GetGenetica/jearx-gemma4-e2b-p07-l4-l13-lora-aligned

P07 L4/L13 KAN-Aware Adapter Target

This repository is a GetGenetica-owned metadata shell for the P07 train-first adapter target. It intentionally does not include adapter weights yet.

The current state is:

• Target repo: GetGenetica/jearx-gemma4-e2b-p07-l4-l13-lora-aligned
• Required weight file: adapter_model.safetensors
• Expected hidden dimension: 1536
• Observed adapter hidden dimension: None
• Adapter weight state: blocked_missing_adapter_model_safetensors
• Merge allowed today: False

Canonical Target Modules

• model.language_model.layers.4.self_attn.q_proj
• model.language_model.layers.4.self_attn.v_proj
• model.language_model.layers.13.self_attn.q_proj
• model.language_model.layers.13.self_attn.v_proj

Current Attach Blockers

• missing_adapter_model.safetensors
• missing_observed_dim=1536
• missing_same_checkpoint_task_ablation

Claim Boundary

This package proves only that the target repo, target layers, target modules, and safetensors policy are explicit. It is not a trained adapter, not a model merge, not a same-checkpoint task ablation, and not an uplift claim.

Next Required Evidence

1. Train the UB11 L4/L13 adapter against the declared base checkpoint.
2. Save adapter_model.safetensors only, with no legacy .bin, .pt, or .pth weight artifacts.
3. Verify observed adapter dimension is 1536.
4. Verify the safetensors header contains LoRA A/B tensor keys for the exact L4/L13 q/v modules above.
5. Run same-checkpoint paired task ablation before any promotion or public quality claim.