noffy/hastejev-500k

⚡ Haste Jev 500k (Micro) (~500k Parameters)

Haste Jev 500k (Micro) is part of the Haste Jev family of open-weights, zero-bias System-1 Decision Engines. Micro System-1 decision model optimized for mobile CPUs, in-browser workers, and client-side web extensions.

---

🔬 Model Specifications

• Total Parameters: 500,091 (~500k)
• Trainable Parameters: 401,787
• Buffer / Projection Table: 98,304
• Hidden Dimension ($d_{\text{model}}$): 96
• Transformer Layers: 3
• Attention Heads: 4
• Target Deployment: Mobile CPU, In-browser workers
• Quantization Formats Available: FP32, FP16 (model_fp16.safetensors), INT8 (model_int8.safetensors), INT4 (model_int4.safetensors)

---

⚡ Quickstart

from hastejev import HasteJevEngine

# 1. Load standard weights directly from Hugging Face Hub
engine = HasteJevEngine.from_pretrained("noffy/hastejev-500k")

# 2. Or load with INT8 / INT4 quantization
engine_int8 = HasteJevEngine.from_pretrained("noffy/hastejev-500k", quantization="int8")

# 3. Execute Decision Primitives
state = "Account balance is $14,850.50 with pending transaction of $3,200.00."
options = ["Approve Transaction", "Flag for Review", "Decline"]

res = engine.choice(state, options)
print(f"Decision: {res.decision} (Confidence: {res.confidence:.3f})")

---

📊 Complete Haste Jev Model Family

Model	Parameters	Hidden Dim	Layers	Heads	RAM (FP32)	RAM (INT8)	Target Use Case
hastejev-100k	~98k	48	2	2	~0.4 MB	~0.1 MB	Microcontrollers, WASM, IoT
hastejev-500k	~500k	96	3	4	~2.0 MB	~0.5 MB	Mobile CPU, in-browser workers
hastejev-1m	~1.1M	128	4	4	~4.4 MB	~1.1 MB	High-throughput API sidecars
hastejev-2m	~1.8M	160	4	4	~7.3 MB	~1.8 MB	Browser automation & bots
hastejev-5m	~5.0M	224	5	4	~20.0 MB	~5.0 MB	Financial & KYC routing
hastejev-10m	~10.0M	320	5	4	~40.0 MB	~10.0 MB	Multimodal agent kernels
hastejev-20m	~20.4M	256	4	4	~81.5 MB	~20.4 MB	Enterprise decision engine

---

📄 License

Apache License 2.0.