20 days ago
-
QLoRA training kernel for Qwen2.5-7B-Instruct on a T4
-
security model doc
-
Synthetic validation data
-
Synthetic training data (messages format)
-
Developer docs + quickstart
-
Security self-test (6/6 passing)
-
QLoRA training kernel for Qwen2.5-7B-Instruct on a T4
-
Synthetic data generator (reasoning, self-edit, security-refusal, correction rows)
-
The security engine: whitelist, locked sections, caps, rate limit, hash-chained audit, rollback
-
initial commit
