Train / Fine-tuning

Bring your JSONL, pick a base. Your first fine-tune runs in minutes.

Managed GPUs run the training loop, the scheduling and the checkpointing. Start from the dashboard with no code, or with one API call, and score the result against the base it started from.

Support triage v3Language · LoRA rank 16
Base model
Qwen3.8 27B
Dataset
tickets-2026.jsonl
TrainingEpoch 3 of 468%
Training lossstep 612 of 900
0.335
ckpt 10.640ckpt 20.384ckpt 3ckpt 4

One call replaces the loop

POST /v1/fine-tuning/jobs takes a base model and a training file, and the platform runs the training, the scheduling and the checkpointing.

Defaults you can leave alone

Epochs, learning rate, batch size and LoRA rank are all optional. Anything you omit is chosen for you, and the job reports the resolved values.

Watch the loss while it spends

The events stream carries the loss curve step by step and warns when validation loss starts to rise.

Cancel, and pay for what ran

Cancelling stops the run and stops the meter. billed_tokens_so_far is the total you are charged.

How it works

From a JSONL file to a scored checkpoint

Upload, choose, train, score. The platform validates the file, trains on managed GPUs and runs the evaluation you ask for. Take the dashboard with no code, the API with one call, or the training loop when the loss is yours to write.

Upload, choose, train, score

Four steps to a scored checkpoint

Upload a JSONL file, or connect a source so the training set keeps itself current. Choose a base from the catalogue. Managed GPUs train it, with a checkpoint at every epoch. Then score the result against the base it started from.

train.jsonlready
file_9fQ2x · purpose fine-tune
val.jsonlready
errorline 812
or
proj_acme · ap-southeast-1
tickets-2026.jsonlsynced
01

Upload or connect

A JSONL of conversations. Validation rejects a bad row with its line number before any GPU spends.

Qwen3-8B32K · rank 64
Qwen3.6 35Bdefault256K · rank 64
Llama 3.3 70B128K · rank 32
02

Choose a base

Every trainable model carries a context window and a LoRA rank ceiling. The default is Qwen3.6 35B.

ftjob_7Kd1prunning
Epoch 2 of 342%
step 120 · loss 0.94
val loss rising
03

Managed GPUs train

The job moves from queued to running. Progress is a fraction, and the events stream carries the loss.

eval_2Wpexact_match
0.813scored
0.642baseline
n = 500
+26.6% over baseline
04

Score the checkpoint

An evaluation with compare_to turns one score into a before and after.

47 trainable bases, six modalities, three rank ceilings

47 of the 49 catalogue models accept a LoRA adapter. A fine-tuning job on a JSONL of conversations starts from one of the 26 text bases.

ModelContextLoRA rank
Rank 64
Nemotron 3.5 Lightningnvidia/Nemotron-3.5-Lightning
131K64
GPT-OSS 20Bopenai/gpt-oss-20b
131K64
Nemotron 3 Nano 30Bnvidia/Nemotron-3-Nano-30B-A3B
66K64
Qwen3 8BQwen/Qwen3-8B
33K64
Qwen3.8 27BQwen/Qwen3.8-27B-A3B
262K64
Qwen3.6 35BQwen/Qwen3.6-35B-A3B
262K64
Rank 32
GLM 5.3 Flashzai-org/GLM-5.3-Flash
1M32
GPT-OSS 120Bopenai/gpt-oss-120b
131K32
Mistral Medium 3.5mistralai/Mistral-Medium-3.5
262K32
DeepSeek V4 Flash 0731deepseek-ai/DeepSeek-V4-Flash-0731
1M32
Llama 3.3 70Bmeta-llama/Llama-3.3-70B
131K32
Rank 16
MiniMax M3MiniMaxAI/MiniMax-M3
1M16
DeepSeek V3.2deepseek-ai/DeepSeek-V3.2
131K16
DeepSeek V3.1deepseek-ai/DeepSeek-V3.1
131K16
GLM 5.2zai-org/GLM-5.2
1M16
Kimi K2.6moonshotai/Kimi-K2.6
131K16
GLM 5.3zai-org/GLM-5.3
1M16
Nemotron 3 Ultra 550Bnvidia/Nemotron-3-Ultra-550B-A55B
66K16
DeepSeek V4 Prodeepseek-ai/DeepSeek-V4-Pro
1M16
Explore the managed fine-tuning guide
Results

Results from LLMs powered by Nucleus

$73.4K

monthly revenue misattributed

Ad Performance Tracking

47

trucks rerouted while still moving

Intelligent Fleet Optimisation

$177K

downtime prevented this month

Predictive Maintenance

14%

yield shortfall traced to one cause

Agriculture Yield Analysis

Stories

Workflows driven by a fine-tuned model

Each of the Workflows runs on a custom model trained on Nucleus.

Fleet operations

47 trucks rerouted while they were still moving

$2.4Kdaily fuel saved across the fleet
23delays resolved without a call
Marketing

$73.4K a month was being credited to the wrong channel

$73.4Kmonthly revenue misattributed
4.12xtrue ROAS against 2.8x reported
Manufacturing

Line 3's next failure, booked into the maintenance window

$177Kdowntime prevented this month
18.3 daysadvance notice on a failure

Fine-tuning at a glance

Data

  • JSONL, one conversation per line, with system, user and assistant roles
  • Validation before submit, with the rejected line and the reason
  • A held-out validation file for val_loss and the overfit warning
  • Connect a source and the training set keeps itself current

Bases

  • 47 trainable models across six modalities
  • A LoRA rank ceiling of 16, 32 or 64 per model
  • Namespaced ids such as Qwen/Qwen3-8B
  • Withdrawal dates published on the model ahead of time

Training

  • sft and dpo
  • Epochs, learning rate, batch size and rank, all optional
  • Progress as a fraction and an events stream with the loss
  • Cancel at any point and pay for the tokens that ran

After the run

  • result_model as a nucleus:// path
  • final_loss and val_loss on the job
  • An evaluation with compare_to against the base
  • A sampler on the OpenAI-compatible route, or export the weights
Pricing

Training tokens by model group

Fine-tuning jobs and training-loop steps are metered per million tokens processed, whichever way you train.

GroupTrainable basesFromTo
Language19$0.22$5.60
Vision7$0.35$6.00
Embedding & reranking6$0.03$0.09
Image generation4$0.50$1.40
Video generation3$1.40$1.80
Audio8$0.28$0.90

USD per million training tokens. Mixture-of-experts models are priced by their active parameters.

Pick your path

More in Train

Training loop

Docs

RL, DPO and distillation as API calls

Explore Training loop

Data connectors

Your systems as training data

Explore Data connectors

Models

Open bases across six modalities

Explore Models

Bring your JSONL. The first fine-tune runs in minutes.

Managed GPUs, a scored checkpoint, and the weights are yours.