Langtrain Logo
Langtrain Docs
GuidesAPI referenceSDKs
GitHubOpen the app

Search documentation

Jump to any docs page

  • REST API reference
  • Python SDK
  • Node.js and TypeScript
GitHub

API reference

Start, follow and export fine-tuning runs from your own code. Every endpoint lives under https://api.langtrain.xyz/api/v1 and returns JSON.

Authentication

Create an API key in the dashboard under API keys; it is shown once. Send it in the X-API-Key header (or as Authorization: Bearer). Keep it in an environment variable, never in code.
python
1export LANGTRAIN_API_KEY=sk-lt-...
2

3curl https://api.langtrain.xyz/api/v1/training/training-methods \
4  -H "X-API-Key: $LANGTRAIN_API_KEY"

Datasets

Upload a dataset in the dashboard (Data), then copy its ID. Runs refer to datasets by ID. The format is chat JSONL, one conversation per line; see Datasets.

Start a run

POST /training/jobs
  • –base_model - a Hugging Face model ID
  • –dataset_id - from the dashboard
  • –training_method - qlora (default), lora, dora, sft, ia3, prefix, dpo, orpo or kto
  • –hyperparameters - optional: n_epochs, learning_rate, batch_size, max_seq_length, lora_rank, lora_alpha, lora_dropout
  • –name - optional
The response is the run, with its id and a status of pending until a GPU picks it up.
python
1curl -X POST https://api.langtrain.xyz/api/v1/training/jobs \
2  -H "X-API-Key: $LANGTRAIN_API_KEY" \
3  -H "Content-Type: application/json" \
4  -d '{
5    "name": "support-assistant",
6    "base_model": "meta-llama/Llama-3.1-8B-Instruct",
7    "dataset_id": "YOUR_DATASET_ID",
8    "training_method": "qlora",
9    "hyperparameters": { "n_epochs": 3, "learning_rate": 0.0002, "lora_rank": 16 }
10  }'

Follow a run

GET /training/jobs/{id} returns the run: status (pending, running, completed, failed or cancelled), progress from 0 to 100, metrics such as loss, and error_message if it failed.
GET /training/jobs/{id}/stream sends the same updates as server-sent events while it trains.
python
1curl https://api.langtrain.xyz/api/v1/training/jobs/$JOB_ID \
2  -H "X-API-Key: $LANGTRAIN_API_KEY"
3

4curl -N https://api.langtrain.xyz/api/v1/training/jobs/$JOB_ID/stream \
5  -H "X-API-Key: $LANGTRAIN_API_KEY"

List and cancel runs

GET /training/jobs?organization_id=…&limit=10 lists a workspace's runs, newest first. POST /training/jobs/{id}/cancel stops a pending or running run.
python
1curl -X POST https://api.langtrain.xyz/api/v1/training/jobs/$JOB_ID/cancel \
2  -H "X-API-Key: $LANGTRAIN_API_KEY"

Export the weights

POST /training/jobs/{id}/export merges the adapter into the base model and pushes the result to a Hugging Face repo you own. Fields: repo_id (required), private (default true), merge_lora (default true), hf_token (or the one saved in your settings). Check progress with GET /training/jobs/{id}/export/{export_id}/status.
python
1curl -X POST https://api.langtrain.xyz/api/v1/training/jobs/$JOB_ID/export \
2  -H "X-API-Key: $LANGTRAIN_API_KEY" \
3  -H "Content-Type: application/json" \
4  -d '{"repo_id": "you/support-assistant", "private": true}'

Errors

Errors come back with an HTTP status and a JSON detail message:
  • –401 - missing or invalid API key
  • –404 - no run with that ID
  • –422 - the request body failed validation; detail says which field
PreviousFine-tune vision modelsNextPython SDK