# Attempts

> Attempts helps companies diagnose where a model falls short on their own task, shape proprietary data, and measure whether tuning is worth deploying.

Training that is available today is supervised fine-tuning (SFT) with LoRA adapters. Preference tuning (DPO) and reinforcement learning are planned and not yet available. If training is not the right answer, Attempts says so. The customer owns their data and the tuned model. The base model keeps its original license. A dedicated GPU deployment uses an authenticated inference API and does not by itself place the model inside the customer's VPC or private network.

Contact: support@attempts.ai

## Product
- [Home](https://attempts.ai/): AI models trained on your data, evaluated against your baseline, and deployed on dedicated GPUs through an authenticated inference API.
- [About](https://attempts.ai/about/): Task diagnosis, data shaping, evaluation, and dedicated GPU deployment when the results justify it.
- [Training methods](https://attempts.ai/methods/): How a training method is matched to a task. SFT is available; DPO and reinforcement learning are not yet available.
- [Model catalog](https://attempts.ai/models/): Open-weight models Attempts can evaluate, tune, and deploy on dedicated GPUs.

## Documentation
- [Quickstart](https://attempts.ai/docs/): From sign-up to your first training run.
- [Preparing your dataset](https://attempts.ai/docs/datasets/): File format, required fields, limits and how data is split.
- [Training settings](https://attempts.ai/docs/training/): Models, method, epochs, evaluation and the cost estimate.
- [Billing and credits](https://attempts.ai/docs/billing/): Credits, your card, final charges, auto top-up, alerts and codes.
- [Using your model](https://attempts.ai/docs/serving/): Deploy a trained model and call it from your application.
- [Usage and API keys](https://attempts.ai/docs/usage-and-keys/): Reading your usage and managing workspace keys.
- [FAQ](https://attempts.ai/docs/faq/): Data handling, failed runs, refunds and support.

## Learning
- [Learning hub](https://attempts.ai/learn/): Practical guides to post-training, fine-tuning, evaluation, and cost.
- [Strategy & foundations](https://attempts.ai/learn/topics/strategy/): How models are built and adapted, open-weight versus proprietary, and when owning a tuned model makes sense.
- [Fine-tuning & post-training](https://attempts.ai/learn/topics/post-training/): Supervised fine-tuning, preference optimization, reinforcement learning and the data that feeds them.
- [Evaluation & quality](https://attempts.ai/learn/topics/evaluation/): Test sets, rubrics, LLM judges and the statistics that tell you whether a change actually helped.
- [Cost & efficiency](https://attempts.ai/learn/topics/cost/): Where LLM spend goes and how smaller tuned models, distillation, caching and serving choices reduce it.
- [Custom AI workflows](https://attempts.ai/learn/topics/workflows/): Prompting, retrieval, tuning and multi-step pipelines, and how to combine them for your task.
- [Tutorials](https://attempts.ai/learn/topics/tutorials/): Step-by-step implementation guides with working code.
- [Fine-tuning vs RAG vs prompt engineering: a practical comparison](https://attempts.ai/learn/fine-tuning-vs-rag-vs-prompt-engineering/): Fine-tuning vs RAG vs prompt engineering compared: what each changes, costs, when each wins, a decision flowchart, and how to test which works on your data.
- [How to build a custom evaluation framework for LLMs](https://attempts.ai/learn/llm-evaluation-framework/): Build an LLM evaluation framework for your own task: eval set design, sample sizes, metrics, calibrated LLM judges, regression suites and monitoring.
- [How to build custom AI workflows that improve output quality](https://attempts.ai/learn/custom-ai-workflows/): Design custom AI workflows that beat a single prompt: chaining, routing, verification, tuned models per step, observability, budgets and a data flywheel.
- [How to build domain-specific models from proprietary data](https://attempts.ai/learn/domain-specific-llm-proprietary-data/): A playbook for building a domain-specific LLM from proprietary data: task choice, data prep, leak-free splits, RAG vs fine-tuning and governance.
- [How to fine-tune an open-weight LLM with LoRA and QLoRA](https://attempts.ai/learn/fine-tune-llm-lora-qlora/): Fine-tune an open-weight LLM with LoRA and QLoRA: current TRL and PEFT code, memory math, 2025 hyperparameter evidence, evaluation and vLLM serving.
- [How to measure whether fine-tuning actually improved your model](https://attempts.ai/learn/measure-fine-tuning-improvement/): Measure fine-tuning improvement properly: strong baselines, a held-out test set, paired bootstrap and McNemar tests, sample sizes, regressions and cost.
- [How to reduce enterprise LLM costs without sacrificing quality](https://attempts.ai/learn/reduce-llm-costs/): A ranked playbook to reduce LLM costs: measurement, output limits, prompt caching, batch APIs, routing, tuned small models and self-hosting break-even math.
- [Model distillation: building smaller, cheaper, specialized models](https://attempts.ai/learn/llm-distillation/): How LLM distillation works: soft targets, sequence-level KD, teacher-generated data and on-policy distillation, plus when it fails and what API terms allow.
- [Open-weight vs proprietary LLMs: what should enterprises choose?](https://attempts.ai/learn/open-weight-vs-proprietary-llms/): Open-weight vs proprietary LLMs compared on licences, quality gap, cost at volume, data control and vendor risk, with a decision framework and checklist.
- [OpenAI fine-tuning is shutting down: how to migrate your fine-tuned models to open-weight models](https://attempts.ai/learn/migrate-off-openai-fine-tuning/): OpenAI fine-tuning shutdown dates verified, plus a dated checklist to export data and evals, re-train with LoRA, prove parity and cut over.
- [SFT vs DPO vs RL: understanding modern post-training methods](https://attempts.ai/learn/sft-vs-dpo-vs-reinforcement-learning/): SFT vs DPO vs RL compared precisely: the loss each optimizes, the data it needs, compute and stability trade-offs, failure modes and when to use which.
- [Synthetic data: how to generate training data for better models](https://attempts.ai/learn/synthetic-data-for-fine-tuning/): How to generate synthetic data for fine-tuning: generation methods, verification, dedup, decontamination, model collapse, licences, PII, plus working code.
- [The complete guide to LLM post-training in 2026](https://attempts.ai/learn/llm-post-training-guide/): LLM post-training explained: SFT, DPO, RLHF, GRPO, LoRA, data and evaluation, plus a decision framework for when it beats prompting or RAG.

## Policies
- [Privacy policy](https://attempts.ai/privacy/): What Attempts collects, where it goes, how long it is kept, and how to ask for deletion.