-
Hi.
When running [demo.ipynb](https://colab.research.google.com/github/callummcdougall/sae_vis/blob/master/demo.ipynb#scrollTo=Zbgc09Y8usD9) notebook
the
```python
ModuleNotFoundError …
-
I have three routers in my Lime mesh network, TP WDR 4300, TP Archer C50 V3 and V4.
All built the firmware according to the instructions.
What is the error?
Thu Dec 24 11:54:04 2020 daemon.not…
-
If using `normalize_activations` = `expected_average_only_in`, we scale activations by a scaling factor being training the SAE. Without calling `fold_activation_norm_scaling_factor()`, the SAE will no…
-
Hi @sae-park,
It seems like there is a small issue with your repo named [VBA-challenge](https://github.com/sae-park/VBA-challenge).
Currently, it is marked as a VBScript repo since more than 50% of i…
-
First of all, nice work on this fantastic and useful package.
For my work I would like to extend the functionality to allow for decoding of NMEA2000 messages when the extended data page (here calle…
-
Hey all! Working on getting deeper integration for this into Neuronpedia. Ran into a few things (not blockers) but may be good to resolve at some point:
1) When FuzzingScorer or DetectionScorer rec…
-
-
### Proposal
Ensure that the `activation_store.from_sae` method creates a store that will remove BOS activations before adding them to the buffer. Setting default seqpos slice to ignore the first …
-
For the refraction models in the Python package, I think the units should be hPa as reported in the paper and not Pa as shown in the documentation.
https://github.com/gschwind/sg2/blob/ca99a3ac35454a…
-
### Proposal
We should add an option to exclude special tokens when adding a SAE into `HookedSAETransformer`. This could take the form of an `exclude_special_tokens` param for `add_sae()` / `run_w…