metadata
license: mit
TopK SAE trained on layer 15 of gpt-oss-20b as part of Goodfire's internal gpt-oss hackathon.
Note: this was trained for a hackathon — it's quick+hacky and may not be 'production-grade'!
license: mit
TopK SAE trained on layer 15 of gpt-oss-20b as part of Goodfire's internal gpt-oss hackathon.
Note: this was trained for a hackathon — it's quick+hacky and may not be 'production-grade'!