Talk to CRATER Tiny

Standalone scripted demonstration of the proposed message-processing path

CRATER Tiny

Ask about angular attention, ternary weights, retrieval, or the proposal's limits.

Local demo
Simulation, not a trained checkpoint. Responses come from a small local knowledge map. The operator trace shows data flow, not hidden chain-of-thought. No network request occurs.
Proposed LLM artifact

crater-tiny.crater

Specification preview

No model file is loaded by this demo. A deployed runtime would verify the signed manifest once, memory-map the packed weights, then reuse the loaded artifact for every message.

Open fileVerify manifestmmap weightsRun tokens
Packed core
Base-3 ternary weights plus 2-bit quaternary heads
Q/K/V/O + experts
Quantization metadata
FP16 output scales γw, learned thresholds Δ, head mask
Applied after accumulation
Sensitive modules
Embeddings, CLN, router, gates, LoRA and output head
FP16 / optional INT8
Manifest + tokenizer
Architecture, packing layout, IRW settings and vocabulary
Signed configuration

The paper's 7B-complex example estimates about 2.8 GB for packed ternary weights and about 3 GB after stated overhead. CRATER Tiny has no released checkpoint or measured file size.

CRATER Tiny simulated

Hello. Ask me about angular attention, ternary weights, retrieval, or the limits of this proposal.