TailorLMBook a diagnostic
Menu
ServicesPricingProcessContact中文Book a diagnostic

SERVICES

Start with the problem,
not the model.

Four ways to work together. Not every problem needs training; start with a diagnostic, then go only as far as the evidence supports.

01

Diagnostic

Decide whether training is the right fix, and what “better” should mean.

$700¥4,8003–5 days

Deliverables

  • Problem framing: task, frequency, impact, current workaround
  • Data readiness: volume, quality, rights, test set
  • Baseline plan: model, metric, evaluation protocol
  • Cost estimate: data, training, evaluation, cloud GPU
  • Go/no-go: train, prompt/RAG, workflow change, or stop

Who it fits

You think you need training, but are not sure it is worth it; prompts and RAG still leave behavior unstable.

Not included

Data cleaning, full training, model deployment.

02

Data & evaluation

Turn raw material into trainable, testable data.

$2,500¥16,8001–2 weeks

Deliverables

  • Cleaning, deduplication, redaction, format validation
  • Train/test isolation and a held-out evaluation set
  • Evaluation items and scoring rules
  • Data quality report and rights record

Who it fits

You have raw material but no trainable data, no independent test set, or no confidence in data quality.

Not included

Training, model delivery, online deployment.

03

LoRA experiment

Validate the direction on a small budget before scaling.

$4,000¥28,8001–2 weeks

Deliverables

  • LoRA / QLoRA / SFT experiment
  • Before/after comparison on the fixed test set
  • Failure-case analysis
  • Recommendation: scale, change approach, or stop

Who it fits

You want a signal before committing to a full dataset and training budget.

Not included

Full training, production delivery, ongoing maintenance. GPU billed at cost.

04

Full delivery

A usable custom model, with a reproducible evaluation and deployment notes.

$9,500+¥68,000+3–6 weeks

Deliverables

  • Adapter or merged model
  • Independent evaluation report: baseline, post-training, regression, failures
  • Inference example and deployment notes
  • Known limitations and risks
  • Two revision rounds

Who it fits

You have a clear acceptance metric and need a model you can run, not a demo.

Not included

24/7 hosting, unlimited retraining, new scope. GPU billed at cost.

→How to choose

Not sure where
to start?

Not sure training is worth itDiagnostic
Messy data, no test setData & evaluation
Want a signal firstLoRA experiment
Need a production modelFull delivery

START HERE

Tell us what
is failing.

Describe the task, the current model, three real failures, and what “better” means. We will tell you whether a diagnostic makes sense.

Book a diagnostic