Welcome to distil labs

distil labs provides a platform for training task-specific small language models (SLMs) with just a prompt and a few dozen examples. Our platform handles the complex machine learning processes behind the scenes. This allows you to focus on your use-case instead of managing large datasets and infrastructure.

Why distil labs?

• Train Specialized Models with Minimal Data: Train specialized SLMs with just a prompt and a few dozen examples.
• Full Ownership of Your Models: Download and deploy your models anywhere - your servers, applications, or at the edge. No vendor lock-in for inference.
• Fully Automated ML Pipeline: No ML expertise required, we handle the technical complexity.
• High Performance with Smaller Footprint: Get LLM-level accuracy with models up to 400x smaller. Run locally on affordable hardware.

Minimal example

Here’s how simple it is to train a specialized model with distil labs (get YOUR_API_KEY):

1
import json
2
import requests
3
4
# Upload your task description and examples
5
data = {
6
    "config.yaml": open("data/config.yaml").read(),
7
    "job_description.json": open("data/job_description.json").read(),
8
    "test.jsonl": open("data/test.jsonl").read(),
9
    "train.jsonl": open("data/train.jsonl").read(),
10
    "unstructured.jsonl": open("data/unstructured.jsonl").read()
11
}
12
auth_header = {"Authorization": f"Bearer {distil_bearer_token()}"}
13
response = requests.post(
14
    "https://api.distillabs.ai/uploads",
15
    headers={"Content-Type": "application/json", **auth_header},
16
    data=json.dumps(data),
17
)
18
upload_id = response.json()["id"]
19
20
# Start SLM training
21
response = requests.post(
22
    f"https://api.distillabs.ai/trainings/{upload_id}",
23
    headers=auth_header,
24
)
25
slm_training_job_id = response.json()["id"]
26
27
# When training completes, get your model download link
28
response = requests.get(
29
    f"https://api.distillabs.ai/trainings/{slm_training_job_id}/model",
30
    headers=auth_header,
31
)
32
print(response.json())

Next steps

Ready to build your own specialized models? Continue to our How to train your SLM guide, see more detailed examples, or jump straight into the API reference.