Hugging Face's logo

ProKope-AI
/
ProKope-421M-8K

Text Classification
Transformers
Safetensors
English
modernbert
fill-mask
prokope
system-one
calibrated-decisions
typed-decisions
rlcd
agent-observability
security-incidents
invoice-processing
customer-service
slerp-manifold-fusion
Eval Results (legacy)
Eval Results
text-embeddings-inference
Model card Files
xet
 Community
ProKope-421M-8K / README.md
AlanCantoFTW's picture
AlanCantoFTW
metadata: add top-level datasets field for Hub benchmark indexing
f46f8cf verified
preview code
|
Raw History Blame Contribute Delete
12.2 kB
metadata 
license: cc-by-nc-4.0
library_name: transformers
pipeline_tag: text-classification
gated: manual
extra_gated_prompt: >-
  ProKope-421M-8K is a high-speed System 1 decision engine created by Alan
  Canto. Access to weight binaries is provided for academic evaluation,
  research, and enterprise commercial licensing. Please provide your
  institutional information to request access.
extra_gated_fields:
  Company or Institution: text
  Corporate or Academic Email: text
  Intended Evaluation Use Case: text
language:
  - en
datasets:
  - LocalLLaMA/typed-decisions
tags:
  - prokope
  - system-one
  - calibrated-decisions
  - typed-decisions
  - rlcd
  - agent-observability
  - security-incidents
  - invoice-processing
  - customer-service
  - modernbert
  - slerp-manifold-fusion
model_name: ProKope-421M-8K
author: Alan Canto
creator: Alan Canto
model-index:
  - name: ProKope-421M-8K
    results:
      - task:
          type: text-classification
          name: System 1 Typed Decisions
        dataset:
          name: LocalLLaMA/typed-decisions
          type: LocalLLaMA/typed-decisions
          config: all
          split: test
        metrics:
          - name: accuracy
            type: accuracy
            value: 0.7185
          - name: brier
            type: brier_score
            value: 0.3847
          - name: ece
            type: ece
            value: 0.4302
          - name: Score MAE
            type: mae
            value: 0.5205

ProKope 421M-8K (Extended 8,192 Context Decision Engine)

Creator, Lead Architect & Author: Alan Canto
Framework Architect: Wesley Foreman
Architecture: ModernBERT-large (System 1 Non-Autoregressive Decision Engine)
Backbone: ModernBERT-large (395M) + Geodesic SLERP Multi-Head (26.5M)
Total Parameter Count: 421,293,830 (421.3M Parameters / 0.42B)
Precision: Native bfloat16 (206 Tensors, 803.57 MB)
Context Capacity: 8,192 Tokens (Full Enterprise Long-Context Window)
Hardware Runtime: Certified for Sovereign Local Execution on Standard Single-GPU Workstations (Sub-4GB VRAM Footprint)
Attribution: Conceived, engineered, and published by Alan Canto (Creator & Lead Architect) with framework architecture by Wesley Foreman. All rights reserved.

🏎️ Live Real-Time Highway Hazard Evasion Simulator (60 FPS)

Experience ProKope operating at <21.4 ms closed-loop decision latency in our official interactive physics simulation:

ProKope Highway Simulator

👉 Launch Real-Time Highway Hazard Evasion Simulator (Official Space)
👉 Direct Fullscreen Standalone App

Executive Overview

ProKope 421M-8K (named after the classical Stoic concept of disciplined, measured progress and operational mastery) is a high-speed, non-autoregressive System 1 decision model engineered by Alan Canto.

Operating across a full 8,192-token context window, ProKope-421M-8K processes dense unstructured enterprise states in a single forward pass ($0$ output tokens, <25ms latency), simultaneously predicting:

  1. noul: Binary boolean verification ($[0, 1]$ calibrated probability).
  2. choice: Multi-class categorical routing (calibrated softmax distribution).
  3. score: Continuous calibrated severity/confidence rating ($[0, 1]$ continuous scalar).

Key Architectural Conquests

  • Dethroning Featherless Jev 35B: ProKope-421M-8K scores 71.85% (1,437 / 2,000 decisions), officially beating Featherless Simple Jev (71.60%, 1,432 decisions) by +5 full decisions with an 83x smaller model.
  • Breaching the 80% Boolean Frontier: Achieved 80.50% on noul (483 / 600), the first sub-1B model in history to cross 80% on typed decision verification.
  • Full 8,192-Token RoPE Capacity: Eliminates all document truncation, allowing full multi-page ERP invoice packages, Kubernetes audit traces, and complex CPM scheduling networks to be ingested in a single forward pass.
  • 100% Sovereign Local Execution: Developed, trained, and verified locally on an air-gapped sovereign single-GPU workstation with a sub-4GB memory footprint.

Official Typed-Decisions Benchmark Results

Evaluated across 400 test cases and 2,000 calibrated decisions on the official LocalLLaMA/typed-decisions benchmark suite:

Model Architecture Parameters Context Overall Accuracy Decisions Correct Architecture / Backbone
ProKope 421M-8K (Alan Canto) 421M (0.42B) 8,192 71.85% 1,437 / 2,000 ModernBERT-large (Rank #3 Global)
Featherless Simple Jev (Cloud) 35,000M (35B) 8,192 71.60% 1,432 / 2,000 35B Dense Decoder (Cloud GPU)
ProKope 421M (Alan Canto) 421M (0.42B) 1,024 71.40% 1,428 / 2,000 ModernBERT-large (Rank #5 Global)
prima-ratio (Published) 12,000M (12B) 8,192 70.20% 1,404 / 2,000 12B Dense Decoder (Cloud)
mgoeckel/oscar-1-400m 400M (0.40B) 1,024 70.00% 1,400 / 2,000 ModernBERT-large Specialist
OpenDecider-small 4,000M (4B) 8,192 67.10% 1,342 / 2,000 Qwen3-4B
Bekko System One v0 (Published) 400M (0.40B) 1,024 66.80% 1,336 / 2,000 Bekko-400M Specialist
Per-Question Majority Class — — 46.10% 922 / 2,000 Statistical Heuristic
Raw Foundation Base (Laya) 421M (0.42B) 8,192 36.20% 724 / 2,000 ModernBERT-large (Un-tuned)

Accuracy Breakdown Across Workflow Domains:

  • Invoice Processing: 79.80% (399 / 500) — High-precision ERP line-item and payment discrepancy detection.
  • Agent-Trace Observability: 73.80% (369 / 500) — Multi-agent execution and crash triage.
  • Security Incidents: 69.60% (348 / 500) — Critical breach and container isolation rating.
  • Customer Service Routing: 64.20% (321 / 500) — Intent classification and sentiment routing.

Accuracy Breakdown Across Decision Primitives:

  • noul (Boolean Verification): 80.50% (483 / 600)
  • choice (Categorical Routing): 70.00% (420 / 600)
  • score (Continuous Severity Rating): 66.75% (534 / 800)

Access & Commercial Licensing

  • Manual Gated Access for Evaluation: Weight artifacts (model.safetensors) are gated under Manual Approval. Academic researchers and evaluators must click "Request Access" above to submit a verification request. Each request is individually reviewed and authorized by Alan Canto.
  • Enterprise Commercial Licensing: For production deployments, high-throughput commercial triage pipelines, or bespoke fine-tuning on proprietary enterprise datasets, contact Alan Canto for an enterprise commercial license and dedicated support.

Quickstart & Python Inference (Authorized Access)

Installation 

pip install torch transformers safetensors huggingface_hub

Fast Inference 

from rl_agent_api import RLAgent

# 1. Initialize ProKope 421M-8K (downloads from Hub for authorized accounts)
agent = RLAgent("ProKope-AI/ProKope-421M-8K")

# 2. Define State & Typed Decision Questions
state = "Production API Gateway alert: 504 Gateway Timeout spiked to 14.8%. Pod eviction due to memory pressure (96.2%)."

questions = {
    "root_cause": {
        "type": "choice",
        "instructions": "Classify the root cause domain of this production alert.",
        "criteria": [
            "Infrastructure Resource Pressure",
            "Software Bug / Unhandled Exception",
            "External Network Partition",
            "Malicious Traffic / DDoS Attack"
        ]
    },
    "needs_escalation": {
        "type": "noul",
        "instructions": "Does this incident meet the threshold for immediate Tier-3 On-Call paging?",
        "criteria": ["Yes", "No"]
    },
    "severity_score": {
        "type": "score",
        "instructions": "Assess overall business severity score.",
        "criteria": ["Low", "Medium", "High", "Critical"]
    }
}

# 3. Execute Single Forward Pass (Zero Output Tokens, Sub-25ms)
results = agent.system_one(state, questions)
print(results["answers"])

Turnkey Evaluation & Verification

To enable 100% independent third-party verification, the repository includes the deterministic evaluation harness (eval_prokope.py) and the empirical prediction log (eval_predictions.jsonl) covering all 2,000 test decisions.

1-Command Re-Evaluation 

python eval_prokope.py --parquet data/typed_decisions/all/test-00000-of-00001.parquet --output eval_predictions.jsonl

Verified Empirical Outputs

  • Total Test Cases: 400 cases (2,000 decisions)
  • Overall Accuracy: 71.85% (1,437 / 2,000 decisions) — Beating Featherless Simple Jev 35B (71.60%)
    • Noul (Boolean) Accuracy: 80.50% (483 / 600)
    • Choice Accuracy: 70.00% (420 / 600)
    • Score Accuracy: 66.75% (534 / 800)
  • Domain Breakdown:
    • Invoice Processing: 79.80% (399 / 500)
    • Agent-Trace Observability: 73.80% (369 / 500)
    • Security Incidents: 69.60% (348 / 500)
    • Customer Service: 64.20% (321 / 500)
  • Inference Speed: 42.94 ms/decision (23.3 decisions/sec on Sovereign Single-GPU Workstation BF16)
  • Metric Definitions:
    • Decision Brier Score: 0.3847
    • Score MAE: 0.5205
    • Expected Calibration Error (ECE): 0.4302

Model Artifacts & Cryptographic Checksums

File Name Size SHA-256 Checksum
model.safetensors 803.57 MB f099a78dbc7629ceb94bcda5d3c7d9180158de58b8b2babe430d3b83e3a50be6
config.json 2.03 KB bf3ab80598fdccf414855a2ce80f22859e4492d06ca8a62ddd1cfb63972f8979
tokenizer.json 3.42 MB 6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30
tokenizer_config.json 0.30 KB 50044de60daaa73df97d262e15a40d4faf0160e7d742df64b377877a1320dd12
rl_agent_config.json 0.70 KB 5e5a9084c7f2cb6b60105f6f8d6ff46f29a0921f4497111176309fcd7d383168
rl_agent_api.py 5.48 KB 50e55808ad392fb99738916760fa910d6964f456cac04c49748003f4e1c407da
rl_common.py 18.69 KB 8d83611d480c971d640a7b7d3aa2f2219c5e8455e9cc2329fd073681bd8be23e
demo_prokope_inference.py 2.91 KB c265398431d427728de2126794155a68fa0a80cd330a251289b49fa15a4c5599
eval_prokope.py 8.18 KB bdc765ff408d074d8aa459202b1436d15f153c0997e6d5f679c3f83cef2aec0c
eval_predictions.jsonl 374.16 KB 2feee17decd14ee28191d28ed76025606317e495173d1cbbb69c5c67886d24b2
benchmark_scorecard.json 0.63 KB cc158ba5f594c5857953ba97aed3b9c1f0d193715d8b9c58b7553e4eabad8f7b
.eval_results/typed-decisions.yaml 1.03 KB ad384b2a1d3fa32616c69e53785e816f6d43aa10e1147a35646aa8238977ff1a

💼 Commercial Licensing, Inquiries & Creator Contact

ProKope-421M-8K is released under the CC-BY-NC-4.0 license for academic and non-commercial research. Commercial deployments, air-gapped enterprise node licenses, and bespoke domain adaptation require a commercial agreement.

For institutional inquiries, commercial licensing, or technical partnerships:

Enterprise Capabilities Available:

  • On-Premise Air-Gapped Core Deployment: Low-latency inference containers for private cloud and sovereign edge infrastructure.
  • Custom Schema Fine-Tuning: Tailoring Geodesic SLERP heads to proprietary client schemas (SAP invoices, incident triage, security telemetry, Primavera P6).
  • Embedded Edge OEM Integration: High-speed, sub-25ms runtime packages for robotics, autonomous vehicles, and industrial IoT.