Support Cache Class for New Versions of Transformers Library #1341
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
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
force-pushed
the
export-D62408520
branch
from
9e412b8 to
3c1fc7d
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 9, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
3c1fc7d to
8b93cf0
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 9, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
8b93cf0 to
8600914
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 9, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
8600914 to
189676a
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 9, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
189676a to
4897dad
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 10, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
4897dad to
ef91a25
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 10, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
ef91a25 to
f52946e
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
added a commit
to craymichael/captum
that referenced
this pull request
Sep 10, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
craymichael
force-pushed
the
export-D62408520
branch
from
f52946e to
6bfe388
Compare
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D62408520 |
craymichael
force-pushed
the
export-D62408520
branch
from
6bfe388 to
bb7213c
Compare
craymichael
pushed a commit
to craymichael/captum
that referenced
this pull request
Sep 10, 2024
…torch#1341) Summary: Pull Request resolved: meta-pytorch#1341 Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for `transformers` models here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742. Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a `_supports_cache_class` flag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility. (minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior). Differential Revision: D62408520
Contributor
|
This pull request has been merged in 7b22550. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Summary:
Fixes D62210529 (now reverted by D62262760). Transformers library is now an optional dependency. We do not depend on it, however, we have some logic for
transformersmodels here. The library will only be imported if a model already has the library in the corresponding environment. This TARGETS configuration prevents transformers version conflicts which e.g. caused T200877742.Add support for new transformers Cache objects. This may need changes in the future as it seems that LLMs handle Caching differently. Some handle Caching themselves, however, some of them do not and some of them don't support Caches yet. Llama models seem to have a
_supports_cache_classflag that indicates whether this new Cache object is supported. If it isn't marked as supported, we assume it takes legacy format (tuple past values). Multiple checks added to ensure compatibility.(minor) Also, changed the defaults for LLM generation to dismiss warnings (does not change generation behavior).
Differential Revision: D62408520