PVL CPU Semantic Router & Invariant Interceptor

Low-Power Negentropy Routing Engine (MiniLM-L6-v2)

Preventative Value Labs (PVL) Protocol


1. Overview & Thermodynamic Role

In the PVL protocol, the CPU Semantic Router acts as the first line of defense before any high-power GPU inference is triggered:

  • Sub-Millisecond Routing on CPU: Consumes <15W–30W CPU energy (fractions of a Joule) rather than spinning up a 300W+ GPU for raw task triage.
  • 4-Aspect Task Decomposition: Decomposes queries into Intent, Symptom, Surface, and Cure to eliminate semantic vector smearing.
  • Reasoning Synthetic Memories (RSM): Matches incoming tasks against a 12,193-entry specification database of formal C/C++ memory safety rules, compiler flags, and EVM invariants.
  • Prefill Negentropy: Injects the matched invariant specifications directly into Cranberry-12B's KV-cache via parallel prefill, preventing wasteful autoregressive decode loops.
  • CPU Router Reimbursement: Every routing operation tracks physical CPU Joules and timestamps, reimbursed directly through user escrow under Section 10 of .agents/AGENTS.md.

2. Directory Structure & Included Files

release/pvl-router/
β”œβ”€β”€ minilm_l6_v2/                 # MiniLM-L6-v2 Transformer Model (86.7 MB)
β”‚   β”œβ”€β”€ model.safetensors         # 22.7M parameter transformer weights
β”‚   β”œβ”€β”€ config.json               # Model architecture config
β”‚   β”œβ”€β”€ tokenizer.json            # WordPiece tokenizer
β”‚   β”œβ”€β”€ tokenizer_config.json     # Tokenizer settings
β”‚   β”œβ”€β”€ vocab.txt                 # Vocabulary dictionary
β”‚   └── special_tokens_map.json   # Special tokens mapping
β”œβ”€β”€ vector_store/                 # PVL 12,193-Invariant Specification DB (35.6 MB)
β”‚   β”œβ”€β”€ db_embeddings.pt          # Pre-computed 384-dim MiniLM tensor embeddings
β”‚   β”œβ”€β”€ db_keys.json              # Invariant lookup identifiers
β”‚   β”œβ”€β”€ rsm_12k_invariants_db.json# Full JSON specification texts
β”‚   β”œβ”€β”€ domain_invariant_embeddings.pt # Specialized domain overlay embeddings
β”‚   β”œβ”€β”€ domain_invariant_keys.json     # Specialized domain keys
β”‚   └── rsm_domain_invariants.json    # Specialized domain invariant texts
β”œβ”€β”€ pvl_router.py                 # Core routing engine (PVLTaskRouter & Toolchain Inspector)
β”œβ”€β”€ test_router.py                # Standalone verification test script
└── README.md                     # Documentation

3. How to Run the Router Locally

Ensure PyTorch and Transformers are installed:

pip install torch transformers

Run the standalone verification test:

python release/pvl-router/test_router.py

4. How the Router Pairs with Cranberry-12B

[Incoming User Task / Bug]
           β”‚
           β–Ό
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚       PVL CPU Semantic Router (MiniLM)       β”‚
β”‚  - Toolchain Inspection (ASan, GCC, Clang)   β”‚
β”‚  - Semantic Cosine Match against 12k Specs   β”‚
β”‚  - Energy: < 0.001 Joules CPU Opex           β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
           β”‚
           β”‚ (Top Verified Invariant Token + Hash)
           β–Ό
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚       Cranberry-12B GPU Engine (LoRA)        β”‚
β”‚  - KV-Cache Prefill of Invariant Constraints β”‚
β”‚  - Pure Patch Synthesis: T_payload = T_promptβ”‚
β”‚  - Avoided Autoregressive Decode Compute     β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
           β”‚
           β–Ό
[Deterministic Compiler / Harness Test Pass]

5. Licensing & Public Distribution

  • MiniLM-L6-v2: Released under the Apache 2.0 License by Microsoft / Sentence-Transformers.
  • PVL Invariant Specifications & Router Engine: Proprietary protocol coordination software of Preventative Value Labs under the PVL Protocol Specification.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for chappedjord/pvl-semantic-router