GitLab

* feat(vllm_causallms): make collator ignore seed when splitting batch into chunks

* fix(collator): revert PR changes

* fix(vllm-causallm): update collator call with groupby None

* feat(sglang-causallms): make generation accept a list of sampling params

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

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

* Update vLLM compatibility
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

* add TokensPrompt to all generate calls

---------
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: Baber <baber@hey.com>

Device has been a deprecated arg for a few releases of vLLM and is now removed in 0.10.0 https://github.com/vllm-project/vllm/pull/21349

* added support for additional chat template arguments

* use `enable_thinking`

* add wrap logging function

* add `chat_template_args` back to HF

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

* feat: add postprocessing for generated text to strip stop sequences and thinking tokens

* nit

* fix: trim leading whitespace after stripping thinking tokens from generation

* feat: add think_end_token to model_args

* nit

* nit

* nit

* add to readme

* nit

* use all answers

* use middle truncation

* maybe fix classification score

* strip classification preds

* [vllm] remove stop tokens post-hoc

* strip all preds

* pacify pre-commit

* start on truncation utility

* add to readme

* add a footgun doc

* fix newline in yaml templates

* do not strip code_sim preds!

* fix pre-commit config

* fix instruction warning

* add not to longbench readme

* add data_parallel for V1

* use Process instead of Queue

* ray used if V0 DP

* better error handling

* fix truncation warning comparison

* fix arguments

* pacify pre-commit

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

* add `sglang-generate`

* nit

* nit

* nit

* pacify pre-commit

* fix: pass device arg in model_ar in vllm_causallms

* casting device arg to str in vLLM model args

* fix resolve_hf_chat_template version

* pre-commit

* Add support for chat templates defined outside of tokenizer_config.json, as supported by vLLM

* Update template name to avoid conflict with other variable

* remove ray.remote resources

* remove kobtest tag (registered as group)

* changed source of eval_logger

* allow eval_logger to be set from args

* removed verbosity arg from non-main methods

* fix logging

* pre-commit

* set verbosity in eval logger

* replace utils.eval_logger

* fix logging in main

* add logging to docs

* add logging message

* nit

* add logging to docs

* refactor setup_logging to utils

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

* fix vllm

* fix data_parallel

* copy to multimodal

* update pre-commit

* add assistant prefix

* add arc_challenge from llama

* nit

* nit

* nit

* add assistant prefix

* add mmlu_llama

* nit

* nit

* Revert "nit"

This reverts commit 6a97f8356237305e375212b966b30e8de59dd4bc.

* fix regex bug

* add assistant_prefix to vllm

* add `Question:`

* add mmlu_pro

* add fewshot assistant_prefix

* use `assistant_prefill`

* typehints

* nits

* nits

* add to docs

* add readme

* batch all rolling token windows

* nit

* copy to vllm

* fix max_length for `get_rolling_token_windows`

* bugfix

* bugfix

* add type hints

* make utility function to handle `until`

* fix text

* fix: use lora_request for data parallel vllm evals

* fix(docs): include type hint

* chore: lint, et pre-commit al

---------
Co-authored-by: Chris Kerwell Gresla <chris@wafer.systems>

* Replace generic exception classes with a more specific ones

* rerun pre-commit to pass linter tests

* Revert "rerun pre-commit to pass linter tests"

This reverts commit 67f88ccf144469853217704520e613196042d859.

* reduce repetitions in errors or so

* Replace generic exception class with a more specific one

* default chat template method fix

* move chat_template to TemplateLM

* remove hotfix

* handle openai `chat_template`

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

* add 'max_tokens' to gen_kwargs

* pre-commit

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

* max_length - 1 (generation always >= 1)

* vllm: fix rolling prefix_token

* nit: add comment

* fixup! max_length should be handled for logliklihoods

* fix revision type

* allow for None-input loglikelihood reqs to be cached

* handle no remaining cache items

* pre-commit

* change cache_hook.add_partial(loglikelihood_rolling...) convention

---------
Co-authored-by: Baber Abbasi <baber@eleuther.ai>