README.md

metadata

license: apache-2.0
datasets:
  - TendieLabs/Tender_Mermaid_Training_V2
  - crownelius/Opus-4.6-Reasoning-2100x-formatted
  - microsoft/rStar-Coder
language:
  - en
base_model:
  - allenai/Olmo-3-7B-Think

Fred 7B is a general-purpose reasoning model built on OLMo-3-7B-Think, developed by Tendie Labs. It continues the Fred model lineage — following Fred 32B and Fred 42B — with a focus on STEM reasoning and structured diagram generation.

Fred 7B was trained on Tender Mermaid V2, an updated and improved version of the dataset pipeline used for Fred 32B and Fred 42B — meaning its Mermaid.js capabilities are built on a stronger foundation than the larger models despite the smaller parameter count. Portions of the training dataset were generated with the assistance of Fred-35B-A3B, Tendie Labs' own MoE model, making this a dataset refined within the Fred lineage itself. It also incorporates reasoning-focused and code-oriented data to improve general problem-solving performance at the 7B scale.

Intended Use

Fred 7B is well-suited for tasks involving:

• Mermaid.js diagram generation (flowcharts, sequence diagrams, ERDs, etc.)
• STEM reasoning and structured problem-solving
• Code generation and technical explanation

Training Data

• TendieLabs/Tender_Mermaid_Training_V2 — Tendie Labs' in-house Mermaid diagram dataset, partially generated with Fred-35B-A3B
• crownelius/Opus-4.6-Reasoning-2100x-formatted — reasoning-focused instruction data
• microsoft/rStar-Coder — code reasoning data from Microsoft