Needs to consider
- LLM and inference workloads
- GPU assessment with real application loads
- Model-to-application connection
Workload and model
Share the model/version, data sensitivity, expected concurrency and response-time target. Memory needs depend on input length and execution behaviour. One sample test does not guarantee performance for all production traffic.
Data and operations
Review access methods, authentication, errors and where inputs, outputs and logs are retained. Consider GPU, storage, networking and management together. Define acceptance and support scope in the proposal. Do not submit customer data or access keys through the general form.
Related Disket services
Share your current systems and goals
We can assess product, integration or infrastructure requirements together. Exact service and fee conditions are defined in the applicable product information and proposal process.
Updated: