Lora #207

mrudat-iais · 2024-08-05T06:47:48Z

What does this PR do?

This PR ..

General Changes

..

Breaking Changes

..

Checklist before submitting final PR

My PR is minimal and addresses one issue in isolation
I have merged the latest version of the target branch into this feature branch
I have reviewed my own code w.r.t. correct implementation, missing type hints, proper documentation, etc.
I have run a sample config for model training
I have checked that all tests run through (python tests/tests.py)

…y stopping of generation

…y. Unit test still needed.

…data. Change symbol for special tokens, which are actaully a single token within the vocab.

…mentation.

…ft_with_main

…enerator

Co-authored-by: Alexander Weber <[email protected]>

SFT sample generator

…script for end-to-end test

… int.

# Conflicts: # src/modalities/__main__.py # src/modalities/dataloader/create_packed_data.py # src/modalities/dataloader/dataset.py # src/modalities/dataloader/dataset_factory.py # src/modalities/models/gpt2/collator.py # src/modalities/registry/components.py

# Conflicts: # src/modalities/__main__.py # src/modalities/dataloader/create_packed_data.py # src/modalities/dataloader/dataset.py # src/modalities/dataloader/dataset_factory.py # src/modalities/models/gpt2/collator.py

…n checkpointing

- include_to_loss_tokens are now adjusted to special tokens in tokenizer - general path for checkpoints - save only last 5 checkpoints instead of all - test config to run training - test to run training

…for lora training.

rrutmann and others added 30 commits July 15, 2024 13:18

Add example config for the construction of chat templates

59f0191

chore: add chat template config based on jinja2

8b60a83

chore: update chat template config based on jinja2

ba2f65c

chore: Add apply chat template feature with role mapping

47e71c3

chore: extend to multiple chat templates

3303147

fix: data driven chat tempalte key retrieval

0c6bbf5

chore: Add 'index' to output JSONL

32f5756

fix: Add süecical token to be kept during treinaing to allow for earl…

482f7af

…y stopping of generation

chore: Update output file

1d72770

build: Add jsonlines dependency

0bd9bfa

chore: integration of collator wrapper with loss masking functionalit…

ed2f4ce

…y. Unit test still needed.

chore: Use SFT config replaction with uuid as file pair identification.

6e24ea2

chore: Add loss masking test

6e716b4

chore: Merge branch 'main' into sft_loss_masking

a0376a6

fix: copy raw config file for truly original content

70dc498

chore: add pbin file for testing loss masking

242e429

chore: add pbin file with more data for testing loss masking

bddcf8b

chore: use a hash not uuid for showing which config belongs to whoch …

f86b6ed

…data. Change symbol for special tokens, which are actaully a single token within the vocab.

chore: add pbin file for testing loss masking

7632a02

chore: Fix loss masking when starting within an assistant answer

15719a3

chore: add lost collator wrappr again

ab0f34c

chore: fix the loss masking test and the implementation. Improve docu…

0a545ca

…mentation.

chore: Merge commit '15ed069beaa2c83dcd15b087e4d0864b1aec4caa' into s…

c6d0a61

…ft_with_main

chore: Merge branch 'sft_with_main' into sft_loss_masking

0c52856

feat(sft): Do not reuse last targets for Instruction Tuning

12c74bc

Merge remote-tracking branch 'origin/sft_with_main' into sft_sample_g…

fc7bec1

…enerator

refactor(sft): Make reuse_last_target optional

25fdcd7

docs: Correct spelling

01109e2

Update comment

7148e1e

Co-authored-by: Alexander Weber <[email protected]>

Merge pull request #193 from Modalities/sft_sample_generator

75611dd

SFT sample generator

mrudat-iais and others added 30 commits September 9, 2024 11:30

refactor(lora): image for readme changed

b148354

refactor(lora): image for readme changed

3879de3

refactor(lora): image for readme changed

8026eb0

refactor(lora): image for readme changed

53c2ab0

refactor(lora): image for readme changed

5b76bf6

test(sft): Use special tokens already existing in tokenizers vocabulary

6594633

test(sft): Add data and config for tests

66f0bea

refactor(lora_tests): Added type hints.

f2acd7a

test(sft): Add documentation for test

0daec5b

chore: Pass black check

125311f

chore: Merge remote-tracking branch 'origin/main' into sft_with_main

9cbc8f3

feat(lora_tests): Adapted the tests for new lora config. Added a new …

2e88e70

…script for end-to-end test

fix(LORA): Type of kernel_size was checked. If tuple, convert it into…

97cbe18

… int.

fix(tests): Adapted the test config.

ccc2cd8

Merge branch 'sft_with_main' into lora

9b29ec4

# Conflicts: # src/modalities/__main__.py # src/modalities/dataloader/create_packed_data.py # src/modalities/dataloader/dataset.py # src/modalities/dataloader/dataset_factory.py # src/modalities/models/gpt2/collator.py

fix: Load huggingface models with use_cache=False to enable activatio…

34a4c45

…n checkpointing

chore: Add activation checkpointing to lora-config

471d1f4

chore: Merge branch 'lora' of github.com:Modalities/modalities into lora

885ad62

refactor(lora): optimized imports

a0a903b

Merge remote-tracking branch 'origin/lora' into lora

dd047e1

feat(lora): Added Lora to components.md

9e76b01

Update README.md

16e4008

refactor(lora): readme changed

2e103a5

Merge remote-tracking branch 'origin/lora' into lora

2a2cc6f

test: Remove empty val dataset

c1a292d

feat(lora): e2e test

0d22826

- include_to_loss_tokens are now adjusted to special tokens in tokenizer - general path for checkpoints - save only last 5 checkpoints instead of all - test config to run training - test to run training

feat(lora): wip: run e2e test on two GPUs

23c542d

feat(lora): Adapted the end-to-end test for lora training.

d12bc98

chore(lora): Added some comments and linting for the end-to-end test …

1bb3be6

…for lora training.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Lora #207

Lora #207

mrudat-iais commented Aug 5, 2024

Lora #207

Are you sure you want to change the base?

Lora #207

Conversation

mrudat-iais commented Aug 5, 2024

What does this PR do?

General Changes

Breaking Changes

Checklist before submitting final PR