GitLab

# Conflicts:
#	lm_eval/api/task.py

* Fix the format of mgsm zh and ja.

* Add change log to mgsm.

* Add newline after changelog.

* fix(zeno): Generate unique ids in case of multiple filters

* fix(zeno): Report even non-aggregable metrics, just not as metrics

* pre-commit

---------
Co-authored-by: Baber <baber@hey.com>

* hf support load gguf file

* code review

* code review

* code clean up

* note about use_fast compat with gguf

---------
Co-authored-by: Qubitium-ModelCloud <qubitium@modelcloud.ai>

* remove yaml extension from phraes_va_common

* remove yaml extension from winogenerated

* remove yaml extension from phrases_es

* no cache debug logging when not used

* update evaluate; update construct requests

* update construct requests to handle `apply_chat_template` kwarg

upgrade transformers and peft in CI

* fix extra_match low if batch_size > 1
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* add sorting to logprobs

* nit

---------
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>
Co-authored-by: Baber <baber@hey.com>

* added aradice

* Added ArabicMMLU Lev Configs

* added ArabicMMLU egy configs

* Added boolq configs

* Added cultural bench configs

* added openbookqa configs

* Added PiQA configs

* added winogrande configs

* Added truthfulQA configs

* Added aradice group config

* Remove deleted files from repository

* modified arabimmlu configs

* modified metadata versions

* fixed formatting using ruff

* added aradice tasks information

* pre-commit

* Uptaded openbookqa utils

* fixed formatting on obqa

---------
Co-authored-by: Basel Mousi <bmousi@hbku.edu.qa>
Co-authored-by: Baber <baber@hey.com>

* add warning for truncation

* add global mmlu lite

* add global mmlu lite

* fix bugs

* add task README.md

* Update README.md

* Update tasks README.md

* Update README.md

* update readme

---------
Co-authored-by: shivi <shivalikasingh95@gmail.com>

* feat: drop Python 3.8 support

* feat: drop Python 3.8 tests

* pre-commit

forgot to increment 0.4.6!

* fix `DeprecationWarning: invalid escape sequence '\s'`

* add type hints

* Revert "add type hints"

This reverts commit 15d8abc626a84e97f8c238ddfbf9e243d6f6eb5c.

* batch all rolling token windows

* nit

* copy to vllm

* fix max_length for `get_rolling_token_windows`

* bugfix

* bugfix

* add type hints

* score readme added

* generate until task's "until" parameter's default value fixed.

* score mmlu-pro and agieval added

* changed macro accuracy to micro for agieval

* Always E removed from agi eval

* redundancies removed

* MATH added

* minor cosmetic changes for math

* Licenses added Readme updated

* changes for flake8 + license header on math

* Score added to readme and precommit was run.

* Score added to readme and precommit was run.

* Import error fixed

* math task bugfix
postprocess minor fix

* CR for math added

* math CR

* math task bugfix
postprocess minor fix

CR for math added

* Math cr fixed

* mmlu_pro non_greedy task added

* non greedy summarizer added

* Non greedy for all score tasks

* Bugfixes for non-greedy

* fixing the until argument

* undoing the change to "until" arguments default behaviour

* minor fix in summarizer

* log naming changes for better readability

* math subtasks naming fix

* agieval subtask naming fix

* logging added for debugging

* path issue fixed

* minor fix

* path fix

* path fix

* non_greedy_math minor fix

* final changes

* changed readme for non-greedy
added Nvidia header
added wxample script for non_greedy
changed prompts to match that fo trt runs

* non greedy summarizer bugfix

* non_greedy summarizer fixed

* make warning prominent

* make warning prominent

* initial support for optimum-intel ipex model. LM model as first step

* format
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* pass dtype
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* update README
Signed-off-by: Yao, Matrix <matrix.yao@intel.com>

---------
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* update import
Signed-off-by: Maanu Grover <maanug@nvidia.com>

* run formatting

---------
Signed-off-by: Maanu Grover <maanug@nvidia.com>

* left truncate for generate_until

* pre-commit

* Handle pipeline_parallel parameter

* Add description of pipeline parallelism with OV models

* Update README.md

add caching tip to readme

* Update README.md

add api link

* avoid timeout errors with high concurrency in api_model

* style

* add timeout

* add docs

---------
Co-authored-by: Baber <baber@hey.com>

Update Unitxt task to  use locally installed unitxt and not download Unitxt code from Huggingface (#2514)

* Moved to require unitxt installation and not download unitxt from HF hub.

This has performance benefits and simplifies the code.
Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Updated watsonx documentation

* Updated installation instructions

* Removed redundant comman

* Allowed unitxt tasks to generate chat APIs

Modified WatsonXI model to support chat apis

* Removed print

* Run precommit formatting

---------
Signed-off-by: Yoav Katz <katz@il.ibm.com>

* make utility function to handle `until`

* fix text

* allow !function filters

* bugfix

* nit

* add `filter` to logged samples

* add `filter` and `metric` to logged samples to identification

* convert `metric` to `metrics`: list

* score readme added

* generate until task's "until" parameter's default value fixed.

* score mmlu-pro and agieval added

* changed macro accuracy to micro for agieval

* Always E removed from agi eval

* redundancies removed

* MATH added

* minor cosmetic changes for math

* Licenses added Readme updated

* changes for flake8 + license header on math

* Score added to readme and precommit was run.

* Score added to readme and precommit was run.

* Import error fixed

* math task bugfix
postprocess minor fix

* CR for math added

* math CR

* math task bugfix
postprocess minor fix

CR for math added

* Math cr fixed

* reverting the default "until" parameter change and adjusting  score task configs

* fix test task

* dont call lm.chat_template each time

* Add metabench (Kipnis et al. 2024)

* Update metabench tasks for full replication of original benchmarks, using publicly available datasets

* Remove unnecessary import

* Add permute versions of each task, where the answer orders are randomly shuffled.

* Add metabench group for easier evaluations

* Fix mmlu counts after removing duplicate

* Add secondary datasets

* Fix f-string error

* Fix f-string error for permute processing

* Add original hash to outputs for easy matching to original results

* Add line break at end of utils files

* Remove extra line from winogrande

* Reformat for linters

* fix multiple input test

* appease pre-commit

* Add metabench to tasks README

* fix multiple input `test_doc_to_text`

---------
Co-authored-by: Baber <baber@hey.com>

* add hf mamba to mamba_lm

* fix _model_generate for hf