Enhancements#55

Open

thiswillbeyourgithub wants to merge 26 commits intovgel:mainfrom thiswillbeyourgithub:enhancements

+438-40

Commits on Dec 4, 2024

add file settings.py to store settings
thiswillbeyourgithub
committed
add a utils.py file that contains a make_dataset function
thiswillbeyourgithub
committed
move DatasetEntry to utils.py
thiswillbeyourgithub
committed
feat: give more controls to how the user can select the layers to modify
thiswillbeyourgithub
committed
feat: add joblib memory for caching model activations
thiswillbeyourgithub
committed
fix: numpy v2 compatibility
thiswillbeyourgithub
committed
feat: add arg norm_type when training that allows setting l1 or l2 (or other) norm
thiswillbeyourgithub
committed
feat: add function to accept chat template as inputs + autocorrect
thiswillbeyourgithub
committed
fix: forgot an import
thiswillbeyourgithub
committed
minor: add tqdm for getting activations and applying the new directions
thiswillbeyourgithub
committed
perf: potentially faster one liner to tokenize
thiswillbeyourgithub
committed
feat: use flag LOW_MEMORY to reduce the amount of memory needed when storing the arrays before computing directions
thiswillbeyourgithub
committed
fix: imports
thiswillbeyourgithub
committed
update default model in example from mistral 7B v0.1 to v0.3
thiswillbeyourgithub
committed
docs: add more details on how to load the model, including quantization, avoiding OOM, etc
thiswillbeyourgithub
committed
docs: use chat messages in example
thiswillbeyourgithub
committed
docs: in example, show how to login for models that require it
thiswillbeyourgithub
committed
docs: add missing declaration of tokenizer
thiswillbeyourgithub
committed
fix: in example, move the tokens directly to the correct device
thiswillbeyourgithub
committed
minor: changed default in the example
thiswillbeyourgithub
committed
docs: add link to github issue about OOM for gguf files
thiswillbeyourgithub
committed
docs: add more examples of models
thiswillbeyourgithub
committed
fix: autocorrect chat templates
thiswillbeyourgithub
committed
docs: mention the new way to specify the layers
thiswillbeyourgithub
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Enhancements#55

Commits on Dec 4, 2024

add file settings.py to store settings

add a utils.py file that contains a make_dataset function

move DatasetEntry to utils.py

feat: give more controls to how the user can select the layers to modify

feat: add joblib memory for caching model activations

fix: numpy v2 compatibility

feat: add arg norm_type when training that allows setting l1 or l2 (or other) norm

feat: add function to accept chat template as inputs + autocorrect

fix: forgot an import

minor: add tqdm for getting activations and applying the new directions

perf: potentially faster one liner to tokenize

feat: use flag LOW_MEMORY to reduce the amount of memory needed when storing the arrays before computing directions

fix: imports

update default model in example from mistral 7B v0.1 to v0.3

docs: add more details on how to load the model, including quantization, avoiding OOM, etc

docs: use chat messages in example

docs: in example, show how to login for models that require it

docs: add missing declaration of tokenizer

fix: in example, move the tokens directly to the correct device

minor: changed default in the example

docs: add link to github issue about OOM for gguf files

docs: add more examples of models

fix: autocorrect chat templates

docs: mention the new way to specify the layers

Commits on Dec 11, 2024

fix: template function for non chats

docs: mention in readme that qwen2.5 7B works out of the box