Something is cooking in the lab...
Fred-42B A3B is a dense Mermaid diagram-focused fine-tune built on top of DavidAU/Qwen3-42B-A3B-Grande-Claude-4.5-Opus-High-Reasoning-2507, trained primarily for generating accurate, well-structured diagrams in academic and STEM contexts.
Intended Use
Fred-42B A3B was developed to assist students and professionals in producing Mermaid-syntax diagrams, with a particular focus on Entity-Relationship (ER) diagrams for database design, as well as flowcharts, sequence diagrams, and class diagrams commonly used in computer science and informatics coursework.
The model is designed to integrate naturally into note-taking and knowledge management workflows, including tools like Obsidian via its native Mermaid rendering support. This makes it well-suited as an in-context diagram assistant for STEM-related academic tasks.
Training Data
- TeichAI/claude-4.5-opus-high-reasoning-250x: High-reasoning instruction data used to reinforce structured analytical output.
- TendieLabs/Tender_Mermaid_Training_V1: Mermaid diagram-specific training examples covering a range of diagram types and complexity levels.
Capabilities
- ER diagram generation from natural language descriptions or schema definitions
- General Mermaid diagram generation (flowcharts, sequence, class, and state diagrams)
- STEM academic task assistance with structured, diagram-augmented responses
- Obsidian-compatible output formatting
Limitations
Fred-42B A3B is optimized for diagram generation tasks and may underperform on general-purpose reasoning tasks relative to the base model. Output quality is best when prompts include clear schema definitions, entity lists, or explicit diagram type requests.
Fred-35B A3B is a Mermaid diagram-focused fine-tune built on top of Qwen/Qwen3.5-35B-A3B, trained primarily for generating accurate, well-structured diagrams in academic and STEM contexts.
Intended Use
Fred-35B A3B was developed to assist students and professionals in producing Mermaid-syntax diagrams, with a particular focus on Entity-Relationship (ER) diagrams for database design, as well as flowcharts, sequence diagrams, and class diagrams commonly used in computer science and informatics coursework.
The model is designed to integrate naturally into note-taking and knowledge management workflows, including tools like Obsidian via its native Mermaid rendering support. This makes it well-suited as an in-context diagram assistant for STEM-related academic tasks.
Training Data
- crownelius/Opus-4.6-Reasoning-2100x-formatted: High-reasoning instruction data used to reinforce structured analytical output.
- TendieLabs/Tender_Mermaid_Training_V1: Mermaid diagram-specific training examples covering a range of diagram types and complexity levels.
Capabilities
- ER diagram generation from natural language descriptions or schema definitions
- General Mermaid diagram generation (flowcharts, sequence, class, and state diagrams)
- STEM academic task assistance with structured, diagram-augmented responses
- Obsidian-compatible output formatting
Frank 9B is a dense general-purpose fine-tune built on Qwen/Qwen3.5-9B, designed for capable all-around reasoning, coding, and writing — optimized to run fast on consumer hardware.
Intended Use
Frank 9B is built for local deployment. It targets the sweet spot between size and capability — fast enough for real-time use on an RTX 3090 (~65 tok/s at IQ4_XS), smart enough to handle complex coding problems, structured reasoning, and creative writing tasks.
Where the Fred models specialize in Mermaid diagram generation, Frank 9B is a generalist — the goal is a capable, direct assistant with Claude-like reasoning style for everyday use.
Training Data
- microsoft/rStar-Coder: Math and code reasoning from competitive programming and STEM problem sets.
- NickyNicky/Code-290k: Multi-language code examples across 15 programming languages, no GPT contamination.
- Crownelius/GLM-5.0-25000x: General reasoning distillation with thinking traces.
- crownelius/Opus-4.6-Reasoning-3300x: Claude Opus 4.6 reasoning chains for structured analytical output.
- Crownelius/Opus-4.5-Writing-Style-formatted: Claude writing style and tone.
- Crownelius/High-Coder-Reasoning-Multi-Turn: Multi-turn coding sessions.
- Crownelius/Creative-Writing-Sonnet4.6-Cleaned: Creative writing from Claude Sonnet 4.6.
Capabilities
- General-purpose reasoning and problem-solving
- Multi-language code generation and debugging
- Mathematical reasoning and structured step-by-step output
- Creative writing with Claude-like tone and style
- Fast local inference — optimized for RTX 3090 / consumer GPUs
Model Details
- Base Model: Qwen/Qwen3.5-9B
- Parameters: 9B
- Architecture: Dense transformer
- Fine-tune Type: LoRA via Unsloth
- Inference Format: GGUF (IQ4_XS primary)
- Developer: TendieLabs
Limitations
Frank 9B is optimized for local consumer hardware use. It is not intended for safety-critical or high-stakes applications. As a personal-use fine-tune trained on Claude-derived datasets, it should not be used commercially or redistributed.
Frank 4B is a dense STEM-focused fine-tune built on Qwen/Qwen3.5-4B, trained for structured reasoning, code generation, and Mermaid diagram synthesis in academic and technical contexts.
Intended Use
Frank 4B was developed to assist students and professionals working across STEM disciplines. The model combines high-reasoning instruction tuning with domain-specific training in Mermaid diagram generation and competitive-level coding, making it well-suited for tasks that require both structured analytical output and accurate syntax generation.
Frank 4B is a sibling model to Fred-35B A3B, which focuses narrowly on Mermaid diagram generation. Frank trades some diagram specialization for broader STEM coverage, particularly in mathematical reasoning and competitive programming.
Training Data
- microsoft/rStar-Coder: Code and mathematical reasoning data derived from competitive programming and STEM problem sets.
- NickyNicky/Code-290k: Practical coding examples across 15 programming languages.
- Crownelius/GLM-5.0-25000x: Math and code with thinking traces.
- crownelius/Opus-4.6-Reasoning-3300x: High-reasoning instruction data for structured analytical output.
- Crownelius/Opus-4.5-Reasoning-782x: Opus reasoning style data.
- TendieLabs identity dataset: Frank persona injection via upsampled identity examples.
Capabilities
- ER diagram generation from natural language descriptions or schema definitions
- General Mermaid diagram generation (flowcharts, sequence, class, and state diagrams)
- Competitive-level code generation and mathematical reasoning
- STEM academic task assistance with structured, step-by-step responses
- Obsidian-compatible Mermaid output formatting
Model Details
- Base Model: Qwen/Qwen3.5-4B
- Parameters: 4B
- Architecture: Dense transformer
- Fine-tune Type: LoRA / QLoRA
- Developer: TendieLabs
Limitations
Frank 4B inherits the general limitations of its base model. It is not intended for use in safety-critical or high-stakes decision-making contexts. Diagram output should be reviewed for syntactic correctness before use in production tooling. Performance on domains outside STEM has not been evaluated.