v0.0.8

Full Changelog: v0.0.7...v0.0.8

v0.0.7

What's Changed

• [P1] Addressing issues in DPO Training (#127) by @AmirZur in #128
• Debug subspace composition notebook impl by @PinetreePantry in #130
• Check in negated reft notebook by @PinetreePantry in #131
• [Minor] Update notebook with newer names (#132) by @frankaging in #133
• [Minor] fix undefined var by @frankaging in #134

Full Changelog: v0.0.6...v0.0.7

v0.0.6: Minor updates and bug fix

What's Changed

• [Minor] Update README with an example. by @frankaging in #55
• [Minor] Update README with Colab links. by @frankaging in #56
• [Minor] Update README by @frankaging in #57
• [Major] Support Llama3 models by @frankaging in #64
• [Minor] More refactory to support Llama3 experiments by @frankaging in #74
• [Minor] fix subspace (#72) by @frankaging in #75
• ReFT + DPO Tutorial by @AmirZur in #76
• Title: Fix: Shape Mismatch during Left Padding Adjustment in compute_metrics (Generated by Ana - AI SDE) by @ana-ai-sde in #89
• [P1] support ReFT+PEFT by using ReftModel to wrap PeftModel (#46) by @frankaging in #93
• [Minor] Enable lora with loreft training by @frankaging in #94
• [Minor] Basic support of quantization by @frankaging in #100
• [P0] Revert back to ortho init as unstable training by @frankaging in #103
• Fix: datasets.exceptions.DatasetNotFoundError when training with alpaca_data_cleaned by @savadikarc in #108
• [P0] Fixing LoReFT rotation layer hot loading problem (#114) by @frankaging in #123

New Contributors

• @AmirZur made their first contribution in #76
• @ana-ai-sde made their first contribution in #89
• @savadikarc made their first contribution in #108

Full Changelog: v0.0.5...v0.0.6

v0.0.5: Zeta version release

What's Changed

• Fix GitHub links to standfordnlp in the README files by @bbrowning in #39
• Update README.md by @Vikrant-Khedkar in #47
• Update chat_model.ipynb by @Vikrant-Khedkar in #48
• Fix loading IntervenableModel for its subclasses by @PinetreePantry in #49
• [Major] Zeta version by @aryamanarora in #44

New Contributors

• @bbrowning made their first contribution in #39
• @Vikrant-Khedkar made their first contribution in #47

Full Changelog: v0.0.4...v0.0.5

v0.0.4

What's Changed

• [P0] Fixing the requirements for Kaggle and Google notebooks env by @frankaging in #38

Full Changelog: v0.0.3...v0.0.4

v0.0.3

Fix requirements.

Full Changelog: v0.0.2...v0.0.3

v0.0.2: Cleanup

Full Changelog: v0.0.1...v0.0.2

v0.0.1: initial release

What's Changed

• separate data processing to data.py by @aryamanarora in #1
• Initial attempt to adapt to HF trainer by @aryamanarora in #2
• add weight decay param by @aryamanarora in #3
• [Bug fix] fix layer parsing step after dataset creation and others by @frankaging in #4
• separate argparse from training fxn by @aryamanarora in #8
• change math and commonsense to LLM adaptor template by @frankaging in #7
• Adding in stsb support by @frankaging in #10
• add an option for normalized input; GLUE in training HF eval by @frankaging in #11
• gsm8k splits by @aryamanarora in #12
• sharing interventions across positions by @frankaging in #13
• fix padding on intervention locations by @aryamanarora in #15
• more update on the padding thing with gsm8k and others by @frankaging in #16
• add gd option by @frankaging in #17
• adjust decode stra by @frankaging in #18
• Zen/gsm8k by @frankaging in #19
• minor fix by @frankaging in #20
• move generation args to config file by @aryamanarora in #21
• Update README.md by @PinetreePantry in #22
• Verified README code, fix a bug preventing proper save and load by @PinetreePantry in #23
• Update README.md by @eltociear in #32

New Contributors

• @aryamanarora made their first contribution in #1
• @frankaging made their first contribution in #4
• @PinetreePantry made their first contribution in #22
• @eltociear made their first contribution in #32

Full Changelog: https://github.com/stanfordnlp/pyreft/commits/v0.0.1