CoolFace
Modelpublic

orivus/Orivus-Architect-v1-7B-Lite-GGUF

sourceHugging Faceotherupdated 5mo agoView on Hugging Face
0likes32downloads
Model Card

Orivus-Architect-v1-7B GGUF

Orivus-Architect-v1 (7B) is the lightweight strategic reasoning engine for the Orivus Edge ecosystem.

It is designed for fast, local architectural analysis with structured outputs. This model prioritizes efficiency and responsiveness while maintaining deterministic behavior within the Orivus pipeline.

Model Details

  • —Base model: DeepSeek-R1-Distill-Qwen-7B
  • —Format: GGUF
  • —Quantization: Q4KM
  • —Prompt format: ChatML
  • —Runtime: llama.cpp-compatible environments
  • —Context window: 8192 tokens

Intended Use

This model is optimized for:

  • —quick architectural assessments
  • —risk estimation on isolated or mid-sized modules
  • —layered system reasoning (Core / Infrastructure / Edge / Domains / Entrypoints)
  • —generation of structured handoffs for execution engines
  • —local-first workflows with limited compute

This model is not intended to:

  • —perform deep multi-system analysis (use 14B variant)
  • —generate production code
  • —act as a general-purpose assistant
  • —operate without structured input signals

Output Contract

The model is expected to:

  • —return structured JSON outputs
  • —avoid speculative reasoning
  • —avoid inventing signals or dependencies
  • —ground all decisions strictly in provided inputs
  • —produce outputs compatible with Orivus-Coder execution pipelines

Non-structured or narrative responses are considered invalid.

Prompt Format (ChatML)

The model expects strict ChatML formatting:

text
<|im_start|>system
You are Orivus-Architect, a software architecture reasoning engine. Analyze structural signals and produce a deterministic, structured response. Do not write code. Do not invent data. Output strict JSON only.<|im_end|>
<|im_start|>user
{structured_input_signals}<|im_end|>
<|im_start|>assistant

Runtime Configuration (llama.cpp)

json
{
  "n_ctx": 8192,
  "temperature": 0.15,
  "top_p": 0.95,
  "repeat_penalty": 1.1
}

Orivus Edge Integration

Orivus-Architect-7B operates as the fast decision layer in Orivus Edge.

Execution flow:

  1. 1.Observe → analyze repository signals (AST, dependency graph, churn)
  2. 2.Orient → map signals to architectural layers and constraints
  3. 3.Decide → produce structured strategy and risk classification
  4. 4.Handoff → pass deterministic instructions to Orivus-Coder

This variant is optimized for:

  • —lower latency
  • —reduced hardware requirements
  • —iterative workflows and rapid feedback loops

Capabilities

  • —fast analysis of structured signals
  • —identification of architectural violations
  • —generation of deterministic handoffs
  • —suitable for iterative and local workflows
  • —efficient on limited hardware environments

Limitations

  • —reduced reasoning depth compared to 14B variant
  • —less effective for large-scale or highly complex systems
  • —requires structured input
  • —does not perform code generation

Available Files

  • —deepseek-r1-distill-qwen-7b.Q4KM.gguf

License

This model is derived from DeepSeek-R1-Distill-Qwen-7B.

Usage must comply with the Qwen License: https://github.com/QwenLM/Qwen2.5/blob/main/LICENSE

Fine-tuning and orchestration strategies are part of the Orivus Edge system design.