dria_agent 0.1.54

speedy, performant and small tool calling agents

Dria Agent

Dria Agent is a lightweight pip package to use fastest and most performant models for function calling on edge devices, tiny-agent-α.

Tiny-Agent-α is an extension of Dria-Agent-a, trained on top of the Qwen2.5-Coder series to be used in edge devices. These models are carefully fine-tuned with quantization aware training to minimize performance degradation after quantization. The smallest model is 0.5B with 4bit quantization (398MB on disk), and the largest model is 3B with 4bit quantization.

Installation

To install the package run: (Best for CPU inference)

pip install dria_agent
pip install 'dria_agent[mlx]' # To use MLX as backend for macOS. 
pip install 'dria_agent[huggingface]' # For HuggingFace backend for GPU.
pip install 'dria_agent[mlx, tools]' # In order to use extra tooling select a backend

Quick Start

Write your functions in pure python, decorate them with @tool to expose them to the agent.

from dria_agent import tool

@tool
def check_availability(day: str, start_time: str, end_time: str) -> bool:
    # Your implementation here
    return True

Create an agent:

from dria_agent import ToolCallingAgentFactory

agent = ToolCallingAgentFactory(
    tools=[check_availability], backend="ollama"
)

Use agent.run(query) to execute tasks with tools.

query = "Check if I'm available on Monday from 10:00 to 11:00"
execution = agent.run(query, print_results=True)

Models

Model	Description	HF Download Link	Ollama Tag	Size
Tiny-Agent-a-3B (8bit)	High performance and reasoning	Download	driaforall/tiny-agent-a:3B-q8_0	3.3 GB
Tiny-Agent-a-3B (4bit)	Tradeoff 3B quality for memory	Download	driaforall/tiny-agent-a:3B-q4_K_M	1.9 GB
Tiny-Agent-a-1.5B (8bit)	Balanced performance and speed	Download	driaforall/tiny-agent-a:1.5B-q8_0	1.6 GB
Tiny-Agent-a-1.5B (4bit)	Faster CPU inference, performance tradeoff	Download	driaforall/tiny-agent-a:1.5B-q4_K_M	986 MB
Tiny-Agent-a-0.5B (8bit)	Ultra-light	Download	driaforall/tiny-agent-a:0.5B-q8_0	531 MB

a fast and powerful tool calling model designed to run on edge devices.

3B 8bit high performance and quality download

3B 4bit balanced performance and speed download

1.5B 8bit fast CPU inference download

1.5B 4bit faster CPU inference download

0.5B 8bit ultra-light download

Evaluation & Performance

We evaluate the model on the Dria-Pythonic-Agent-Benchmark (DPAB): The benchmark we curated with a synthetic data generation +model-based validation + filtering and manual selection to evaluate LLMs on their Pythonic function calling ability, spanning multiple scenarios and tasks. See blog for more information.

Below are the DPAB results:

Current benchmark results for various models (strict):

Model Name	Pythonic	JSON
Closed Models
Claude 3.5 Sonnet	87	45
gpt-4o-2024-11-20	60	30
Open Models
> 100B Parameters
DeepSeek V3 (685B)	63	33
MiniMax-01	62	40
Llama-3.1-405B-Instruct	60	38
> 30B Parameters
Qwen-2.5-Coder-32b-Instruct	68	32
Qwen-2.5-72b-instruct	65	39
Llama-3.3-70b-Instruct	59	40
QwQ-32b-Preview	47	21
< 20B Parameters
Phi-4 (14B)	55	35
Qwen2.5-Coder-7B-Instruct	44	39
Qwen-2.5-7B-Instruct	47	34
Tiny-Agent-a-3B	72	34
Qwen2.5-Coder-3B-Instruct	26	37
Tiny-Agent-a-1.5B	73	30

Citation

@misc{Dria-Agent-a,
      url={https://huggingface.co/blog/andthattoo/dria-agent-a},
      title={Dria-Agent-a},
      author={"andthattoo", "Atakan Tekparmak"}
}