Kwokou/Spyra-20B-v.1.1-GGUF

Spyra-20B — Domain-Specific LLM for Architectural Design Reasoning (FP16 GGUF)

Spyra-20B is a domain-specific Large Language Model for the AEC industry (Architecture, Engineering, Construction). It combines Tree-of-Thought (ToT) and Chain-of-Thought (CoT) reasoning to decompose complex architectural design problems—mirroring how experienced architects explore alternatives and make decisions.

This version is provided in GGUF format (FP16 precision), optimized for use with Ollama, llama.cpp, or LM Studio.

Model Details

Model Description

Spyra-20B was developed to bridge the gap between general-purpose LLMs and the multidimensional reasoning required in architectural design. The model balances creativity, building code compliance, sustainability, structural constraints, and functional zoning.

The model utilizes a two-channel architecture: an Analysis Channel for structured reasoning (ToT/CoT) and a Final Channel for the user-facing answer.

• Developed by: [Nik Ansre & Yara Hirsekorn / Jade Hochschule]
• Model type: Causal Language Model (GGUF Quantization)
• Language(s): German (primary), English
• License: Apache License 2.0
• Base model: unsloth/gpt-oss-20b-BF16
• Format: GGUF (FP16 High Precision)
• Architecture: Mixture-of-Experts (MoE), 20B parameters

Intended Use

Primary Use Cases

Spyra-20B is designed as a professional assistant for architects, urban planners, and AEC professionals:

• Design Logic: Parametric design, spatial planning, and zoning strategies.
• Urban Densification: Evaluating strategies such as extensions, additions, or courtyard developments with trade-off analysis.
• Permit Processes: Navigating building codes (e.g., German BauO) and regulatory compliance logic.
• Renovation & Heritage: Handling historic preservation constraints and archaeological findings.
• Process Logic: Construction management, stakeholder coordination, and phased planning.

---

Training Details

Training Procedure

The model was fine-tuned using QLoRA on an expert-curated dataset and subsequently exported to the GGUF format with FP16 precision to ensure maximum reasoning quality and numerical stability compared to lower-bit quantizations.

Training Hyperparameters (Fine-tuning Phase)

Parameter	Value
Base model	unsloth/gpt-oss-20b-BF16
Method	QLoRA (Fine-tuning) -> GGUF Export
Precision	FP16 (GGUF)
Max steps	3,500
Training regime	bf16 mixed precision

---

How to Use

With Ollama

1. Create a file named Modelfile (content below).
2. Ensure the file Spyra-20B-f16.gguf is in the same directory.
3. Run the following commands:

ollama create spyra-20b -f Modelfile
ollama run spyra-20b