Issues
- 8
- 7
Can not run the example of cifar.
#108 opened by yaolu-zjut - 2
FSDP error
#93 opened by sangkeun00 - 1
Metadata loading bug
#91 opened by sangkeun00 - 6
Add `flatten` option to storage config
#54 opened by sangkeun00 - 4
Diet on metadata
#77 opened by eatpk - 2
Analysis interface design
#56 opened by sangkeun00 - 1
Investigating torch.linalg float64 operations
#62 opened by pomonam - 1
- 6
Config Improvements
#76 opened by sangkeun00 - 5
- 1
Configurable CPU Offloading
#81 opened by sangkeun00 - 1
analog.eval() doesn't turn on grad=True
#75 opened by hage1005 - 1
Integration with Huggingface Trainer
#72 opened by sangkeun00 - 5
Store influence scores in DataFrame
#41 opened by sangkeun00 - 2
add_lora() sets init_strategy="random" if using initialize_from_log() with state computed by PCA
#74 opened by hage1005 - 5
Refactoring proposal for logging
#65 opened by sangkeun00 - 1
Separating `log_state` from `AnaLogState`
#63 opened by sangkeun00 - 1
Non-writable tensor warning
#64 opened by sangkeun00 - 2
LoRAEmbedding
#59 opened by hwijeen - 1
Deprecate torch.save in flush_unsafe
#24 opened by sangkeun00 - 2
It seems like the log extracting part( generation of `log_chunk_{idx}.mmap` part) gotten bit slower after `AnalogState` deployment.
#48 opened by eatpk - 1
Fall back to random initialization when PCA initialization encounters errors
#36 opened by sangkeun00 - 3
InfluenceFunction precondition function fails when logging does not set hessian=True
#26 opened by YoungseogChung - 0
call `StorageHandler.flush` only when necessary
#20 opened by hwijeen - 8
Too many open files error
#22 opened by hwijeen - 0
Fix the way modules are added
#16 opened by hwijeen - 1
Register only the necessary hooks
#17 opened by hwijeen