Update neuron backend #2314
Merged
Update neuron backend #2314
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
dacorvo
requested review from
haileyschoelkopf,
lintangsutawika and
baberabb
as code owners
|
|
|
Hi! Thank you very much for the PR. Could you sign the CLA and run the pre-commit so it can be merged. |
dacorvo
force-pushed
the
update_neuron
branch
from
118ae95
to
8008231
Compare
The evaluation of log likelihood was not working for neuron models using continuous batching, such as all cached neuron LLama models.
dacorvo
force-pushed
the
update_neuron
branch
from
8008231
to
6570b81
Compare
|
@baberabb it should be ok now. |
dacorvo
force-pushed
the
update_neuron
branch
from
6570b81
to
54c55ca
Compare
baberabb
reviewed
Sep 18, 2024
baberabb
approved these changes
Sep 18, 2024
|
Thanks for the PR! |
jmercat
pushed a commit
to TRI-ML/lm-evaluation-harness
that referenced
this pull request
Sep 25, 2024
* feat(neuron): align with latest optimum-neuron * feat(neuron): support pre-exported neuron models * fix(neuron): correctly use max_length * fix(neuron): adapt loglikelihood The evaluation of log likelihood was not working for neuron models using continuous batching, such as all cached neuron LLama models. * refactor(neuron): remove dead code
giuliolovisotto
pushed a commit
to giuliolovisotto/lm-evaluation-harness
that referenced
this pull request
Sep 27, 2024
* feat(neuron): align with latest optimum-neuron * feat(neuron): support pre-exported neuron models * fix(neuron): correctly use max_length * fix(neuron): adapt loglikelihood The evaluation of log likelihood was not working for neuron models using continuous batching, such as all cached neuron LLama models. * refactor(neuron): remove dead code
shachardon
pushed a commit
to shachardon/lm-evaluation-harness
that referenced
this pull request
Oct 1, 2024
* feat(neuron): align with latest optimum-neuron * feat(neuron): support pre-exported neuron models * fix(neuron): correctly use max_length * fix(neuron): adapt loglikelihood The evaluation of log likelihood was not working for neuron models using continuous batching, such as all cached neuron LLama models. * refactor(neuron): remove dead code
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.
This modifies the
CustomNeuronModelForCausalLMclass to align on the latestoptimum-neuronversion and better use the underlyingNeuronModelForCausalLMclass. This in particular letsNeuronModelForCausalLMdecide the default values when exporting.In addition, this fixes the evaluation of loglikelihood for neuron models using continuous batching.
This also modifies the class initialization to allow evaluating models that have been previously exported.
Finally, this properly supports the
max_lengthparameter, allowing to select neuron model configurations that are actually cached on the Hugging Face hub.