online_training synced with main - #1221
Merged
Merged
Conversation
* add examples and minor updates * update gsm8k dataset example notebook * Update doc/source/notebooks/models/load_model.ipynb Co-authored-by: Artem Kozhevnikov <kozhevnikov.artem@gmail.com> * add example * update model notebook * nit * format * format --------- Co-authored-by: Artem Kozhevnikov <kozhevnikov.artem@gmail.com>
* Compiles and passes pytest when disabled * Working implementation missing save/restore * All passing tests * Simplify thread pool class * Revert minor test change * Python lint * Fix C++ missing include * C++ lint * adding extra test * Lint * Add TBB task_arena as default thread pool implementation * Remove useless non local * isort * Format C++ --------- Co-authored-by: Artem Kozhevnikov <kozhevnikov.artem@gmail.com> Co-authored-by: Artyom Kozhevnikov <artyomko@fb.com>
* move configs * move stopes utils * move stopes utils * fix ci * add dependencies for arrow * add configs * update parquet utils * add parquet transform fns * add draft pipeline * add pytest fixture for pq * cleanup * cleanup * fix ci * try fix ci * try fix ci * try fix ci * fix pipeline ci * revise pipeline * cleanup dependencies * remove parquet_fragments_to_pipeline_builder * cleanup non-generic transform * cleanup * cleanup * after rebase * rebase continue * fixing tests * first config split * more options * first restructuring commit * loading code * tests on loading * tests loading * more tests params * bucketing * some tests for bucketing * fixing imports * some minor fixes * ff * ff * typo * trace flatten * parsing manifest * fixing primary direction * black * pq_mt_dataset debug * caching mt directions per split * typo * small order change * moving functions * more moving * moving between transform and utils * fixing tests * adding extra test * rm joblib * more readme * comment * fixing integration tests and adding some basic integration * typos * to_device -> to * fix * filters directly in fragment loading * typo * cannot filter directly if adding traces * rm unused e * better assert * mypy * generic length column * nit * add doc * nit * nit * update readme * typo * typos * doc update * hf integration notebook example * update readme * add sphinx doc * nit * update doc * update doc * adding non_deterministic_read options * flake8 --------- Co-authored-by: Artyom Kozhevnikov <artyomko@fb.com> Co-authored-by: Paul-Ambroise Duquenne <duquenne.pa@gmail.com> Co-authored-by: Artem Kozhevnikov <kozhevnikov.artem@gmail.com> Co-authored-by: artyomko <artyomko@etinyclues.com>
facebook-github-bot
added
the
CLA Signed
Removed duplicate parameter in DPO configuration which otherwise leads to an error.
Trace for visibility:
```
> fairseq2 lm instruction_finetune $OUTPUT_DIR --config-file $CONFIG_FILE
(...)
File "/Users/aerben/miniconda3/envs/fairseq2/lib/python3.12/site-packages/ruamel/yaml/constructor.py", line 281, in check_mapping_key
raise DuplicateKeyError(*args)
ruamel.yaml.constructor.DuplicateKeyError: while constructing a mapping
in "temp.config", line 3, column 9
found duplicate key "max_seq_len" with value "4096" (original value: "4096")
in "temp.config", line 6, column 9
```
* add parquet based text dataloader with packing, plus minor refactors * missing file * using stric_state=False to avoid big buffer serialization * mypy * annotations * flake 8 * pinned_memory * comments * rm recipes * rm npc=1 * stricter split selection * adding split pattern * copy --------- Co-authored-by: Artyom Kozhevnikov <artyomko@fb.com>
* Pairwise J1 prompt * Adding Pairwise-J1 support * Minor changes * Simplifying * Add logging back in * Add generation prompt. * removing debug statements * More logging for scores out of range * Adding reward name as a reward class attribute * Cleaning up generative judges with extractor classes * Typing * logger label --------- Co-authored-by: swarna <swarnadeep@meta.com> Co-authored-by: Ilia Kulikov <kulikov@meta.com>
* add gener_verifier for hf backend * working? with HF background * fix bug * comments * working with vllm now * add generative_judge * merge * remove GeneralVerifier * move wrap_text to extractor * rename wrap_text --> format_prompt * string * remove comment * comment * comments * comment
* bug fix * new parser * fix * fix parsing * dont parse ref answer * remove unused
Co-authored-by: Kevin Heffernan <kevinheffernan@meta.com>
* add optional regex customization * look up the regex expression in base cards as well * type annotation
* create the adafactor optimizer * lint * further lint * Add missing docstring
…cipes (#1228) * chat template overwriting, chat mode in SFT dataset * qwen template added * formatting * dpo recipe with chat mode * extra newline fix in the template * introducing chat template interface * lint * lint * lint * removing v0.4 compatibility --------- Co-authored-by: Ilia Kulikov <kulikov@meta.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
14 participants
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.
What does this PR do? Please describe:
Many changes that reflect main updates, major ones:
this way we support qwen and other archs available in fairseq2 and huggignface for online training now