locuslab/tofu issues and pull requests

#47 - Request for the original author profiles of TOFU dataset

Issue - State: open - Opened by the-jb about 1 month ago

#46 - Upload Llama Unlearning Checkpoints

Issue - State: open - Opened by shariqahn about 2 months ago - 8 comments

#45 - Unusual Model Utility Gap: Gradient Difference vs Ascent (Llama2)

Issue - State: open - Opened by jeongjin0 2 months ago - 1 comment

#44 - Is the dpo loss wrong?

Issue - State: closed - Opened by HaomingX 2 months ago - 3 comments

#43 - Which dataset should we use for evaluate?

Issue - State: open - Opened by Yuda-Jin 5 months ago - 1 comment

#42 - RuntimeError: 'weight' must be 2-D During Fine-Tuning with Single GPU

Issue - State: open - Opened by ouerwt 5 months ago - 1 comment

#41 - Can not find 'adapter_config.json' in ckpt or huggingface

Issue - State: closed - Opened by Yuda-Jin 6 months ago - 2 comments

#40 - Raise error while evaluate.

Issue - State: closed - Opened by Yuda-Jin 6 months ago - 5 comments

#39 - Could you please provide finetuned model weight of phi-1.5 and lamma2, this will unify the basis of our research.

Issue - State: closed - Opened by Yuda-Jin 6 months ago - 1 comment

#38 - DeepSpeed Zero-3 is not compatible with `low_cpu_mem_usage=True` or with passing a `device_map`

Issue - State: open - Opened by zhmzm 6 months ago - 1 comment

#37 - Why finetuned model and retained model have similar model utility?

Issue - State: closed - Opened by Carol-gutianle 7 months ago - 2 comments

#36 - About the deepspeed

Issue - State: closed - Opened by LetheSec 8 months ago - 3 comments

#35 - Error loading Phi Finetuned model

Issue - State: closed - Opened by pomonam 8 months ago - 2 comments

#34 - Added support for LLaMa3

Pull Request - State: closed - Opened by mikeFore4 8 months ago - 1 comment

#33 - Inquiry about Constructing Datasets with Elaborate Prompt

Issue - State: closed - Opened by tbozhong 9 months ago - 1 comment

#32 - Inconsistent number of forget samples when evaluating the retain model (forget10 task)

Issue - State: closed - Opened by LetheSec 9 months ago - 2 comments

#31 - Breaking change in Huggingface Phi-1-5?

Issue - State: closed - Opened by ajyl 9 months ago - 1 comment

#30 - Bug in calculating loss when using DPO

Issue - State: closed - Opened by zeta-zl 9 months ago - 1 comment

#29 - Trying to get model parallelism and lower precision working

Pull Request - State: closed - Opened by molereddy 10 months ago

#28 - Finetuning with LORA causes DeepSpeed error

Pull Request - State: closed - Opened by mikeFore4 10 months ago - 1 comment

#27 - Fixing order of directory creation for forget.py to prevent exiting

Pull Request - State: closed - Opened by mikeFore4 10 months ago - 1 comment

#26 - The implementation of Truth Ratio and Probability is different from the definition in the paper

Issue - State: open - Opened by wzunknown 10 months ago - 13 comments

#25 - Support for different num_processes in interleave_eval_result_dict

Issue - State: closed - Opened by molereddy 10 months ago - 1 comment

#24 - question about retain_perturbed.json in datasets locuslab/TOFU

Issue - State: open - Opened by wtma1999 10 months ago - 1 comment

#23 - TOFU-finetuned Phi-1.5 is not on the huggingface page

Issue - State: closed - Opened by molereddy 10 months ago - 1 comment

#22 - The huggingface leaderboard page is showing Runtime Error

Issue - State: open - Opened by wzunknown 10 months ago

#21 - changed bf16 to fp16 and fixed some model paths

Pull Request - State: closed - Opened by akshayneema 10 months ago

#20 - One of the inputs missing for DPO loss

Issue - State: open - Opened by chrisliu298 10 months ago - 4 comments

#19 - Getting truth ratio always 1

Issue - State: open - Opened by shaswati1 10 months ago - 13 comments

#18 - Dataset contents issues

Issue - State: open - Opened by molereddy 10 months ago - 4 comments

#17 - Eval log file limiting examples

Issue - State: closed - Opened by molereddy 10 months ago - 1 comment

#16 - eval generates answer same as dataset

Issue - State: closed - Opened by shaswati1 11 months ago - 9 comments

#15 - Refactor eval

Pull Request - State: closed - Opened by zhilif 11 months ago

#15 - Refactor eval

Pull Request - State: closed - Opened by zhilif 11 months ago

#14 - Issues introduced by refactoring and other miscellaneous

Issue - State: open - Opened by molereddy 11 months ago - 9 comments

#13 - Refactor eval

Pull Request - State: closed - Opened by pratyushmaini 11 months ago

#12 - End to generated text

Issue - State: closed - Opened by molereddy 11 months ago - 5 comments

#11 - Hyperparameter issues in configs

Issue - State: closed - Opened by molereddy 11 months ago - 3 comments

#10 - Unable to train fintuned LoRA on forget

Issue - State: open - Opened by shaswati1 11 months ago - 3 comments

#9 - Where is eval_log_aggregated.json generated?

Issue - State: closed - Opened by molereddy 11 months ago - 1 comment

#8 - Issues with deepspeed

Issue - State: closed - Opened by molereddy 11 months ago - 3 comments

#7 - Unable to save finetuned llama2

Issue - State: closed - Opened by shaswati1 11 months ago - 8 comments

#6 - requirements.txt needs a fix

Issue - State: closed - Opened by molereddy 12 months ago - 2 comments

#5 - Finetune LLAMA2 with LoRA

Issue - State: closed - Opened by petezone 12 months ago - 1 comment

#4 - Is anyone getting a problem with the command for forget.py?

Issue - State: closed - Opened by sriramvema 12 months ago - 3 comments

#3 - Where are the evals inside the data folder being generated?

Issue - State: closed - Opened by rthapa84 12 months ago - 5 comments

#2 - Command for evaluations

Issue - State: closed - Opened by yujianll about 1 year ago - 2 comments

#1 - Unable to load the dataset from HuggingFace hub, throws a ValueError

Issue - State: closed - Opened by archit31uniyal about 1 year ago - 1 comment

GitHub / locuslab/tofu issues and pull requests