MohitML10/qwen2.5-32b-agentic-orchestrator
MohitML10/qwen2.5-32b-agentic-orchestrator is a 32.8 billion parameter model fine-tuned from Qwen2.5-32B-Instruct, specifically optimized for agentic tool-calling workflows. It excels at structured function calling, multi-turn tool chaining, and orchestrator decision-making, having been trained on 22,000 agentic conversations. This model is designed to act as a decision layer in complex AI systems, rather than for general question answering, and was trained on AMD Instinct MI300X hardware.
Loading preview...
Overview
MohitML10/qwen2.5-32b-agentic-orchestrator is a 32.8 billion parameter model, fine-tuned from Qwen/Qwen2.5-32B-Instruct, with a context length of 32768 tokens. Unlike general LLMs, this model is specifically trained to act as an agentic orchestrator, making decisions on when and how to use tools. It was fine-tuned using LoRA on 22,000 multi-turn agentic conversations, leveraging the full precision of the 32B model on AMD Instinct MI300X (192GB VRAM).
Key Capabilities
- Tool call decision-making: Determines when to use a tool versus responding directly.
- Structured tool call formatting: Generates consistent
<tool_call>and<tool_response>patterns. - Multi-turn chaining: Maintains context and orchestrates tool use across extended conversations.
- Escalation and handoff patterns: Identifies when to transfer tasks to humans or other agents.
- Constraint awareness: Adheres to system policies while fulfilling user goals.
Intended Use Cases
- Agentic orchestrators: Serves as the decision layer for tool selection.
- Multi-agent systems: Functions as the planning component above specialized agents.
- Tool-calling pipelines: Generates structured JSON tool calls.
- AI infrastructure research: Facilitates the study of agentic behavior at scale.
Hardware Requirements
- GGUF Q4_K_M: Requires a minimum of 24 GB RAM (24GB VRAM for Mac M2 Pro+ or single A100).
- BF16 safetensors: Requires 64 GB VRAM (80 GB VRAM recommended for A100/H100).