[FSDP()][9/N] Refactor ctor (continued) #87923
Closed
Conversation
This file contains hidden or 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
[ghstack-poisoned]
awgu
requested review from
H-Huang,
kwen2501,
mrshenli,
pritamdamania87,
rohan-varma and
zhaojuanmao
as code owners
pytorch-bot
bot
added
the
release notes: distributed (fsdp)
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/87923
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Failures, 3 PendingAs of commit 5a5cd73: The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Oct 27, 2022
[ghstack-poisoned]
[ghstack-poisoned]
This PR makes a second pass over the constructor. The logic has been grouped into `_init_<...>` functions based on intent (e.g. `_init_prefetching_state()` or `_init_runtime_state()`). This makes the initialization code for composable FSDP much cleaner than having to re-write the same sequences of lower-level helper calls. This PR also moves `_ExecOrderData` into its own file `_exec_order_utils.py`. [ghstack-poisoned]
This PR makes a second pass over the constructor. The logic has been grouped into `_init_<...>` functions based on intent (e.g. `_init_prefetching_state()` or `_init_runtime_state()`). This makes the initialization code for composable FSDP much cleaner than having to re-write the same sequences of lower-level helper calls. This PR also moves `_ExecOrderData` into its own file `_exec_order_utils.py`. [ghstack-poisoned]
This was referenced Oct 31, 2022
awgu
pushed a commit
to awgu/pytorch
that referenced
this pull request
Oct 31, 2022
ghstack-source-id: fa1f6c7 Pull Request resolved: pytorch#87923
This PR makes a second pass over the constructor. The logic has been grouped into `_init_<...>` functions based on intent (e.g. `_init_prefetching_state()` or `_init_runtime_state()`). This makes the initialization code for composable FSDP much cleaner than having to re-write the same sequences of lower-level helper calls. This PR also moves `_ExecOrderData` into its own file `_exec_order_utils.py`. [ghstack-poisoned]
awgu
pushed a commit
to awgu/pytorch
that referenced
this pull request
Oct 31, 2022
ghstack-source-id: 1b0ac6e Pull Request resolved: pytorch#87923
zhaojuanmao
reviewed
Nov 1, 2022
| FSDP_SYNCED = "_fsdp_synced" | ||
|
|
||
| # TODO (awgu): Refactor this later | ||
| SHARDING_STRATEGY_MAP = { |
Contributor
There was a problem hiding this comment.
curious why it needs this map if the key and value are the same?
Collaborator
Author
There was a problem hiding this comment.
Key is public facing ShardingStrategy. Value is private/internal HandleShardingStrategy. I think we should be able to refactor later to only use the public facing ShardingStrategy. I had two separate ones to avoid the circular import issues a while ago, but now that ShardingStrategy is in its own file api.py, hopefully this should be resolved.
Contributor
|
ideally we still want to run the basic FSDP benchmarks to sanity check no crash, no regression before landing the PRs as it has huge changes to the core codes |
This was referenced Nov 1, 2022
kulinseth
pushed a commit
to kulinseth/pytorch
that referenced
this pull request
Nov 5, 2022
This PR makes a second pass over the constructor. The logic has been grouped into `_init_<...>` functions based on intent (e.g. `_init_prefetching_state()` or `_init_runtime_state()`). This makes the initialization code for composable FSDP much cleaner than having to re-write the same sequences of lower-level helper calls. This PR also moves `_ExecOrderData` into its own file `_exec_order_utils.py`. Pull Request resolved: pytorch#87923 Approved by: https://github.com/mrshenli
kulinseth
pushed a commit
to kulinseth/pytorch
that referenced
this pull request
Dec 10, 2022
This PR makes a second pass over the constructor. The logic has been grouped into `_init_<...>` functions based on intent (e.g. `_init_prefetching_state()` or `_init_runtime_state()`). This makes the initialization code for composable FSDP much cleaner than having to re-write the same sequences of lower-level helper calls. This PR also moves `_ExecOrderData` into its own file `_exec_order_utils.py`. Pull Request resolved: pytorch#87923 Approved by: https://github.com/mrshenli
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.
Stack from ghstack:
unflat_param_name->fqnfor consistency #88123 [FSDP] Renameunflat_param_name->fqnfor consistency_get_buffer_names()#88122 [FSDP] Simplify_get_buffer_names()torch.no_grad()context when offloading to CPU #88121 [FSDP] Remove unneededtorch.no_grad()context when offloading to CPU_lazy_init()into_fsdp_root_pre_forward()#87941 [FSDP()][26/N] Move_lazy_init()into_fsdp_root_pre_forward()_post_forward_reshard()#87940 [FSDP()][25/N] Add_post_forward_reshard()_lazy_init()#87939 [FSDP()][24/N] Refactor_lazy_init()_reset_lazy_init()#87937 [FSDP] Simplify_reset_lazy_init()_cast_buffers()in_lazy_init()#87936 [FSDP()][22/N] Refactor_cast_buffers()in_lazy_init()_cast_buffers()#87935 [FSDP()][21/N] Refactor_buffer_name_to_orig_dtypecomputationdtypetobuffer_name_to_dtype#87934 [FSDP] Renamedtypetobuffer_name_to_dtypedevicearg from_cast_buffers()#87933 [FSDP] Removedevicearg from_cast_buffers()pre_forward_unshard()#87931 [FSDP()][18/N] Refactorpre_forward_unshard()_fsdp_root_pre_forward()#87930 [FSDP()][17/N] Refactor_fsdp_root_pre_forward()_init_streams()#87928 [FSDP()][15/N] Refactor_init_streams()This PR makes a second pass over the constructor. The logic has been grouped into
_init_<...>functions based on intent (e.g._init_prefetching_state()or_init_runtime_state()). This makes the initialization code for composable FSDP much cleaner than having to re-write the same sequences of lower-level helper calls.This PR also moves
_ExecOrderDatainto its own file_exec_order_utils.py.