GitLab

The OpenAI interface supports batch size as an argument to the completions API, but does not seem to support specification of this on the CLI i.e. `lm_eval --model openai-completions --batch_size 16 ...` because of a simple lack of str->int conversion.

This is confirmed by my usage and stacktrace from running `OPENAI_API_KEY=dummy lm_eval --model local-completions --tasks gsm8k --batch_size 16 --model_args model=nm-
testing/zephyr-beta-7b-gptq-g128,tokenizer_backend=huggingface,base_url=http://localhost:8000/v1`:
```
Traceback (most recent call last):
  File "/home/michael/venv/bin/lm_eval", line 8, in <module>
    sys.exit(cli_evaluate())
  File "/home/michael/code/lm-evaluation-harness/lm_eval/__main__.py", line 341, in cli_evaluate
    results = evaluator.simple_evaluate(
  File "/home/michael/code/lm-evaluation-harness/lm_eval/utils.py", line 288, in _wrapper
    return fn(*args, **kwargs)
  File "/home/michael/code/lm-evaluation-harness/lm_eval/evaluator.py", line 251, in simple_evaluate
    results = evaluate(
  File "/home/michael/code/lm-evaluation-harness/lm_eval/utils.py", line 288, in _wrapper
    return fn(*args, **kwargs)
  File "/home/michael/code/lm-evaluation-harness/lm_eval/evaluator.py", line 390, in evaluate
    resps = getattr(lm, reqtype)(cloned_reqs)
  File "/home/michael/code/lm-evaluation-harness/lm_eval/models/openai_completions.py", line 263, in generate_until
    list(sameuntil_chunks(re_ord.get_reordered(), self.batch_size)),
  File "/home/michael/code/lm-evaluation-harness/lm_eval/models/openai_completions.py", line 251, in sameuntil_chunks
    if len(ret) >= size or x[1] != lastuntil:
TypeError: '>=' not supported between instances of 'int' and 'str'
```

* add basqueglue

* add eus_exams

* add eus_proficiency

* add eus_reading

* add eus_trivia

* run pre-commit

* Integration of NeMo models into LM Evaluation Harness library

* rename nemo model as nemo_lm

* move nemo section in readme after hf section

* use self.eot_token_id in get_until()

* improve progress bar showing loglikelihood requests

* data replication or tensor/pipeline replication working fine within one node

* run pre-commit on modified files

* check whether dependencies are installed

* clarify usage of torchrun in README

* 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

* add logging of model args

* nit

* Add warnings.

* nit

* add warning

* nit

* Add task ACLUE

* fix minor bug

* fix code style

* fix code style

* 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>

* Fix eval_logger import for mmlu/_generate_configs.py

* linter

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

* Update interface.md

* fix: make caching reqs always work with accelerate launch

* remove stale task migration checklist

* remove deprecation warnings

* make informative TypeErrors for get_task_dict

* bump version metadata

* fix num_fewshot printing bug

* add fewshot value to cache key

* 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>

* New tests for CLI args

* fix spacing

* change tests for parsing

* add tests, fix parser

* remove defaults for store_true

* Link to vllm integration

* add pip install .[vllm] cmd

* add manual tqdm disabling management

* add typing to all new args

* apply precommit changes

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

* add agieval

* fix typo

* add cloze / math exactmatch agieval tasks, rename

* update exact-match agieval tasks, allow for multiple-correct answers

* add more detail to readme

* don't parse_math_answer twice

---------
Co-authored-by: Alex Bäuerle <alex@a13x.io>

* add Arabic EXAMS benchmark

* fixed the linter issue, and add more information on the readme

* Update README.md

---------
Co-authored-by: Lintang Sutawika <lintang@sutawika.com>

* Support jinja templating for "description"

* Update task_guide.md

* Update lm_eval/api/task.py

* fix format?

* whitespace errors

* fix whitespace

* fix bad variable reference

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

* update gen_kwargs in code2-text-go.yaml

* update gen_kwargs in rest code2-text

* Add compatibility for vLLM's new Logprob object

* Fix

* Update lm_eval/models/vllm_causallms.py

* fix format?

* trailing whitespace

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

Update installation commands in openai_completions.py and contributing document and, update wandb_args description (#1536)

* Update openai completions and docs/CONTRIBUTING.md

* Update wandb args description

* Update docs/interface.md

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

* Remove unused `decontamination_ngrams_path` and all mentions (still no alternative path provided)

* Fix improper import of LM and usage of evaluator in one of scripts

* update type hints in instance and task api

* raising errors in task.py instead of asserts

* Fix warnings from ruff

* raising errors in __main__.py instead of asserts

* raising errors in tasks/__init__.py instead of asserts

* raising errors in evaluator.py instead of asserts

* evaluator: update type hints and remove unused variables in code

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

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

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

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

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

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

* pre-commit induced fixes

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

update printed num-fewshot ; prevent fewshots from erroneously being used by cot which hardcodes fewshot prompt (#1502)

* init wmdp yaml file

* Add WMDP Multiple-choice

* fix linter issues

* Delete lm_eval/tasks/wmdp/_wmdp.yaml

---------
Co-authored-by: Lintang Sutawika <lintang@sutawika.com>

* Start adding eq-bench

* Start adding to yaml and utils

* Get metric working

* Add README

* Handle cases where answer is not parseable

* Deal with unparseable answers and add percent_parseable metric

* Update README