| Name | dria_agent JSON |
| Version |
0.1.58
JSON |
| download |
| home_page | None |
| Summary | speedy, performant and small tool calling agents |
| upload_time | 2025-02-17 14:35:54 |
| maintainer | None |
| docs_url | None |
| author | andthattoo |
| requires_python | <4.0,>=3.12 |
| license | MIT |
| keywords |
|
| VCS |
|
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# Dria Agent
`tiny-agent-α` is a tiny model for building tool calling agents on edge devices.
It's fast and veeeery good compared to it's size. See for yourself:
https://github.com/user-attachments/assets/909656ad-4924-435a-8b4d-ac1b6d664d9c
### Features
Tiny-Agent-α is an extension of [Dria-Agent-a](https://huggingface.co/collections/driaforall/dria-agent-a-67a61f4b7d3d544fe5d3cd8a=), 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.
- **One-shot Parallel Multiple Function Calls**
- **Free-form Reasoning and Actions**
- **On-the-fly Complex Solution Generation**
#### Demo:
https://github.com/user-attachments/assets/5f7cbd26-7ba3-46aa-926f-4ac68de5ccb0
#### Edge Device Optimized:
- Supports mlx, ollama, and transformers (Hugging Face).
- Includes built-in support for macOS, Gmail, search, and more.
- Uses similarity search to efficiently select relevant tools.
- Optimized for Edge
`tiny-agent-a-0.5b` gets a whopping 72 on the _DPAB_ benchmark and run with `183.49 tokens/s` on a M1 macbook pro. Yet it's only **530MB**!
### Installation
To install the package run:
```bash
pip install dria_agent # Best for CPU inference
pip install 'dria_agent[mlx]' # To use MLX as backend for macOS.
pip install 'dria_agent[huggingface]' # HuggingFace/transformers backend for GPU.
pip install 'dria_agent[tools]' # In order to use factory tools in package
```
### Quick Start
#### CLI Mode
You can run the agent with pre-defined tools using the CLI.
```bash
dria-agent --chat
```
#### Using your own tools
Write your functions in pure python, decorate them with @tool to expose them to the agent.
````python
from dria_agent import tool
@tool
def check_availability(day: str, start_time: str, end_time: str) -> bool:
"""
Checks if a given time slot is available.
:param day: The date in "YYYY-MM-DD" format.
:param start_time: The start time of the desired slot (HH:MM format, 24-hour).
:param end_time: The end time of the desired slot (HH:MM format, 24-hour).
:return: True if the slot is available, otherwise False.
"""
# Mock implementation
if start_time == "12:00" and end_time == "13:00":
return False
return True
````
Create an agent:
```python
from dria_agent import ToolCallingAgent
agent = ToolCallingAgent(
tools=[check_availability]
)
```
Use agent.run(query) to execute tasks with tools.
```python
execution = agent.run("Check my calendar for tomorrow noon", print_results=True)
```
#### Run Modes
`agent.run()`
- **query (str)**: The user query to process.
- **dry_run (bool, default=False)**: If True, only performs inference—no tool execution.
- **show_completion (bool, default=True)**: Displays the model’s raw output before tool execution.
- **num_tools (int, default=2)**: Selects the best K tools for inference (using similarity search).
- *Allows handling thousands of tools efficiently*.
- * perform best with 4-5 tools max*.
- **print_results (bool, default=True)**: Prints execution results.
`agent.run_feedback()`
Same as run, but if there are errors in the execution, it will feed the errors back until execution is successful.
#### Tool Library
See [tool's library](dria_agent/tools/library/__init__.py) for implemented tools.
## Models
A fast and powerful tool calling model designed to run on edge devices.
| Model | Description | HF Download Link | Ollama Tag | Size |
|------------------------|--------------------------------------------|--------------------------------------------------------------------------------------------------------------------------|-------------------------------------|--------|
| Tiny-Agent-a-3B (8bit) | High performance and reasoning | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-3B/resolve/main/dria-agent-a-3b.Q8_0.gguf?download=true) | driaforall/tiny-agent-a:3B-q8_0 | 3.3 GB |
| Tiny-Agent-a-3B (4bit) | Tradeoff 3B quality for memory | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-3B/resolve/main/dria-agent-a-3b.Q4_K_M.gguf?download=true) | driaforall/tiny-agent-a:3B-q4_K_M | 1.9 GB |
| Tiny-Agent-a-1.5B (8bit) | Balanced performance and speed | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-1.5B/resolve/main/dria-agent-a-1.5b.Q8_0.gguf?download=true) | driaforall/tiny-agent-a:1.5B-q8_0 | 1.6 GB |
| Tiny-Agent-a-1.5B (4bit) | Faster CPU inference, performance tradeoff | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-1.5B/resolve/main/dria-agent-a-1.5b.Q8_0.gguf?download=true) | driaforall/tiny-agent-a:1.5B-q4_K_M | 986 MB |
| Tiny-Agent-a-0.5B (8bit) | Ultra-light | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-1.5B/resolve/main/dria-agent-a-1.5b.Q4_K_M.gguf?download=true) | driaforall/tiny-agent-a:0.5B-q8_0 | 531 MB |
## Evaluation & Performance
We evaluate the model on the **Dria-Pythonic-Agent-Benchmark ([DPAB](https://github.com/firstbatchxyz/function-calling-eval)):** 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](https://huggingface.co/blog/andthattoo/dpab-a) 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"}
}
```
Raw data
{
"_id": null,
"home_page": null,
"name": "dria_agent",
"maintainer": null,
"docs_url": null,
"requires_python": "<4.0,>=3.12",
"maintainer_email": null,
"keywords": null,
"author": "andthattoo",
"author_email": "omer@firstbatch.xyz",
"download_url": "https://files.pythonhosted.org/packages/c1/89/707f15a7818c1670159dcdf9fe76267f36810b6d6d6414f6524f7d7b05ca/dria_agent-0.1.58.tar.gz",
"platform": null,
"description": "# Dria Agent\n\n`tiny-agent-\u03b1` is a tiny model for building tool calling agents on edge devices.\n\nIt's fast and veeeery good compared to it's size. See for yourself:\n\nhttps://github.com/user-attachments/assets/909656ad-4924-435a-8b4d-ac1b6d664d9c\n\n\n### Features\n\nTiny-Agent-\u03b1 is an extension of [Dria-Agent-a](https://huggingface.co/collections/driaforall/dria-agent-a-67a61f4b7d3d544fe5d3cd8a=), trained on top of the Qwen2.5-Coder series to be used in edge devices. \nThese models are carefully fine-tuned with quantization aware training to minimize performance degradation after quantization. \nThe smallest model is 0.5B with 4bit quantization (398MB on disk), and the largest model is 3B with 4bit quantization.\n\n- **One-shot Parallel Multiple Function Calls**\n\n- **Free-form Reasoning and Actions**\n\n- **On-the-fly Complex Solution Generation**\n\n#### Demo:\n\nhttps://github.com/user-attachments/assets/5f7cbd26-7ba3-46aa-926f-4ac68de5ccb0\n\n#### Edge Device Optimized:\n- Supports mlx, ollama, and transformers (Hugging Face).\n- Includes built-in support for macOS, Gmail, search, and more.\n- Uses similarity search to efficiently select relevant tools.\n- Optimized for Edge\n\n`tiny-agent-a-0.5b` gets a whopping 72 on the _DPAB_ benchmark and run with `183.49 tokens/s` on a M1 macbook pro. Yet it's only **530MB**!\n\n### Installation\n\nTo install the package run:\n```bash\npip install dria_agent # Best for CPU inference\npip install 'dria_agent[mlx]' # To use MLX as backend for macOS. \npip install 'dria_agent[huggingface]' # HuggingFace/transformers backend for GPU.\npip install 'dria_agent[tools]' # In order to use factory tools in package\n```\n\n### Quick Start\n\n#### CLI Mode\n\nYou can run the agent with pre-defined tools using the CLI.\n\n```bash\ndria-agent --chat\n```\n\n#### Using your own tools\n\nWrite your functions in pure python, decorate them with @tool to expose them to the agent.\n\n````python\nfrom dria_agent import tool\n\n@tool\ndef check_availability(day: str, start_time: str, end_time: str) -> bool:\n \"\"\"\n Checks if a given time slot is available.\n\n :param day: The date in \"YYYY-MM-DD\" format.\n :param start_time: The start time of the desired slot (HH:MM format, 24-hour).\n :param end_time: The end time of the desired slot (HH:MM format, 24-hour).\n :return: True if the slot is available, otherwise False.\n \"\"\"\n # Mock implementation\n if start_time == \"12:00\" and end_time == \"13:00\":\n return False\n return True\n\n````\n\nCreate an agent:\n\n```python\nfrom dria_agent import ToolCallingAgent\n\nagent = ToolCallingAgent(\n tools=[check_availability]\n)\n\n```\n\nUse agent.run(query) to execute tasks with tools.\n\n```python\nexecution = agent.run(\"Check my calendar for tomorrow noon\", print_results=True)\n```\n\n#### Run Modes\n\n`agent.run()`\n\n- **query (str)**: The user query to process.\n- **dry_run (bool, default=False)**: If True, only performs inference\u2014no tool execution.\n- **show_completion (bool, default=True)**: Displays the model\u2019s raw output before tool execution.\n- **num_tools (int, default=2)**: Selects the best K tools for inference (using similarity search).\n - *Allows handling thousands of tools efficiently*.\n - * perform best with 4-5 tools max*.\n- **print_results (bool, default=True)**: Prints execution results.\n\n`agent.run_feedback()`\n\nSame as run, but if there are errors in the execution, it will feed the errors back until execution is successful.\n\n#### Tool Library\n\nSee [tool's library](dria_agent/tools/library/__init__.py) for implemented tools.\n\n\n## Models\n\nA fast and powerful tool calling model designed to run on edge devices.\n\n| Model | Description | HF Download Link | Ollama Tag | Size |\n|------------------------|--------------------------------------------|--------------------------------------------------------------------------------------------------------------------------|-------------------------------------|--------|\n| Tiny-Agent-a-3B (8bit) | High performance and reasoning | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-3B/resolve/main/dria-agent-a-3b.Q8_0.gguf?download=true) | driaforall/tiny-agent-a:3B-q8_0 | 3.3 GB |\n| Tiny-Agent-a-3B (4bit) | Tradeoff 3B quality for memory | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-3B/resolve/main/dria-agent-a-3b.Q4_K_M.gguf?download=true) | driaforall/tiny-agent-a:3B-q4_K_M | 1.9 GB |\n| Tiny-Agent-a-1.5B (8bit) | Balanced performance and speed | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-1.5B/resolve/main/dria-agent-a-1.5b.Q8_0.gguf?download=true) | driaforall/tiny-agent-a:1.5B-q8_0 | 1.6 GB |\n| Tiny-Agent-a-1.5B (4bit) | Faster CPU inference, performance tradeoff | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-1.5B/resolve/main/dria-agent-a-1.5b.Q8_0.gguf?download=true) | driaforall/tiny-agent-a:1.5B-q4_K_M | 986 MB |\n| Tiny-Agent-a-0.5B (8bit) | Ultra-light | [Download](https://huggingface.co/driaforall/Tiny-Agent-a-1.5B/resolve/main/dria-agent-a-1.5b.Q4_K_M.gguf?download=true) | driaforall/tiny-agent-a:0.5B-q8_0 | 531 MB |\n\n\n## Evaluation & Performance\n\nWe evaluate the model on the **Dria-Pythonic-Agent-Benchmark ([DPAB](https://github.com/firstbatchxyz/function-calling-eval)):** 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](https://huggingface.co/blog/andthattoo/dpab-a) for more information.\n\nBelow are the DPAB results: \n\nCurrent benchmark results for various models **(strict)**:\n\n| Model Name | Pythonic | JSON |\n|---------------------------------|----------|------|\n| **Closed Models** | | |\n| Claude 3.5 Sonnet | 87 | 45 |\n| gpt-4o-2024-11-20 | 60 | 30 |\n| **Open Models** | | |\n| **> 100B Parameters** | | |\n| DeepSeek V3 (685B) | 63 | 33 |\n| MiniMax-01 | 62 | 40 |\n| Llama-3.1-405B-Instruct | 60 | 38 |\n| **> 30B Parameters** | | |\n| Qwen-2.5-Coder-32b-Instruct | 68 | 32 |\n| Qwen-2.5-72b-instruct | 65 | 39 |\n| Llama-3.3-70b-Instruct | 59 | 40 |\n| QwQ-32b-Preview | 47 | 21 |\n| **< 20B Parameters** | | |\n| Phi-4 (14B) | 55 | 35 |\n| Qwen2.5-Coder-7B-Instruct | 44 | 39 |\n| Qwen-2.5-7B-Instruct | 47 | 34 |\n| **Tiny-Agent-a-3B** | **72** | 34 |\n| Qwen2.5-Coder-3B-Instruct | 26 | 37 |\n| **Tiny-Agent-a-1.5B** | **73** | 30 |\n\n\n#### Citation\n\n```\n@misc{Dria-Agent-a,\n url={https://huggingface.co/blog/andthattoo/dria-agent-a},\n title={Dria-Agent-a},\n author={\"andthattoo\", \"Atakan Tekparmak\"}\n}\n```\n\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "speedy, performant and small tool calling agents",
"version": "0.1.58",
"project_urls": null,
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "a41ddf6f3daab51ceed25f83ef4c0fb4da948ec6421881b24c7b9bbee4395df0",
"md5": "463a1751331fd9ef8ed0b18e5afd7620",
"sha256": "7b87b4a6e7356c3b898591cd4f648a5a967876f6b2a211345e28435ba862af4f"
},
"downloads": -1,
"filename": "dria_agent-0.1.58-py3-none-any.whl",
"has_sig": false,
"md5_digest": "463a1751331fd9ef8ed0b18e5afd7620",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": "<4.0,>=3.12",
"size": 47827,
"upload_time": "2025-02-17T14:35:52",
"upload_time_iso_8601": "2025-02-17T14:35:52.207675Z",
"url": "https://files.pythonhosted.org/packages/a4/1d/df6f3daab51ceed25f83ef4c0fb4da948ec6421881b24c7b9bbee4395df0/dria_agent-0.1.58-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "c189707f15a7818c1670159dcdf9fe76267f36810b6d6d6414f6524f7d7b05ca",
"md5": "2c8b65cad02af2faf9a239cb722adde5",
"sha256": "18db7b02d753a9c21f12a1ede402fee6fdc83ed9ca3846d7a2b6ec01495cdc59"
},
"downloads": -1,
"filename": "dria_agent-0.1.58.tar.gz",
"has_sig": false,
"md5_digest": "2c8b65cad02af2faf9a239cb722adde5",
"packagetype": "sdist",
"python_version": "source",
"requires_python": "<4.0,>=3.12",
"size": 38382,
"upload_time": "2025-02-17T14:35:54",
"upload_time_iso_8601": "2025-02-17T14:35:54.252078Z",
"url": "https://files.pythonhosted.org/packages/c1/89/707f15a7818c1670159dcdf9fe76267f36810b6d6d6414f6524f7d7b05ca/dria_agent-0.1.58.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-02-17 14:35:54",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "dria_agent"
}
