test VS benchmark

Sources [1] MMLU Benchmark (Multi-task Language Understanding) | Papers With Code https://paperswithcode.com/sota/multi-task-language-understanding-on-mmlu [2] MMLU Dataset | Papers With Code https://paperswithcode.com/dataset/mmlu [3] hendrycks/test: Measuring Massive Multitask Language Understanding | ICLR 2021 - GitHub https://github.com/hendrycks/test [4] lukaemon/mmlu · Datasets at Hugging Face https://huggingface.co/datasets/lukaemon/mmlu [5] [2009.03300] Measuring Massive Multitask Language Understanding - arXiv https://arxiv.org/abs/2009.03300

With a score of 90.0%, Gemini Ultra is the first model to outperform human experts on MMLU (massive multitask language understanding), which uses a combination of 57 subjects such as math, physics, history, law, medicine and ethics for testing both world knowledge and problem-solving abilities.

One method for comparison is the MMLU https://arxiv.org/abs/2009.03300.

Because there are many benchmarks that measure different things.

You need to look at the benchmark that reflects your specific interest.

So in this case ("I wasn't impressed that 30B didn't seem to know who Captain Picard was") the closest relevant benchmark they performed is MMLU (Massive Multitask Language Understanding"[1].

In the LLAMA paper they publish a figure of 63.4% for the 5-shot average setting without fine tuning on the 65B model, and 68.9% after fine tuning. This is significantly better that the original GPT-3 (43.9% under the same conditions) but as they note:

> "[it is] still far from the state-of-the-art, that is 77.4 for GPT code-davinci-002 on MMLU (numbers taken from Iyer et al. (2022))"

InstructGPT[2] (which OpenAI points at as most relevant ChatGPT publication) doesn't report MMLU performance.

[1] https://github.com/hendrycks/test

[2] https://arxiv.org/abs/2203.02155

Benchmark result is 67.6% which is 7.6% improvement from Gopher. MMLU is multiple choice Q&A over various subjects. Questions can be found linked in this github repo (see data).

I'm aware of one by Google that I used a couple times, but IMO it's better to capture real runtime data from a fully-operational process than to carve out the benchmarkable bits and test them in isolation, so I track information during program testing and print it all to a log instead of using things like that.

If you just want to do some quick benchmarks, you can just use std::chrono::high_resolution_clock::now(). Call it before the code that you are benchmarking and then immediately after. Take them away and you have your duration. If you want to use a proper benchmarking tool then I can totally recommend Google Benchmark. Fantastic benchmarking tool. Honourable mention would be Quick Bench which is an online tool that uses Google Benchmark.

I tried to install google benchmark(https://github.com/google/benchmark) in my ubuntu machine by :

We don't know if it's a lie until we verify it and that's not difficult, you have a quicksort implementation in a couple of languages, you'll need to pass the necessary parameters to show the time needed by a function call to execute to the compiler or interpreter or you may use use a library(like benchmark for C++) and you're good to go.

If you are uncertain about the performance characteristics of a function you should ALWAYS benchmark it. Googles Benchmark library is wonderful for quick micro benchmarks. For more complex things, perhaps look into profiling and then look at invocation counts of copy constructors.

Your compiler may be optimising away your loop, there. I typically use a micro-benchmarking tool for these types of tests. You could try Google Benchmark. It’s available in most OS’ package managers, but pretty easy to build from source if not

C++: C++ has quite a number of benchmarking libraries some of the recent ones involving C++ 20's flexibility. The most notable being Google Bench and UT. C does not have many specific benchmarking libraries, but you can easily integrate C code with C++ benchmarking libraries in order to test the performance of your C code.