Repository navigation
Default to Abliterated Large v2 (served via the TEE gateway) - #41
Merged
Merged
Conversation
The gateway now serves abliteration/abliterated-model-large-v2 (tee-gateway#183 / chat-api#446): an open model with refusals removed from the weights (not steered past), 1M context. Add it to the scrubber catalog/allow-list, make it the shipped default, and update the model line-up in the README and the catalog test.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The TEE gateway now serves
abliteration/abliterated-model-large-v2(tee-gateway#183 adds the abliteration provider + registers the model, chat-api#446 adds the catalog entry, both merged and deployed). It is an open model with the refusal directions removed from the weights, not steered past, with a 1M-token context, positioned for authorized offensive-security, red-team and long-horizon agent work. The gateway proxies to the provider inside the enclave, so ghost reaches it over the same OHTTP/TEE/verify path as every other model, with no third-party key stored locally.Changes
privacy/scrubbing_proxy.py: add the model to_CATALOG_MODELS, the single source of truth for the/modelpicker and the bridge allow-list_ALLOWED_GATEWAY_MODELS. No steer entry (uncensored by weights). Refresh the supported-model comment + catalogupdated_at.profile/config.yaml: make it the defaultmodelandopengradient.default_model. DeepSeek V4 Pro and Hermes 4 405B/70B stay selectable via/model.tests/test_scrub.py: add the id to the open-weight allow-list assertion.README.md: new default + model line-up row.Verification
abliteration/abliterated-model-large-v2returns content withopengradient_verification.verified = true(served modelabliterated-model-large-v2); a closed model (claude-opus-4-8) is still rejected by the allow-list.pytest tests/: 32 passed, includingtest_catalog_is_open_weight_only. The 3 failures are the presidio golden-canary tests (presidio-analyzernot installed in my local env); identical 3 failures onmain, and CI installs presidio.