DGV-TC-055L2: Component & ProvenanceGER-203-WEIGHT-TAMPERVersion: 1.0.0
LLM Weight Integrity Across Nodes
Specific test for model weights in distributed inference, verifying weight hashes dynamically before matrix multiplication.
Operational Purpose & Scope
Halts inference if layer weights fail hash check in VRAM.
Verification Context
- Target Scope:
Model weights, distributed inference. - Audit Strategy:
exact-match - Verification Metrics:
weight_hash_verification_time - Framework Origin:DGV v0.5.0 Governance Extensions
- Mandatory:Yes — must pass for certification
Governance Risk & Penalty
Weight divergence across nodes produces inconsistent inference results, breaking determinism guarantees.
Expected Outcome
- verdict:
"hard_deny" - action:
"halt_inference"
Explore this card in 3D
Open DGV-TC-055 in the interactive scenario replay. This illustrates the test flow; it does not run the verifier or generate a cryptographic receipt.
Open DGV Live →