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": {
"type": "veri"
},
"hf_push": {
"repo": "<string>",
"artifact": "<string>",
"private": true
},
"requires_heartbeat": true,
"volume": "<string>",
"resume_from": "<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": { "type": "veri" },
"hf_push": {
"repo": "<string>",
"artifact": "<string>",
"private": True
},
"requires_heartbeat": True,
"volume": "<string>",
"resume_from": "<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: {type: 'veri'},
hf_push: {repo: '<string>', artifact: '<string>', private: true},
requires_heartbeat: true,
volume: '<string>',
resume_from: '<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": {},
"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>",
"attribution": "<string>"
},
"resolved_parallelism": "<unknown>",
"refund": {
"status": "<string>",
"amount_usd": 123
},
"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",
"launched_at": "2023-11-07T05:31:56Z",
"provision_started_at": "2023-11-07T05:31:56Z",
"terminated_at": "2023-11-07T05:31:56Z",
"user_rate_per_hour_usd": 123,
"cost_so_far_usd": 123,
"visibility": "private",
"published_at": "2023-11-07T05:31:56Z"
}Authorizations
API key with the vk_ prefix. Create one from the dashboard.
Body
Show child attributes
Show child attributes
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.
The BYO-script payload. Persisted verbatim as TrainingJob.script_config_json and read back by the worker launch path (forward contract).
Show child attributes
Show child attributes
Inline reward source (Python text, TRL signature). Attached to the job at submit and stored job-scoped; there is no stored reward resource.
Where the checkpoint is stored: omit it or send {"type": "veri"}. Checkpoints are always stored in Veri storage; another type (s3, gs, az), a uri or credentials is rejected with a 400.
Show child attributes
Show child attributes
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
Show child attributes
A volume to mount read-only at /mnt/ on the training machine
(custom scripts see it there and in VERI_VOLUME_DIR). Coming soon:
behind VERI_FEATURE_VOLUME_MOUNTS, a non-blank value is a 400 until
the flag is on. To train on one file in a volume, connect it as a
dataset (datasets.from_volume) instead.
Continue training from an intermediate checkpoint of an earlier job:
a checkpoint id (ckpt_...) or <job_id>@<step>. The checkpoint must
be ready and resumable and its job finished; method, base_model,
dataset, LoRA settings and gpu_count must match the source job, and
the source job's reward functions are inherited (send none).
max_steps and num_epochs may change. The result is a NEW job with its
own id and billing.
Response
The created training job
queued, provisioning, configuring, running, completed, failed, cancelled 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.
How many inline reward sources the job was submitted with (null for reward-free methods). Sources are viewable per job via the dashboard.
The resolved hyperparameters, defaults filled in. The shape depends on
method: GrpoHyperparameters (grpo), AgenticGrpoHyperparameters
(grpo_agentic), HarnessGrpoHyperparameters (grpo_harness),
SftTextHyperparameters (sft_text), DpoHyperparameters (dpo),
SftVideoGenHyperparameters (sft_video_gen). Null for custom_script
jobs.
Pydantic GPUInfo (with type field). Rust uses gpu_type internally
but serializes as "type" for JSON parity.
Show child attributes
Show child attributes
Pydantic GPUInfo (with type field). Rust uses gpu_type internally
but serializes as "type" for JSON parity.
Show child attributes
Show child attributes
Show child attributes
Show child attributes
The multi-GPU launch plan the worker resolved (dp, zero_stage, micro_batch_size, gradient_accumulation_steps, global_batch_size, precision, ...). Null for single-GPU jobs and before the worker has planned.
Credit state for a platform-attributed failure; null when not applicable.
Show child attributes
Show child attributes
The hf_push request echoed back (null when the job didn't opt in).
Show child attributes
Show child attributes
Set after a successful end-of-training push to Hugging Face; the pushed repo's URL (mirrors wandb_run_url).
When the job first reached running (training started).
When the launch returned the instance: queued is
created_at..launched_at. Null until launched (and on jobs launched
before the column existed).
Billing start: the instance's metering anchor that cost_usd and
cost_so_far_usd count from. Stamped at launch on providers that bill
from launch, otherwise when the job reaches configuring.
When the instance was released (set on every terminal status).
The job's price in USD per hour: every GPU of every node, at the rate it is billed. Null when no price is known for its GPU shape.
Live cost in USD while the job is provisioning, configuring or
running: user_rate_per_hour_usd times the hours since
provision_started_at. Null otherwise; a finished job reports its
settled cost_usd.
Public-run state (Explore): private (default), public, or
admin_hidden (removed by Veri; cannot be re-published).
private, public, admin_hidden 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": {
"type": "veri"
},
"hf_push": {
"repo": "<string>",
"artifact": "<string>",
"private": true
},
"requires_heartbeat": true,
"volume": "<string>",
"resume_from": "<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": { "type": "veri" },
"hf_push": {
"repo": "<string>",
"artifact": "<string>",
"private": True
},
"requires_heartbeat": True,
"volume": "<string>",
"resume_from": "<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: {type: 'veri'},
hf_push: {repo: '<string>', artifact: '<string>', private: true},
requires_heartbeat: true,
volume: '<string>',
resume_from: '<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": {},
"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>",
"attribution": "<string>"
},
"resolved_parallelism": "<unknown>",
"refund": {
"status": "<string>",
"amount_usd": 123
},
"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",
"launched_at": "2023-11-07T05:31:56Z",
"provision_started_at": "2023-11-07T05:31:56Z",
"terminated_at": "2023-11-07T05:31:56Z",
"user_rate_per_hour_usd": 123,
"cost_so_far_usd": 123,
"visibility": "private",
"published_at": "2023-11-07T05:31:56Z"
}