Create training job

POST

training_jobs

Create training job

curl --request POST \
  --url https://api.veri.studio/v1/training_jobs \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "output_name": "<string>",
  "gpu": {
    "gpu_type": "<string>",
    "gpu_count": 123
  },
  "job_type": "<string>",
  "base_model": "<string>",
  "dataset_id": "<string>",
  "script": {
    "base_image": "<string>",
    "code_artifact_id": "<string>",
    "entrypoint": "<string>",
    "deps": {
      "kind": "<string>",
      "content": "<string>"
    },
    "env": {}
  },
  "reward_source": "<string>",
  "reward_sources": [
    "<string>"
  ],
  "reward_weights": [
    123
  ],
  "method": "<string>",
  "hyperparameters": "<unknown>",
  "provider": "<string>",
  "checkpoint_destination": {
    "uri": "<string>",
    "credentials": "<unknown>"
  },
  "hf_push": {
    "repo": "<string>",
    "artifact": "<string>",
    "private": true
  },
  "environments": {},
  "requires_heartbeat": true,
  "volume": "<string>",
  "region": "<string>",
  "num_nodes": 123
}
'

import requests

url = "https://api.veri.studio/v1/training_jobs"

payload = {
    "output_name": "<string>",
    "gpu": {
        "gpu_type": "<string>",
        "gpu_count": 123
    },
    "job_type": "<string>",
    "base_model": "<string>",
    "dataset_id": "<string>",
    "script": {
        "base_image": "<string>",
        "code_artifact_id": "<string>",
        "entrypoint": "<string>",
        "deps": {
            "kind": "<string>",
            "content": "<string>"
        },
        "env": {}
    },
    "reward_source": "<string>",
    "reward_sources": ["<string>"],
    "reward_weights": [123],
    "method": "<string>",
    "hyperparameters": "<unknown>",
    "provider": "<string>",
    "checkpoint_destination": {
        "uri": "<string>",
        "credentials": "<unknown>"
    },
    "hf_push": {
        "repo": "<string>",
        "artifact": "<string>",
        "private": True
    },
    "environments": {},
    "requires_heartbeat": True,
    "volume": "<string>",
    "region": "<string>",
    "num_nodes": 123
}
headers = {
    "Authorization": "Bearer <token>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {Authorization: 'Bearer <token>', 'Content-Type': 'application/json'},
  body: JSON.stringify({
    output_name: '<string>',
    gpu: {gpu_type: '<string>', gpu_count: 123},
    job_type: '<string>',
    base_model: '<string>',
    dataset_id: '<string>',
    script: {
      base_image: '<string>',
      code_artifact_id: '<string>',
      entrypoint: '<string>',
      deps: {kind: '<string>', content: '<string>'},
      env: {}
    },
    reward_source: '<string>',
    reward_sources: ['<string>'],
    reward_weights: [123],
    method: '<string>',
    hyperparameters: '<unknown>',
    provider: '<string>',
    checkpoint_destination: {uri: '<string>', credentials: '<unknown>'},
    hf_push: {repo: '<string>', artifact: '<string>', private: true},
    environments: {},
    requires_heartbeat: true,
    volume: '<string>',
    region: '<string>',
    num_nodes: 123
  })
};

fetch('https://api.veri.studio/v1/training_jobs', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

{
  "object": "<string>",
  "id": "<string>",
  "status": "queued",
  "method": "<string>",
  "output_name": "<string>",
  "created_at": "2023-11-07T05:31:56Z",
  "updated_at": "2023-11-07T05:31:56Z",
  "base_model": "<string>",
  "dataset_id": "<string>",
  "dataset_snapshot_id": "<string>",
  "reward_count": 123,
  "hyperparameters": {
    "learning_rate": 123,
    "num_epochs": 123,
    "max_steps": 123,
    "rollouts_per_prompt": 123,
    "kl_coef": 123,
    "max_prompt_length": 123,
    "max_response_length": 123,
    "global_batch_size": 123,
    "seed": 123,
    "use_unsloth": true,
    "lora_rank": 123,
    "lora_alpha": 123,
    "load_in_4bit": true,
    "checkpoint_every": 123,
    "checkpoint_keep": 123,
    "harness_protocol": "<string>",
    "max_turns": 123,
    "harness_timeout_s": 123,
    "max_parallel_rollouts": 123,
    "vllm_gpu_memory_utilization": 123,
    "max_model_len": 123,
    "trajectory_retention_days": 123
  },
  "base_image": "<string>",
  "entrypoint": "<string>",
  "gpu": {
    "type": "<string>",
    "count": 123
  },
  "gpu_requested": {
    "type": "<string>",
    "count": 123
  },
  "provider": "<string>",
  "region": "<string>",
  "num_nodes": 123,
  "error": {
    "code": "<string>",
    "message": "<string>"
  },
  "dashboard_url": "<string>",
  "download_url": "<string>",
  "wandb_run_url": "<string>",
  "hf_push": {
    "repo": "<string>",
    "artifact": "<string>",
    "private": true
  },
  "hf_repo_url": "<string>",
  "current_step": 123,
  "total_steps": 123,
  "current_loss": 123,
  "cost_usd": 123,
  "duration_seconds": 123,
  "started_at": "2023-11-07T05:31:56Z",
  "completed_at": "2023-11-07T05:31:56Z"
}

Authorizations

Authorization

string

header

required

API key with the vk_ prefix. Create one from the dashboard.

Body

application/json

output_name

string

required

gpu

object

required

Show child attributes

job_type

string

base_model

string | null

dataset_id

string | null

Dataset reference: a dataset id or name, optionally pinned to a snapshot — ds_…@snap-3 trains on that snapshot, ds_…@latest cuts (or reuses) a snapshot at the current head and pins it. A plain reference to a stream dataset behaves like @latest; a plain reference to a one-shot upload trains on it as before.

script

null | object

The BYO-script payload. Persisted verbatim as TrainingJob.script_config_json and read back by the worker launch path (forward contract).

Show child attributes

reward_source

string | null

Inline reward source (Python text, TRL signature). Attached to the job at submit and stored job-scoped; there is no stored reward resource.

reward_sources

string[] | null

reward_weights

number<double>[] | null

method

string

hyperparameters

any

provider

string | null

checkpoint_destination

null | object

Show child attributes

hf_push

null | object

Push the trained artifact to your Hugging Face account when the job completes. Requires a connected HF integration (PUT /v1/settings/integrations/huggingface). Validated at submit: managed grpo/grpo_harness/sft_text/dpo jobs only, and artifact="adapter" needs hyperparameters.lora_rank. On success the job carries hf_repo_url.

Show child attributes

environments

object | null

Show child attributes

requires_heartbeat

boolean

volume

string | null

region

string | null

num_nodes

integer<int32>

Response

The created training job

object

string

required

string

required

status

enum<string>

required

Available options:

queued,

provisioning,

configuring,

running,

completed,

failed,

cancelled

method

string

required

output_name

string

required

created_at

string<date-time>

required

updated_at

string<date-time>

required

base_model

string | null

dataset_id

string | null

dataset_snapshot_id

string | null

The dataset snapshot this job was pinned to at submit ("<dataset_id>@snap-N"); set when dataset_id was given as …@snap-N or …@latest (or names a stream dataset). Null for legacy submits.

reward_count

integer<int32> | null

How many inline reward sources the job was submitted with (null for reward-free methods). Sources are viewable per job via the dashboard.

hyperparameters

object

Harness-in-the-loop GRPO (VS-372): the user's own agent harness (Claude Agent SDK / LangChain / OpenAI Agents SDK) drives multi-turn rollouts against the in-training policy. Shares GRPO's core knobs; the harness-specific ones tune rollout execution and the colocated vLLM policy server. The harness code itself arrives via the job's script payload (same shape as custom_script).

Show child attributes

base_image

string | null

entrypoint

string | null

gpu

null | object

Pydantic GPUInfo (with type field). Rust uses gpu_type internally but serializes as "type" for JSON parity.

Show child attributes

gpu_requested

null | object

Pydantic GPUInfo (with type field). Rust uses gpu_type internally but serializes as "type" for JSON parity.

Show child attributes

provider

string | null

region

string | null

num_nodes

integer<int32>

error

null | object

Show child attributes

dashboard_url

string | null

download_url

string | null

wandb_run_url

string | null

hf_push

null | object

The hf_push request echoed back (null when the job didn't opt in).

Show child attributes

hf_repo_url

string | null

Set after a successful end-of-training push to Hugging Face; the pushed repo's URL (mirrors wandb_run_url).

current_step

integer<int32> | null

total_steps

integer<int32> | null

current_loss

number<double> | null

cost_usd

number<double> | null

duration_seconds

number<double> | null

started_at

string<date-time> | null

completed_at

string<date-time> | null

Create training job

curl --request POST \
  --url https://api.veri.studio/v1/training_jobs \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "output_name": "<string>",
  "gpu": {
    "gpu_type": "<string>",
    "gpu_count": 123
  },
  "job_type": "<string>",
  "base_model": "<string>",
  "dataset_id": "<string>",
  "script": {
    "base_image": "<string>",
    "code_artifact_id": "<string>",
    "entrypoint": "<string>",
    "deps": {
      "kind": "<string>",
      "content": "<string>"
    },
    "env": {}
  },
  "reward_source": "<string>",
  "reward_sources": [
    "<string>"
  ],
  "reward_weights": [
    123
  ],
  "method": "<string>",
  "hyperparameters": "<unknown>",
  "provider": "<string>",
  "checkpoint_destination": {
    "uri": "<string>",
    "credentials": "<unknown>"
  },
  "hf_push": {
    "repo": "<string>",
    "artifact": "<string>",
    "private": true
  },
  "environments": {},
  "requires_heartbeat": true,
  "volume": "<string>",
  "region": "<string>",
  "num_nodes": 123
}
'

import requests

url = "https://api.veri.studio/v1/training_jobs"

payload = {
    "output_name": "<string>",
    "gpu": {
        "gpu_type": "<string>",
        "gpu_count": 123
    },
    "job_type": "<string>",
    "base_model": "<string>",
    "dataset_id": "<string>",
    "script": {
        "base_image": "<string>",
        "code_artifact_id": "<string>",
        "entrypoint": "<string>",
        "deps": {
            "kind": "<string>",
            "content": "<string>"
        },
        "env": {}
    },
    "reward_source": "<string>",
    "reward_sources": ["<string>"],
    "reward_weights": [123],
    "method": "<string>",
    "hyperparameters": "<unknown>",
    "provider": "<string>",
    "checkpoint_destination": {
        "uri": "<string>",
        "credentials": "<unknown>"
    },
    "hf_push": {
        "repo": "<string>",
        "artifact": "<string>",
        "private": True
    },
    "environments": {},
    "requires_heartbeat": True,
    "volume": "<string>",
    "region": "<string>",
    "num_nodes": 123
}
headers = {
    "Authorization": "Bearer <token>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {Authorization: 'Bearer <token>', 'Content-Type': 'application/json'},
  body: JSON.stringify({
    output_name: '<string>',
    gpu: {gpu_type: '<string>', gpu_count: 123},
    job_type: '<string>',
    base_model: '<string>',
    dataset_id: '<string>',
    script: {
      base_image: '<string>',
      code_artifact_id: '<string>',
      entrypoint: '<string>',
      deps: {kind: '<string>', content: '<string>'},
      env: {}
    },
    reward_source: '<string>',
    reward_sources: ['<string>'],
    reward_weights: [123],
    method: '<string>',
    hyperparameters: '<unknown>',
    provider: '<string>',
    checkpoint_destination: {uri: '<string>', credentials: '<unknown>'},
    hf_push: {repo: '<string>', artifact: '<string>', private: true},
    environments: {},
    requires_heartbeat: true,
    volume: '<string>',
    region: '<string>',
    num_nodes: 123
  })
};

fetch('https://api.veri.studio/v1/training_jobs', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

{
  "object": "<string>",
  "id": "<string>",
  "status": "queued",
  "method": "<string>",
  "output_name": "<string>",
  "created_at": "2023-11-07T05:31:56Z",
  "updated_at": "2023-11-07T05:31:56Z",
  "base_model": "<string>",
  "dataset_id": "<string>",
  "dataset_snapshot_id": "<string>",
  "reward_count": 123,
  "hyperparameters": {
    "learning_rate": 123,
    "num_epochs": 123,
    "max_steps": 123,
    "rollouts_per_prompt": 123,
    "kl_coef": 123,
    "max_prompt_length": 123,
    "max_response_length": 123,
    "global_batch_size": 123,
    "seed": 123,
    "use_unsloth": true,
    "lora_rank": 123,
    "lora_alpha": 123,
    "load_in_4bit": true,
    "checkpoint_every": 123,
    "checkpoint_keep": 123,
    "harness_protocol": "<string>",
    "max_turns": 123,
    "harness_timeout_s": 123,
    "max_parallel_rollouts": 123,
    "vllm_gpu_memory_utilization": 123,
    "max_model_len": 123,
    "trajectory_retention_days": 123
  },
  "base_image": "<string>",
  "entrypoint": "<string>",
  "gpu": {
    "type": "<string>",
    "count": 123
  },
  "gpu_requested": {
    "type": "<string>",
    "count": 123
  },
  "provider": "<string>",
  "region": "<string>",
  "num_nodes": 123,
  "error": {
    "code": "<string>",
    "message": "<string>"
  },
  "dashboard_url": "<string>",
  "download_url": "<string>",
  "wandb_run_url": "<string>",
  "hf_push": {
    "repo": "<string>",
    "artifact": "<string>",
    "private": true
  },
  "hf_repo_url": "<string>",
  "current_step": 123,
  "total_steps": 123,
  "current_loss": 123,
  "cost_usd": 123,
  "duration_seconds": 123,
  "started_at": "2023-11-07T05:31:56Z",
  "completed_at": "2023-11-07T05:31:56Z"
}

Overview

Training jobs

Datasets

Deployments

Volumes

Models

Code artifacts

Regions

Billing

Workspace settings

Integrations

API keys

SSH keys

Webhooks

Authorizations

Body

Response