ -262,7 +262,7 @@ def compute_logprobs_fn(output, data, use_dynamic_bsz=False, indices=None):
entropys = entropys[revert_indices]
else:
entropys = torch.empty(
size=(batch_size, response_length), dtype=torch.float32, device=input_ids.device
size=(batch_size, response_length), device=input_ids.device
)
# broadcast across pp ranks
entropys = entropys.to(get_device_id())
Expand Down Expand Up @@ -295,10 +295,10 @@ def make_minibatch_iterator(self, data: DataProto) -> Iterable[DataProto]:
``responses``: tensor of shape [batch_size, response_length]. torch.int64. Note that
responses = input_ids[:, -response_length:]

``old_log_probs``: tensor of shape [batch_size, response_length]. torch.float32. The log probability
``old_log_probs``: tensor of shape [batch_size, response_length]. The log probability
of responses.

``advantages``: tensor of shape [batch_size, response_length]. torch.float32. The advantages of
``advantages``: tensor of shape [batch_size, response_length]. The advantages of
responses.
See PPO paper for details. https://arxiv.org/abs/1707.06347

Expand Down