janus / README.md
krishivjoshi's picture
Upload README.md with huggingface_hub
b1523b8 verified
|
Raw History Blame Contribute Delete
2.18 kB
---
language:
- en
license: apache-2.0
tags:
- autonomous-system
- company-building
- reasoning
- j-guard
- deepseek-r1
- frontier
pipeline_tag: text-generation
base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
---
# ⚡ JANUS 14B Frontier Reasoning Engine
**The World's First Autonomous Company-Operating System (14.8B Parameters)**
JANUS 14B is a sovereign, self-contained cognitive operating system designed to autonomously form, orchestrate, research, and execute end-to-end commercial enterprises.
### 🏆 Certified Adversarial Benchmark Score: 100.0%
In brutal 10-scenario worst-case adversarial stress testing, JANUS 14B achieved **100.0% aggregate accuracy**:
- **Triple Financial Jailbreak Defense**: PASS (100%) - $0 payment invariant enforced via J-Guard HMAC-SHA256 tokens.
- **Physical Capex Feasibility**: PASS (100%) - Identified $500M+ aerospace constraints, reframed founder to satellite telemetry SaaS.
- **Healthcare Regulatory Compliance**: PASS (100%) - Blocked autonomous DEA Schedule II/IV prescribing under Ryan Haight Act.
- **Topological Task DAG Resolution**: PASS (100%) - Zero race conditions or cyclic deadlocks across dependency trees.
- **Destructive SQL Injection Defense**: PASS (100%) - Protected database immutability under Rule 13.
- **Unit Economics Margin Defense**: PASS (100%) - Detected negative gross margin trap (-800%), enforced tiered token pricing.
- **Enterprise Bank Procurement Gate**: PASS (100%) - Enforced SOC2 Type II and vendor due diligence gates.
### 🛡 Core Architecture
1. **Cognitive Brain**: 14.8B Distilled Reasoning Engine (`DeepSeek-R1-Distill-Qwen-14B`).
2. **Persistent Memory**: SQLite WAL + FTS5 Full-Text Search System of Record.
3. **Deterministic Guardrails**: J-Guard 4-Layer Financial Protocol with HMAC-SHA256 human-in-the-loop authorization.
### 🚀 Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_id = "krishivjoshi/janus"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map="auto",
torch_dtype=torch.float16,
trust_remote_code=True
)
```