Switch adapters slightly faster #353
Merged
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
borzunov
reviewed
Jul 14, 2023
src/petals/server/backend.py
Outdated
| @@ -17,6 +17,7 @@ | |||
| from petals.server.memory_cache import MemoryCache | |||
| from petals.server.task_pool import PrioritizedTaskPool | |||
| from petals.utils.misc import is_dummy | |||
| from petals.utils.peft import using_global_adapter | |||
There was a problem hiding this comment.
You can't import it here, this triggers bnb import
borzunov
reviewed
Jul 14, 2023
src/petals/server/handler.py
Outdated
| @@ -142,6 +144,7 @@ async def rpc_inference( | |||
| requested_backends = tuple(self.module_backends[uid] for uid in requested_uids) | |||
| max_length = metadata.get("max_length") | |||
| active_adapter = metadata.get("active_adapter", "") | |||
| assert not active_adapter or active_adapter in self.adapters | |||
There was a problem hiding this comment.
Please add a error message here, otherwise the client will receive P2PHandlerError("") and won't know what happened
borzunov
reviewed
Jul 14, 2023
src/petals/utils/peft.py
Outdated
Comment on lines
125
to
126
| ADAPTER_NOT_SET = "__ADAPTER_NOT_SET" | ||
| GLOBAL_ACTIVE_ADAPTER = ADAPTER_NOT_SET |
There was a problem hiding this comment.
Suggested change
| ADAPTER_NOT_SET = "__ADAPTER_NOT_SET" | |
| GLOBAL_ACTIVE_ADAPTER = ADAPTER_NOT_SET | |
| global_active_adapter = None |
GLOBAL_ACTIVE_ADAPTERis not a constant, it should not be in caps- Why use a predefined string constant instead of just
None?
There was a problem hiding this comment.
- ok
- None means no adapter, ADAPTER_NOT_SET means you didn't even bother entering the context and you should be chastized for that
borzunov
approved these changes
Jul 14, 2023
borzunov
changed the title
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.
Currently, each
TransformerBackend.inference_steplooks for adapters and sets the correct adapter type for each block. This is not very expensive, but it can measurably affect inference time:This pull request uses faster adapter switching with just one variable assignment, without iterating over block.modules().