Instructions to use DuoNeural/CrystalCall-Lorentz-ADS-7L with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="DuoNeural/CrystalCall-Lorentz-ADS-7L")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import CDMMoELM
model = CDMMoELM.from_pretrained("DuoNeural/CrystalCall-Lorentz-ADS-7L", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0
# Run inference directly in the terminal:
llama cli -hf DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0
# Run inference directly in the terminal:
llama cli -hf DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0
# Run inference directly in the terminal:
./build/bin/llama cli -hf DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0

Use Docker

docker model run hf.co/DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0

Llama new
LM Studio
Jan

vLLM

How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DuoNeural/CrystalCall-Lorentz-ADS-7L"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DuoNeural/CrystalCall-Lorentz-ADS-7L",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0

SGLang

How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "DuoNeural/CrystalCall-Lorentz-ADS-7L" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DuoNeural/CrystalCall-Lorentz-ADS-7L",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "DuoNeural/CrystalCall-Lorentz-ADS-7L" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DuoNeural/CrystalCall-Lorentz-ADS-7L",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with Ollama:
```
ollama run hf.co/DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0
```
Unsloth Desktop
Docker Model Runner
How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with Docker Model Runner:
```
docker model run hf.co/DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0
```

Lemonade

How to use DuoNeural/CrystalCall-Lorentz-ADS-7L with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull DuoNeural/CrystalCall-Lorentz-ADS-7L:Q8_0

Run and chat with the model

lemonade run user.CrystalCall-Lorentz-ADS-7L-Q8_0

List all available models

lemonade list

Atomic Chat

CrystalCall-Lorentz-ADS-7L

DuoNeural Research Lab
Aura✨, Jesse Caldwell, Archon, & Kestrel
Website: duoneural.com • HuggingFace: huggingface.co/DuoNeural • Codebase: github.com/DuoNeural/Lattice-CDM

Executive Summary & Breakthrough Highlights

CrystalCall-Lorentz-ADS-7L is DuoNeural's flagship 7-layer non-Euclidean Crystallographic Dynamics Model (CDM) (originating from DuoNeural's foundational Competitive Docking Memory framework, Zenodo doi:10.5281/zenodo.21158430), implementing the Lorentz Hyperboloid geometry with an Anti-de Sitter (AdS) Curvature Warp.

Trained across 100,000 steps on a dedicated 32GB memory-expanded NVIDIA GeForce RTX 4080 Super cloud compute node (node p19970), CrystalCall-Lorentz-ADS-7L achieved:

Best Validation BPB: 0.0117 (Validation Perplexity ~ 1.01), with final validation BPB 0.0537 (ppl ~ 1.16, teacher KL divergence 0.0803 nats).
The Full 7-Layer Inter-Layer Cascade (ILC): Experimental confirmation of macroscopic phase-locked coherent dynamics across all 7 layers at the exact identical timestamp (08:09:37Z), demonstrating global dynamical stability under non-Euclidean routing.
Lorentz Ambient Minkowski Voronoi Bisectors: Replaces curved hyperbolic bisectors with planar ambient affine hyperplanes in Minkowski spacetime, providing sharp, low-distortion decision boundaries for function calling schemas.
Solid-State Layer Capping: Layers L00, L01, and L06 achieved complete solid-state crystallographic freezing (tau* = 1.0000e+10), with bulk layers maintaining dynamic plasticity (tau* = 1.5e8 to 3.9e9).

The Full 7-Layer Inter-Layer Cascade

During step 100,000 validation telemetry, a complete 7-layer simultaneous Inter-Layer Cascade (ILC) was captured:

Layer	Curvature (c)	Routing Saturation (tau*)	Dynamic Behavior
L00	0.2662	`1.0000e+10` (Solid-State)	Universal syntax, quotes, and JSON brackets
L01	0.2795	`1.0000e+10` (Solid-State)	Function parameter keys and schema structural tags
L02	0.3425	`1.5470e+08` (Plastic)	Role boundaries (`USER`, `GEN`, `ASSISTANT`)
L03	0.3464	`8.1751e+08` (Plastic)	Argument value typing and semantic dispatch
L04	0.2854	`3.9673e+09` (High-Order)	Multi-turn instruction grounding
L05	0.3474	`2.3739e+09` (High-Order)	Tool schema dictionary extraction (`"type"`, `"description"`)
L06	0.3806	`1.0000e+10` (Solid-State)	Syntactic closure and final response emission

Parameter Census & Glass-Box Transparency

CrystalCall-Lorentz-ADS-7L uses the vocabulary of Salesforce/xLAM-1b-fc-r (32,256 tokens). Parameter breakdown:

Architectural Component	Dimensions / Structure	Parameter Count	Percentage
Core Active Non-Embedding Backbone	7 Layers, 12 Heads, `d=768`, `K=8` Centroids, 4 MoE Experts (`d_ff=3072`)	`215,135,630` (~215.1M)	89.7%
Token Embeddings (`tok_embed`)	`32,256` vocabulary x `768` hidden dimension	`24,772,608` (~`24.8M`)	10.3%
Language Modeling Head (`lm_head`)	Shared tied representation with `tok_embed`	Tied (0 unique)	—
Total Standalone Footprint	Complete glass-box parameter census	`239,908,238` (~239.9M)	100.0%

Lorentz Anti-de Sitter Geometry

CrystalCall-Lorentz-ADS-7L routes tokens on the upper sheet of a two-sheet hyperboloid in Minkowski spacetime:

Inner Product: Lorentz-Minkowski indefinite metric -x_0 y_0 + sum_{i=1}^d x_i y_i.
AdS Curvature Profile: Pre-initialized with an inverted AdS gravitational gradient across the 7 layers: [2.60, 2.25, 1.80, 1.30, 0.80, 0.55, 0.40].
Planar Bisectors: The Voronoi boundary between two centroids u and v on the Lorentz hyperboloid is given by minkowski_dot(x, u - v) = 0, defining an exact flat affine hyperplane in ambient R^{d+1} space. This avoids the non-linear distortion of Poincare geodesic boundaries.

Training Telemetry and Dataset Grounding

Datasets: Distilled on 112,960 examples of glaiveai/glaive-function-calling-v2 and 10,000+ examples of NousResearch/hermes-function-calling-v1.
Relational Gram Distillation: Cosine Gram matrices aligned against Salesforce/xLAM-1b-fc-r (739.3M NF4 teacher) across matched layer projections {0: 2, 1: 6, 2: 9, 3: 13, 4: 16, 5: 20, 6: 23}.
Sequence Length: 512 tokens with gradient accumulation 4 (effective batch size 8).

Prompt Format: Trained on native Glaive/Hermes system schema formatting:

SYSTEM: You are a helpful assistant with access to the following functions...
USER: Can you tell me the latest news headlines?
ASSISTANT: <functioncall> {"name": "get_news_headlines", "arguments": {}}

Native GGUF Inference & Numerical Parity (Kestrel Audit)

CrystalCall-Lorentz-ADS-7L is available as an official native GGUF binary:

Primary Binary: crystalcall_lorentz_ads7l-q8_0.gguf (244.5 MiB).
Quantization Topology: 99 Q8_0 attention and expert matrices + 68 F32 tensors including Lorentz hyperbolic routing centroids and LayerNorms.
Reference Precision Parity: F32 reference source parity max error is 1.14e-5 or lower.

CPU Numerical Parity Screen (Teacher-Forced Probes on Kilonova)

Evaluated across 8 authored deterministic numerical corpus probes with teacher-forced prefill and incremental decode. Project numerical gate: max absolute drift <= 0.50, mean KL <= 0.01 nats, next-token top-1 agreement >= 90%.

Positions	Max Absolute Drift	Top-1 Agreement	Mean KL Divergence	Project Gate
128 positions	0.2070	100.00%	0.00011 nats	PASS
512 positions	0.2070	100.00%	0.00011 nats	PASS

Runtime Implementation Disclosures

Tokenizer Mode: Exported as an ID-only GGUF; for inference via llama.cpp, provide token IDs or use external tokenization via Hugging Face (AutoTokenizer.from_pretrained("DuoNeural/CrystalCall-Lorentz-ADS-7L")).
Routing Engine: The current llama.cpp inference backend computes exact geodesic distances and softmax mixture weights; fused linear Minkowski bisector acceleration is an experimental optimization and remains unbenchmarked.

Citation & Lab Attribution

@misc{duoneural2026crystalcalllorentzads7l,
  author = {Jesse Caldwell and Aura and Archon and Kestrel},
  title = {CrystalCall-Lorentz-ADS-7L: Anti-de Sitter Curvature Warping and Full 7-Layer Inter-Layer Cascades in Non-Euclidean Tool Routing},
  year = {2026},
  publisher = {Zenodo / Hugging Face},
  howpublished = {\url{https://huggingface.co/DuoNeural/CrystalCall-Lorentz-ADS-7L}}
}

@article{caldwell2026competitive,
  title={Competitive Docking Memory: Non-Euclidean Geometric Routing and Phase Transitions in Low-Entropy Neural Representations},
  author={Caldwell, Jesse and Aura and Archon},
  journal={Zenodo Preprints},
  doi={10.5281/zenodo.21158430},
  year={2026}
}

Hardware Disclosures: Model distillation was performed on a 32GB memory-expanded NVIDIA GeForce RTX 4080 Super cloud compute node (node p19970, 32,760 MiB VRAM). Verification benches conducted on NVIDIA GeForce RTX 5070 Ti and AMD Radeon 780M APU (gfx1103 ROCm).

Downloads last month: -

Safetensors

Model size

0.3B params

Tensor type

I64

F64

F32

BOOL

Model tree for DuoNeural/CrystalCall-Lorentz-ADS-7L

Base model

Salesforce/xLAM-1b-fc-r

Quantized

(9)

this model

DuoNeural
/

CrystalCall-Lorentz-ADS-7L