Instructions to use independently-platform/Tasky with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use independently-platform/Tasky with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="independently-platform/Tasky")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("independently-platform/Tasky")
model = AutoModelForCausalLM.from_pretrained("independently-platform/Tasky")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use independently-platform/Tasky with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "independently-platform/Tasky"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "independently-platform/Tasky",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/independently-platform/Tasky

SGLang

How to use independently-platform/Tasky with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "independently-platform/Tasky" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "independently-platform/Tasky",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "independently-platform/Tasky" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "independently-platform/Tasky",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use independently-platform/Tasky with Docker Model Runner:
```
docker model run hf.co/independently-platform/Tasky
```

Tasky / README.md

independently-platform

Update README.md

b2343db verified 6 months ago

preview code

Raw

History Blame Contribute Delete

2.25 kB

	---
	datasets:
	- independently-platform/tasky
	language:
	- en
	- it
	base_model:
	- google/functiongemma-270m-it
	library_name: transformers
	---

	# Tasky

	## About the model
	This model is a fine-tuned function-calling assistant for a todo/task application. It maps user requests to one of four tools and produces valid tool
	arguments according to the schema in `AI-TRAINING-TOOLS.md`.

	- Base model: `google/functiongemma-270m-it`
	- Primary languages: English and Italian (with light spelling errors/typos to mimic real users)
	- Task: Structured tool selection + argument generation

	## Intended Use
	Use this model to translate natural language task requests into tool calls for:
	- `create_tasks`
	- `search_tasks`
	- `update_tasks`
	- `delete_tasks`

	It is designed for task/todo management workflows and should be paired with strict validation of tool arguments before execution.

	### Example
	Input (user):

	Aggiungi un task per pagare la bolletta della luce domani mattina


	Expected output (model):
	```json
	{
	"tool_name": "create_tasks",
	"tool_arguments": "{\"tasks\":[{\"content\":\"pagare la bolletta della luce\",\"dueDate\":\"2026-01-13T09:00:00.000Z\"}]}"
	}

	## Training Data

	Synthetic, bilingual tool-calling data built from the tool schema, including:

	- Multiple phrasings and paraphrases
	- Mixed English/Italian prompts
	- Light typos and user mistakes in user_content
	- Broad coverage of optional parameters

	Splits:

	- Train: 1,500 examples
	- Eval: 500 examples

	## Training Procedure

	- Fine-tuning on synthetic tool-calling samples
	- Deduplicated examples
	- Balanced coverage of all tools and key parameters

	## Evaluation

	Reported success rate: 99.5% on the 500‑example eval split vs 0% base model.
	Success was measured as exact match on the predicted tool name and the JSON arguments after normalization.

	## Limitations

	- Trained for a specific tool schema; not a general-purpose assistant.
	- Outputs may include incorrect or incomplete tool arguments; validate before execution.
	- Language coverage is strongest in English and Italian.
	- Synthetic data may not capture all real-world user phrasing or ambiguity.