-
Notifications
You must be signed in to change notification settings - Fork 51
Issues
is:issue state:open
is:issue state:open
Issue creation is restricted in this repository
Search results
Stage-4 deferred: JAX↔torch persistent-adversary / source parity (tracker)
parityVPD JAX↔torch parity trackingVPD JAX↔torch parity trackingstage-4Deferred full-parity feature (backlog; revisit later)Deferred full-parity feature (backlog; revisit later)subsystem:adversaryParity subsystem: adversaryParity subsystem: adversaryStatus: Open.#918 In goodfire-ai/param-decomp;Investigate correlation between norm and frequency.
exp/investigationA research investigation or experimentA research investigation or experimentPriority: nInUnot Important & not Urgentnot Important & not UrgentStatus: Open.#266 In goodfire-ai/param-decomp;Investigate different options for splitting up target model matrices
exp/investigationA research investigation or experimentA research investigation or experimentPriority: nInUnot Important & not Urgentnot Important & not UrgentStatus: Open.#243 In goodfire-ai/param-decomp;What's with the outlier highly active component?
exp/investigationA research investigation or experimentA research investigation or experimentgood first issueGood for newcomersGood for newcomersPriority: InUImportant & not UrgentImportant & not UrgentStatus: Open.#242 In goodfire-ai/param-decomp;Remove pydantic pins when wandb fixes warnings issue
devopsdependency/testing/development related issuesdependency/testing/development related issuesPriority: nInUnot Important & not Urgentnot Important & not UrgentStatus: Open.#233 In goodfire-ai/param-decomp;Create a component model config
code-qualityimproving code quality/consistencyimproving code quality/consistencyfeatureNew feature or requestNew feature or requestPriority: nInUnot Important & not Urgentnot Important & not UrgentStatus: Open.#174 In goodfire-ai/param-decomp;Is p-annealing just increasing the importance minimality loss?
exp/investigationA research investigation or experimentA research investigation or experimentPriority: InUImportant & not UrgentImportant & not UrgentStatus: Open.#144 In goodfire-ai/param-decomp;Toy model: A multi-layer model doing Computation in Superposition
exp/investigationA research investigation or experimentA research investigation or experimentPriority: InUImportant & not UrgentImportant & not Urgenttoy-modelAdding a new toy modelAdding a new toy modelStatus: Open.#20 In goodfire-ai/param-decomp;Toy model: Add Tracr models
featureNew feature or requestNew feature or requestPriority: InUImportant & not UrgentImportant & not Urgenttoy-modelAdding a new toy modelAdding a new toy modelStatus: Open.#19 In goodfire-ai/param-decomp;Toy model: LM/Simplestories with embed/unembed only
exp/investigationA research investigation or experimentA research investigation or experimentPriority: nInUnot Important & not Urgentnot Important & not Urgenttoy-modelAdding a new toy modelAdding a new toy modelStatus: Open.#18 In goodfire-ai/param-decomp;Removing layer privileging in the importance loss
exp/investigationA research investigation or experimentA research investigation or experimentfeatureNew feature or requestNew feature or requestPriority: InUImportant & not UrgentImportant & not UrgentStatus: Open.#8 In goodfire-ai/param-decomp;Investigating whether SPD is stable to training with a subset of the data
exp/investigationA research investigation or experimentA research investigation or experimentPriority: InUImportant & not UrgentImportant & not UrgentStatus: Open.#4 In goodfire-ai/param-decomp;