WIP: feature(pu): adapt to unizero-multitask ddp, and adapt ppo to support jericho config #858
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…n mask), e.g. detective env
PaParaZz1
requested changes
Mar 9, 2025
| if param.grad is not None: | ||
| allreduce(param.grad.data) | ||
| else: | ||
| # 如果梯度为 None，则创建一个与 param.grad_size 相同的零张量，并执行 allreduce |
There was a problem hiding this comment.
remove commented code and add English comment, then these modifications will be merged
| # dist.init_process_group(backend=backend, rank=rank, world_size=world_size) | ||
| # TODO： | ||
| import datetime | ||
| dist.init_process_group(backend=backend, rank=rank, world_size=world_size, timeout=datetime.timedelta(seconds=60000)) |
| # if self._rank == 0: | ||
| # self._monitor = get_simple_monitor_type(self._policy.monitor_vars())(TickTime(), expire=10) | ||
|
|
||
| self._monitor = get_simple_monitor_type(self._policy.monitor_vars())(TickTime(), expire=10) |
There was a problem hiding this comment.
add an argument named only_monitor_rank0 to control the logic, defaults to True
| for k in engine.log_buffer: | ||
| engine.log_buffer[k].clear() | ||
| return | ||
| # if engine.rank != 0: |
There was a problem hiding this comment.
also pass the only_monitor_rank0 argument to the hook class
| self._global_state_encoder = nn.Identity() | ||
| elif len(global_obs_shape) == 3: | ||
| self._mixer = Mixer(agent_num, embedding_size, embedding_size, activation=activation) | ||
| self._global_state_encoder = ConvEncoder(global_obs_shape, hidden_size_list=hidden_size_list, activation=activation, norm_type='BN') |
There was a problem hiding this comment.
why BN rather than using LN as default here
| agent_state, global_state = agent_state.unsqueeze(0), global_state.unsqueeze(0) | ||
| agent_state = agent_state.unsqueeze(0) | ||
| if single_step and len(global_state.shape) == 2: | ||
| global_state = global_state.unsqueeze(0) |
| @@ -205,7 +214,10 @@ def forward(self, data: dict, single_step: bool = True) -> dict: | |||
| agent_q_act = torch.gather(agent_q, dim=-1, index=action.unsqueeze(-1)) | |||
| agent_q_act = agent_q_act.squeeze(-1) # T, B, A | |||
| if self.mixer: | |||
| global_state_embedding = self._global_state_encoder(global_state) | |||
| if len(global_state.shape) == 5: | |||
| @@ -265,12 +265,17 @@ def compute_actor(self, x: torch.Tensor) -> Dict: | |||
| >>> assert actor_outputs['logit'].shape == torch.Size([4, 64]) | |||
| """ | |||
| if self.share_encoder: | |||
| x = self.encoder(x) | |||
| # import ipdb;ipdb.set_trace() | |||
There was a problem hiding this comment.
modify the corresponding API comments, and the isinstance(x, dict) to control the logic
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Related Issue
TODO
Check List