GitLab

remove batch size compute in loglikelihood rolling, logelikelihood token is now taking care of computing the batch size

This reverts commit d859d1ca.

* initial chat template

* tokenizer attribute check

* variable rename

* interface update

* system instruction

* system inst default update

* fewshot as multiturn

* typing update

* indent update

* added comments

* Adding a fewshot in a more readable way

* linting

* Moved apply chat template to LM

* multiturn alternation fix

* cache key update

* apply chat template method fix

* add system prompt hash to cache_key

* tokenizer name property for cache_key

* property name fix

* linting backward compatibility fix

* docs and errors update

* add documentation on adding chat template compatibility to model_guide

* fewshot as multiturn check fix

* saving system inst and chat template in results

* eval tracker update

* docs update

* Apply suggestions from code review
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

---------
Co-authored-by: haileyschoelkopf <hailey@eleuther.ai>
Co-authored-by: Clémentine Fourrier <22726840+clefourrier@users.noreply.github.com>
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* [HFLM]Add support for Ascend NPU
Co-authored-by: jiaqiw09 <jiaqiw960714@gmail.com>
Co-authored-by: zhabuye <2947436155@qq.com>

* bump accelerate dependency version to 0.26.0 for NPU compat.

---------
Co-authored-by: jiaqiw09 <jiaqiw960714@gmail.com>
Co-authored-by: zhabuye <2947436155@qq.com>
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* resize model embeddings

* resize only

* tokenizer help

* load tokenizer before model

* add comment and run precommit lint

* Add log message
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* fix auto-batch size bug for seq2seq models

* alphabetize task + group tables ; fix eval tracker bug

* fix eval tracker bug

* evaluation tracker implementation

* OVModelForCausalLM test fix

* typo fix

* moved methods args

* multiple args in one flag

* loggers moved to dedicated dir

* improved filename sanitization

* added delta weights

* removed debug

* readme update

* better error handling

* autogptq warn

* warn update

* peft and delta error, explicitly deleting _model_delta

* linter fix

* fix on --task list

* add fixes to tokeniation

* differentiate encoding for seq2seq and decoder

* return token setting

* format for pre-commit

* Seq2seq fix, pt2 (#1630)

* getting model class only when defined

* encode_pair handles None, add_special_tokens turned into dict with default value

---------
Co-authored-by: achervyakov <77295913+artemorloff@users.noreply.github.com>

* peft Version Assertion

* fix the linter issue

* make vllm use prefix_token_id ; have prefix_token_id be optional method to define

* custom_prefix_token_id wasn't set if not passed

This reverts commit b7923a84.

* use BOS token in loglikelihood

* improve comments

* add model arg

* log prefix token id

* log prefix token id

* Update lm_eval/api/model.py
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* change name to prefix_token_id

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* Differentiate _encode_pair setting for decoder and enc-dec models

* tok_decode to not skip special token so that eos doen't become empty string

* Update model.py

* Update model.py

* Update huggingface.py

* Update lm_eval/models/huggingface.py
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* Update model.py

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* add manual tqdm disabling management

* add typing to all new args

* apply precommit changes

---------
Co-authored-by: haileyschoelkopf <hailey@eleuther.ai>

* use `@ray.remote` with distributed vLLM

* update versions

* bugfix

* unpin vllm

* fix pre-commit

* added version assertion error

* Revert "added version assertion error"

This reverts commit 8041e9b78e95eea9f4f4d0dc260115ba8698e9cc.

* added version assertion for DP

* expand DP note

* add warning

* nit

* pin vllm

* fix typos

* model_type attribute error

Getting attribute error when using a model without a 'model_type'

* fix w/ and w/out the 'model_type' specification

* use getattr(), also fix other config.model_type reference

* Update huggingface.py

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* change `all_gather` to `gather`

* add TaskOutput utility class

* Add FilterResults class and refactor task handling.

* Rename `key` to `filter_key` for clarity

* Add `print_writeout` function in utils.py

* Add function to calculate limit size.

* Add doc_iterator method to Task class

* Refactor `doc_iterator` and cleanup in Task class

* remove superfluous bits

* change `all_gather` to `gather`

* bugfix

* bugfix

* fix `gather`

* Refactor `gather` loop

* Refactor aggregate metrics calculation

* Refactor and simplify aggregate metrics calculation
Removed unused code

* Simplify metrics calculation and remove unused code.

* simplify the metrics calculation in `utils.py` and `evaluator.py`.

* Fix group metric

* change evaluate to hf_evaluate

* change evaluate to hf_evaluate

* add docs

* add docs

* nits

* make isslice keyword only

* nit

* add todo

* nit

* nit

* nit: swap order samples_metrics tuple

* move instance sorting outside loop

* nit

* nit

* Add __repr__ for ConfigurableTask

* nit

* nit

* Revert "nit"

This reverts commit dab8d9977a643752a17f840fd8cf7e4b107df28f.

* fix some logging

* nit

* fix `predict_only` bug. thanks to `@LSinev`!

* change `print_tasks` to `prepare_print_tasks`

* nits

* move eval utils

* move eval utils

* nit

* add comment

* added tqdm descriptions

* Update lm_eval/evaluator_utils.py
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* fix mgsm bug

* nit

* fix `build_all_requests`

* pre-commit

* add ceil to limit

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

This reverts commit c1145dfd.

* add add_bos_token to HFLM

* add BOS token flag to other local model classes

---------
Co-authored-by: Lintang Sutawika <lintang@eleuther.ai>

* loglikelihood refactor using template lm

* linter

* fix whitespace in target + prompt for CoT gsm8k (#1275)

* Make `parallelize=True` vs. `accelerate launch` distinction clearer in docs (#1261)

* Make parallelize=True distinction clearer in documentation.

* run linter

* Allow parameter edits for registered tasks when listed in a benchmark (#1273)

* benchmark yamls allow minor edits of already registered tasks

* add documentation

* removed print

* Fix data-parallel evaluation with quantized models (#1270)

* add WIP device_map overrides

* update handling outside of accelerate launcher

* change .to(device) log to debug level

* run linter

* Rework documentation for explaining local dataset (#1284)

* rewor documentation for explaining local dataset

* fix typo

* Update new_task_guide.md

* Re-add citation

It looks like Google Scholar has [already noticed](https://scholar.google.com/scholar?hl=en&as_sdt=0%2C9&authuser=2&q=%22A+framework+for+few-shot+language+model+evaluation%2C+12+2023%22&btnG=

) the updated citation block so let's add it back in.

* Update CITATION.bib (#1285)

Bumping CITATION.bib to match re-adding the citation in readme. 

cc @StellaAthena

* Update nq_open.yaml (#1289)

* Update README.md with custom integration doc (#1298)

* Update README.md

* punctuation

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* Update nq_open.yaml (#1305)

* Update nq_open.yaml

change regex

* Bump NQ version

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* Update task_guide.md (#1306)

* Update pyproject.toml (#1312)

* Fix polemo2_in.yaml config name (#1313)

* Update pyproject.toml (#1314)

* Fix group register (#1315)

* tuple should be considered as well

* set option to keep callable as callable

* Update task_guide.md (#1316)

* Update polemo2_in.yaml (#1318)

* don't pass extra kwargs to mamba any more (#1328)

* Fix Issue regarding stderr (#1327)

* add fix fordeciding if stderr is N/A or not

* process N/A

* Add `local-completions` support using OpenAI interface (#1277)

* Add `local-completions` support using OpenAI interface

* Refactor oa_completion

* Address tokenizer comments and change request chunks to batch size

* Add warning message for tiktoken backend

* fix formatting

* fix whitespace

* Update README.md

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* fallback to classname when LM doesnt have config (#1334)

* fix a trailing whitespace that breaks a lint job (#1335)

* skip "benchmarks" in changed_tasks (#1336)

* Update migrated HF dataset paths (#1332)

* Update arc_easy.yaml

* Update flan_cot.yaml

* update HF dataset path

* Update freeform.yaml

* Update flan_cot.yaml

---------
Co-authored-by: Lintang Sutawika <lintang@eleuther.ai>

* Don't use `get_task_dict()` in task registration / initialization (#1331)

* don't use get_task_dict() as a helper, it will download the dataset!

* pre-commit

* Update README.md

---------
Co-authored-by: lintangsutawika <lintang@eleuther.ai>

* manage default (greedy) gen_kwargs in vllm (#1341)

* manage default (greedy) gen_kwargs in vllm better

* mirror HF `do_sample`

* just need to set temp=0 for greedy

* modified default gen_kwargs to work better with CLI; changed prompt_logprobs=1 (#1345)

* update links to task_guide.md (#1348)

* `Filter` docs not offset by `doc_id`  (#1349)

* get `doc` from instance

* acceletate bugfix: get ground doc from instance

* convert filter to `process_result`

* get docs from instances in `FilterEnsemble`

* rename

* nit

* better looping

* fix typehint

* Add FAQ on `lm_eval.tasks.initialize_tasks()` to README (#1330)

* Update README.md

* [!Tip]

* Refix issue regarding stderr (#1357)

* Add causalLM OpenVino models (#1290)

* added intel optimum

* added intel optimum in readme

* modified intel optimum

* modified intel optimum

* modified intel optimum

* modified install optimum

* modified path of IR file

* added openvino_device

* added openvino_device2

* changed optimum-causal to openvino-causal

* Update README.md

* Update README.md

* remove `lm_eval.base` import

* update openvino-causal -> openvino ; pass device through super().__init__()

* Update README.md

* Add optimum to tests dependencies

* apply pre-commit

* fix so tests pass

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>
Co-authored-by: haileyschoelkopf <hailey@eleuther.ai>

* Apply some best practices and guideline recommendations to code (#1363)

* raise Exception, not a string

Additional info https://peps.python.org/pep-0352/#exception-hierarchy-changes
https://docs.python.org/3.8/tutorial/errors.html#raising-exceptions

* Apply PEP8 recommendation to prefer isinstance

"Object type comparisons should always use isinstance() instead of comparing types directly"
https://peps.python.org/pep-0008/

* Remove dangerous default mutable values in arguments

https://pylint.readthedocs.io/en/stable/user_guide/messages/warning/dangerous-default-value.html

* Format logging messages with fstring (not with format)

Additional info
https://pylint.readthedocs.io/en/stable/user_guide/messages/warning/logging-format-interpolation.html
There are also discussions about the speed of formatting while logging or some unintended code executions
https://github.com/pylint-dev/pylint/issues/2395
https://stackoverflow.com/a/54368109
but at least one format (fstring one) will be used throughout the project

* Specify utf-8 encoding for `open` explicitly

If not specified, it may be supposed differently in different environments, OSes, and Python versions. See
https://peps.python.org/pep-0597/
https://docs.python.org/3.11/library/locale.html#locale.getencoding
https://docs.python.org/3.10/library/os.html#utf8-mode
https://pylint.readthedocs.io/en/stable/user_guide/messages/warning/unspecified-encoding.html

Helps also if some code from English language tasks is taken as inspiration for tasks in non-English languages.

* Use inline-ignoring comments to pass pre-commit instead of identity process

https://flake8.pycqa.org/en/3.0.1/user/ignoring-errors.html#in-line-ignoring-errors
https://www.flake8rules.com/rules/F841.html

flake8 comments are supported by ruff: https://docs.astral.sh/ruff/linter/#error-suppression



* serialize callable functions in config (#1367)

* delay filter init; remove `*args` (#1369)

* delay filter init; remove `*args`

* bugfix

* optimize

* type hint

* Fix unintuitive `--gen_kwargs` behavior (#1329)

* don't override do_sample if no value for it is passed

* Update gen_kwargs override condition

* Update huggingface.py

* Update huggingface.py

* run linters

* silence an erroneous warning

* Publish to pypi (#1194)

* publish to pypi

* lint

* Update publish.yml

* minor

* Make dependencies compatible with PyPI (#1378)

* make deps not point to github urls

* formatting

* try making PyPI only run on tag pushes

* Add support for RWKV models with World tokenizer (#1374)

* Add support for RWKV models with World tokenizer

The RWKV line of model with the World tokenizer, does not allow the padding token to be configured, and has its value preset as 0

This however fails all the "if set" checks, and would cause the tokenizer to crash.

A tokenizer class name check was added, in addition to a model type check, as there exists RWKV models which uses the neox tokenizers

* Update huggingface.py

Genericized so that this supports any RWKVWorld tokenizer, and added a fall-back for if the HF implementation name changes.

* Comply with formatting guidelines

* fix format

---------
Co-authored-by: Stella Biderman <stellabiderman@gmail.com>
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* add bypass metric (#1156)

* add bypass metric

* fixed `bypass` metric.

* add task attributes if predict_only

* add `predict_only` checks

* add docs

* added `overide_metric`, `override_config` to `Task`

* nits

* nit

* changed --predict_only to generations; nits

* nits

* nits

* change gen_kwargs warning

* add note about `--predict_only` in README.md

* added `predict_only`

* move table to bottom

* nit

* change null aggregation to bypass (conflict)

* bugfix; default `temp=0.0`

* typo

* loglikelihood refactor using template lm

* lint

* code review

* neuron optimum

* Mention TemplateLM in model_guide.md

* Update lm_eval/api/model.py

* fix linter

* fix format

* fix format

* fix format

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>
Co-authored-by: Lintang Sutawika <lintang@eleuther.ai>
Co-authored-by: Stella Biderman <stellabiderman@gmail.com>
Co-authored-by: Mark Saroufim <marksaroufim@meta.com>
Co-authored-by: Hannibal046 <38466901+Hannibal046@users.noreply.github.com>
Co-authored-by: Danielle Pintz <38207072+daniellepintz@users.noreply.github.com>
Co-authored-by: Quentin Lhoest <42851186+lhoestq@users.noreply.github.com>
Co-authored-by: kwrobel.eth <djstrong@gmail.com>
Co-authored-by: Michael Goin <michael@neuralmagic.com>
Co-authored-by: Brian Vaughan <nairbv@users.noreply.github.com>
Co-authored-by: Baber Abbasi <92168766+baberabb@users.noreply.github.com>
Co-authored-by: thnkinbtfly <70014488+thnkinbtfly@users.noreply.github.com>
Co-authored-by: NoushNabi <33136068+NoushNabi@users.noreply.github.com>
Co-authored-by: haileyschoelkopf <hailey@eleuther.ai>
Co-authored-by: LSinev <LSinev@users.noreply.github.com>
Co-authored-by: Eugene Cheah <PicoCreator@users.noreply.github.com>

* add key lookup for same contexts

* nit

* appease pre-commit

* nit

* use `expand` (in-place view) rather than `repeat`

* try mixed grouping

* add docs.

* nit

* nit

* nits

* fix tests

* Move greedy_tokens calculation out of cache loop

* nit

* nits

* add test

* nits

* fix name conflict

* fix name conflict

* chunk tensor

* move Collator

* nits/docstring

* fixup

* fixup

* group contexts only for decoders

* pre-commit

* fix `generate_until` test

* fix `generate_until` test

* Update lm_eval/models/huggingface.py
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* add docs

* nit

* add docs

* add docs

* add 'logits_cache' arg

* bugfix

---------
Co-authored-by: Hailey Schoelkopf <65563625+haileyschoelkopf@users.noreply.github.com>

* Fix watchdog timeout

* Pre-commit fix

* Timedelta

* Fixes https://github.com/EleutherAI/lm-evaluation-harness/issues/1416

Sets `do_sample = False` if `temperature == 0.0` and `do_sample = None`

* Update huggingface.py

* Update huggingface.py

making linter happy