An NLA trained with random-length truncation of the verbalizer's output, so the most important information comes first. Checkpoints, data, baseline.
Smitty PRO
syvb
·
AI & ML interests
None yet
Recent Activity
updated a model about 1 hour ago
syvb/oracle-lens-qwen3-8b-recon updated a dataset about 22 hours ago
syvb/oracle-lens-qwen3-8b-artifacts published a model about 22 hours ago
syvb/oracle-lens-qwen3-8b-oracle-sftOrganizations
None yet
An NLA trained with random-length truncation of the verbalizer's output, so the most important information comes first. Checkpoints, data, baseline.
16 repeated injection markers w/ per-slot learned affine vs 1-slot control. Warm-started from nanonla-qwen3-8b-L24-av.
v2 item-truncation matryoshka NLA (Qwen2.5-7B L20): warm-start AV verbalizer + AR critic checkpoints, warm-start data, base NLA.