Tina-3.1-8B-Reasoning

Tina-3.1-8B-Reasoning

Tina-3.1-8B-Reasoning is a fine-tuned version of Llama 3.1 8B, optimized for complex reasoning, tool usage, and agentic workflows. It serves as the primary intelligence for multi-agent systems using the Agent-to-Agent (A2A) protocol.

🚀 Quantized Versions

For faster inference and lower VRAM usage, please use the following GGUF versions provided by the community:

• GGUF (Imatrix & Static): mradermacher/Tina-3.1-8B-Reasoning-GGUF

---

Model Details

Model Description

This model is a 16-bit implementation verified for agentic orchestration. It is designed to act as a "Purchasing Concierge" brain, capable of managing specialized agents.

• Developed by: Shubham Tiwari
• Model type: LLM Fine-tune (Reasoning/Agentic)
• Language(s): English
• Finetuned from model: meta-llama/Llama-3.1-8B-Instruct
• Training Framework: Unsloth

Uses

Direct Use

Ideal for:

• Complex reasoning tasks.
• Multi-agent orchestration (CrewAI, LangGraph).
• Tool-calling and API interaction.

Hardware Compatibility

Tested on AMD Instinct™ GPUs and compatible with Google Agent Development Kit (ADK). The GGUF versions allow this model to run on consumer hardware (MacBooks, NVIDIA RTX GPUs) via LM Studio or Ollama.

Training Details

• Method: GRPO / Fine-tuning via Unsloth.
• Precision: 16-bit.
• Focus: Logic, task decomposition, and tool-use reliability.

Model Card Contact

Shubham Tiwari