Timeline for RAG for sophisticated question-answering (Q&A) chatbots
Current License: CC BY-SA 4.0
4 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 24 at 16:42 | comment | added | Danielle França | The most common way to evaluate is through benchmarks. With each model release, it's disclose its performance on a few well-known benchmarks. Each benchmark is tied to a particular skill of the model; for instance, HellaSwag is common for evaluating a model's reasoning abilities. You can also create your own specific benchmarks to test their performance in your problem, a few frameworks helpt with this, such as Langchain or RAGAS. In terms of human evaluation, the platform I am familiar with is ChatArena. There, numerous models are available for testing, and humans rate their performances. | |
| Mar 24 at 5:07 | comment | added | cogitoergosum | What metrics are generally used to measure the success of an LLM for a given task? Or, only human eye-ball confirmation? | |
| Mar 24 at 5:06 | vote | accept | cogitoergosum | ||
| Mar 23 at 13:16 | history | answered | Danielle França | CC BY-SA 4.0 |