If you are asking for help, separate these parts of the experience:

• Time from sending the request to the first visible output.
• Time spent generating the rest of the answer.
• Whether the first request behaves differently from later requests.
• Whether the issue appears with a short input, a long input, or both.
• What changes when other applications or requests are running.

Include your hardware, runtime and model versions, exact model identifier, settings, and a sanitized test input. Record the command or method used for each timing.

If you reply with a proposed test, change one condition at a time and explain what the result would help rule out. Please post follow-up observations, including tests that did not improve the situation.

• Environment: Hardware, operating system, runtime version, and background workload.
• Model: Exact identifier, quantization, and relevant settings.
• Test: The same shareable inputs and success criteria for each candidate.
• Procedure: Number of runs, run order, and whether loading time is included.
• Results: Individual timings, quality checks, and failed runs.
• Limits: What this test does not establish.

Keep quality judgments separate from speed measurements. If an answer is short but incomplete, explain that in the results instead of treating the shorter duration as an automatic improvement.

Which fields are missing? Share a completed worksheet or suggest changes that would make a comparison easier for another member to repeat.

• Operating system and runtime version.
• CPU, GPU, system RAM, and GPU memory.
• Exact model name or identifier and quantization, if known.
• Configured context length and number of simultaneous requests.
• A representative task and how you judge the output.
• Observed limitations, with the measurement method if you report timings.

Mark any unknown details as unknown. If you have not measured performance, a description of your experience is still welcome; please distinguish it from a benchmark.

For planning, the Local Model Memory Estimator provides estimates. Include your actual observations when comparing an estimate with your own setup.