pyreft

pyreft copied to clipboard

[P0] Multigpu and model sharding

2

**Descriptions:** `pyvene` library was designed for model interpretability, not for some production use case which requires training and inference efficiency. `pyreft` is different. It will have some practical use cases,...

frankaging

I would appreciate support for the OpenAI clip model, Note that PEFT also still has issues with OpenAI CLIP model support. https://github.com/huggingface/peft/issues/761 https://openai.com/research/clip

sailfish009

[P1] How to attend to memorized intervention?

2

When memorizing a sequence (1D intervention) is it possible to attend to it, as in 'where is GO-> located' (Stanford).? I'd be interested in using pyreft for 'online-learning' similar to...

chris-aeviator

[P1] Compatibility with tooling that expects a HF transformer model

3

I'm raising the issue that in terms of "production readyness" (statet goal) pyreft, designed as a very thoughtful library, will need to work together with tooling that expects a loadable...

chris-aeviator

[P1] Getting key error in parameter while training REFT using LLAMA3

8

code: import torch import transformers from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments import pyreft from huggingface_hub import login login(token="***") model_name_or_path = "meta-llama/Meta-Llama-3-8B" device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') model =...

AkashGhosh

[P1] Eval time model is not loaded: Unable to replicate results from paper for RoBERTa Base for Glue tasks like CoLa

50

Using below configuration but unable to replicate paper results. Is there anything different that the authors have done in the paper? Got {'validation_matthews_correlation': 0.40104291315665774} finally instead of ~61%. Should SEED,...

m-dev12

[P1] Cannot load trained model anymore - "type must be tuple of ints,but got NoneType"

9

After updating pyreft recently I'm encountering errors when loading a trained model. This applies to newly trained models as well as prev. trained models. I'm loading from disk. the error...

chris-aeviator

From @Jemoka when trying to save/load a bert. ``` File ~/Documents/Projects/dropval/playground/dropval/trainers/reft.py:213, in ReFTrainer.load(self, path) 210 del model.config.__dict__["use_cache"] 211 model = model.train() --> 213 self.model = pyreft.ReftModel.load( 214 str(Path(path)/"intervention"), 215 model...

aryamanarora

[P1] Intervention Locations more than Prefix and Suffix

5

Does the backbone pyvene support more than two intervention blocks on one layer? I met `anaconda3/envs/reft_train/lib/python3.10/site-packages/pyvene/models/intervenable_base.py", line 1092, in _wait_for_forward_with_parallel_intervention unit_locations_base[ IndexError: list index out of range` when I tried...

comeandcode

Much simplified codebase of ReFT-VLBart implementation. Note: transformers 4.33.1 is required, to be compatible with the Bart model version used by this branch. Therefore, this ReFT version needs to talk...

PinetreePantry