12 GPU types · Per-second billing

The AI Training Platform Built for Scale

250+ models across 80+ architectures, 15+ fine-tuning methods, 6 alignment algorithms. Fine-tune LLMs and vision-language models on managed infrastructure. No token counting, billed per second.

$10 welcome creditBilled per secondDedicated GPU per job

12

GPU Types

Entry to enterprise, 24-288 GB VRAM

250+

Models Supported

LLMs, VLMs, and MoE across 80+ architectures

15+

Fine-Tuning Methods

LoRA, QLoRA, full fine-tune, and more

3

Regions

US, EU, and Asia-Pacific

Why USF BIOS

The deepest AI training platform

Everything you need to train, align, and deploy custom AI models, from adapter methods to managed infrastructure, with per-second billing and transparent pricing.

Deep Training Control

  • 6 alignment algorithms: DPO, SimPO, CPO, ORPO, KTO, Reward Modeling
  • 15+ fine-tuning methods: LoRA, QLoRA, AdaLoRA, LoHa, BOFT, and more
  • 120+ vision-language models for multimodal training
  • Megatron-LM, DeepSpeed, and sequence parallelism built in
  • Automatic checkpointing and fault recovery

Production-Ready Infrastructure

  • Fully managed GPU orchestration, no cluster setup needed
  • Dedicated GPU pod per job, no shared compute between users
  • Multi-region training, run jobs closest to your data
  • Team workspaces and shared model registry
  • Replaces 3-5 separate tools with one platform

Transparent Pricing

  • No token counting, pay only for GPU compute time used
  • Billed per second, stop anytime, no minimum commitments
  • Cost estimates shown before every training job starts
  • 12 GPU types from $0.44/hr, same rate for SFT, RLHF, VLM, and CPT
  • 70-90% cheaper than building your own training infrastructure

Training Capabilities

Every training method. Every model type.

Not just LoRA fine-tuning. Full-spectrum training from parameter-efficient adapters to preference alignment to continued pre-training, for both text and vision models.

15+ methods

Supervised Fine-Tuning

SFT with 15+ fine-tuning methods: LoRA, QLoRA, AdaLoRA, LoHa, LoKr, BOFT, ReFT, VeRA, and full parameter training up to 1T.

6 algorithms

RLHF & Alignment

DPO, SimPO, CPO, ORPO, KTO, and Reward Modeling. Align models to human preferences without building your own RL pipeline.

120+ VLMs

Vision-Language Models

Fine-tune 120+ vision-language models: InternVL, Qwen-VL, LLaVA, DeepSeek-VL, MiniCPM-V, Phi Vision, and more.

Domain-specific

Continued Pre-Training

Extend any foundation model with your proprietary data. Inject domain knowledge: medical, legal, financial, or code.

Deep Dive

More depth than any other managed platform

15+ fine-tuning methods, 6 alignment algorithms, 250+ models across 80+ architectures. The technical depth your ML team needs, without the infrastructure overhead.

15+ Fine-Tuning Methods

Parameter-efficient and full fine-tuning

From lightweight LoRA adapters to complete parameter updates. Pick the right method for your model size and budget.

LoRAQLoRAAdaLoRALoHaLoKrBOFTOFTVeRAReFTFull Fine-Tune

6 Alignment Algorithms

Preference optimization at scale

Align models to human preferences without building your own RL pipeline. No separate reward model needed for most methods.

DPOSimPOCPOORPOKTOReward Modeling

Online RL (PPO, GRPO, GKD) coming soon.

250+ Models

80+ architectures: LLMs, VLMs, and MoE

Every major open model family supported: text, vision-language, mixture-of-experts, reward models, and embeddings.

QwenLlamaDeepSeekPhiGemmaMistralInternVLLLaVAMixtralChatGLM

Across 80+ architectures, plus live Hugging Face Hub search.

Plan Your Training

Pick your method. See what you need.

Select your training goal and we show you the right adapters, data format, and supported models so you can plan before you start.

Fine-Tuning Methods (15+)

QLoRA (4-bit)Recommended
LoRA (8-bit)
AdaLoRA
LoHa
LoKr
BOFT
OFT
VeRA
FourierFT
BoNE
Bottleneck Adapter
ReFT
LLaMA-Pro
LongLoRA
Full Fine-Tune (bf16)

Supported Models

250+ LLMs120+ VLMsMoE Models

Required Data Format

Instruction-Response Pairs

Each example contains an instruction and the expected model response. Supports single-turn and multi-turn conversations.

Alpaca Format
{"instruction": "Summarize this article",
 "input": "The study found that...",
 "output": "Researchers discovered..."}
Conversational
{"messages": [
  {"role": "user", "content": "..."},
  {"role": "assistant", "content": "..."}
]}

Accepted File Types

JSONLParquetCSV

Why USF BIOS

Enterprise infrastructure without the enterprise overhead

We handle distributed training, GPU orchestration, checkpointing, and fault tolerance. Your team focuses on data quality and model performance.

Multi-GPU Training

From 1 to 8 GPUs per node. DeepSpeed and Megatron-LM built in, no cluster management needed.

Sequence Parallelism

Ulysses and ZigZag Ring Attention for long-context training efficiently across GPUs.

Automatic Checkpointing

Training state saved to S3 automatically. Resume from any checkpoint if a job is interrupted.

Multi-Region Availability

Train in the region closest to your data. US, EU, and Asia-Pacific availability.

Dedicated Compute

Each training job runs in its own GPU pod. No shared compute between users, your job gets the full GPU.

Time to Value

Go from raw dataset to production-ready model in hours, not months. No MLOps team required.

Platform Comparison

See how we compare

Honest, side-by-side comparison with the platforms you're evaluating. Every fact sourced from public documentation.

Swipe sideways to see every platform

CapabilityUSF BIOSFireworks AITogether AITinkerSageMaker
SFT (LoRA)DIY
Full Fine-Tuning-DIY
VLM Training120+ modelsLimited-LimitedDIY
Fine-Tuning Methods15+221DIY
RLHF Algorithms6 (9 soon)313DIY
Continued Pre-Training---DIY
Reward Modeling---DIY
Megatron-LM---DIY
Sequence ParallelismPartial--DIY
Models Supported250+~30~4022Any
Model Architectures80+~30~4022Any
Max Parameters1T1T100B550BAny
Pricing ModelPer GPU-hourPer GPU-hourPer GPU-hourPer GPU-hourPer GPU-hour
Managed InfraPartial
Setup RequiredNoneNoneNoneNoneSignificant

Data sourced from public documentation as of June 2026. "DIY" means the platform supports it if you build it yourself.

Pricing

Simple, transparent pricing

12 GPU types, billed per second. No token counting, no method surcharges. Every training type (SFT, RLHF, CPT, VLM) uses the same rate.

Find the right GPU for your model

Select your model and training configuration. We estimate memory with built-in headroom for optimizer states, evaluation passes, and peak memory spikes, so your run completes without interruption.

Loading live pricing and availability...

Sign up to see live pricing and availability.

How It Works

Production-quality models in three steps

Built for teams where model quality and reliability are non-negotiable. Go from raw data to a deployable model, with full control over every training decision.

01

Bring Your Data

Upload your dataset or import directly from HuggingFace. Supports JSONL, Parquet, CSV, and image datasets for vision-language training.

02

Configure Training

Select from 250+ models, 15+ fine-tuning methods, and 6 alignment algorithms. Choose your GPU, set your hyperparameters. Every decision is yours.

03

Train & Deploy

Launch training on managed GPUs with automatic checkpointing and fault recovery. Monitor in real time. Export to HuggingFace or SafeTensors when ready.

Built for serious AI teams

If your product depends on model quality, you need a training platform that gives you complete control, not a simplified wrapper with half the methods stripped out. Start training in minutes.

$10 welcome credit included. Per-second billing on all GPUs.