DGPL-BR: Autonomous System-1 Neural Decision Foundation Model

DGPL-BR is an ultra-fast, non-autoregressive 48.8M parameter System-1 neural decision foundation model developed by Durbhasi Gurukulam Private Limited (DGPL).

Designed for real-time robotic perception, autonomous vehicle trajectory selection, and spatial action planning, DGPL-BR outputs continuous 256-dimensional spatial-action latent embeddings and softmax probability distributions directly through neural weights in single-digit milliseconds.

---

⚡ Performance & Latency Benchmarks

Runtime Environment	Batch Size	Latency	Speedup vs Autoregressive LLM
ONNX Runtime (CPU / WebAssembly)	1	16.77 ms	173.8× faster
PyTorch CPU (torch.no_grad)	1	61.06 ms	47.7× faster
Standard 0.8B/2B LLM Baseline	1	2,915.73 ms	1.0× baseline

---

📦 Model Artifacts Included

1. dgpl_system1_v2.onnx (94.6 MB): Level-3 Graph Optimized ONNX runtime binary for edge devices, C++, Python (onnxruntime), and in-browser WebAssembly (onnxruntime-web).
2. dgpl_system1_v2_final.pt (133 MB): Complete PyTorch model checkpoint.

---

🚀 Quickstart: ONNX Runtime (Python)

import numpy as np
import onnxruntime as ort

# Load ONNX Session with all graph optimizations enabled
session = ort.InferenceSession("dgpl_system1_v2.onnx", providers=["CPUExecutionProvider"])

# Tokenize state description string
def text_to_tokens(text: str, max_len: int = 128):
    tokens = np.ones((1, max_len), dtype=np.int64)
    for i, ch in enumerate(text[:max_len]):
        tokens[0, i] = max(1, min(ord(ch) + 100, 31999))
    return tokens

# Inference
input_tokens = text_to_tokens("speed_50kmh_turn_left_dist_20m")
outputs = session.run(None, {"input_ids": input_tokens})

# Pooled latent vector representation [1, 256]
latent_vector = outputs[0]
print("Latent decision vector shape:", latent_vector.shape)

---

🚀 Quickstart: In-Browser WebAssembly (JavaScript)

import * as ort from "onnxruntime-web";

const session = await ort.InferenceSession.create("/onnx/dgpl_system1_v2.onnx", {
  executionProviders: ["wasm"],
  graphOptimizationLevel: "all"
});

// Run client-side zero-latency inference directly in the browser
const inputTensor = new ort.Tensor("int64", tokenArray, [1, 128]);
const results = await session.run({ input_ids: inputTensor });

---

🏛️ Corporate Identity & Attribution

• Organization: Durbhasi Gurukulam Private Limited (DGPL)
• Website: https://durbhasigurukulam.com
• License: GNU General Public License v3.0 (GPL-3.0)