async example, and link documentation

[cbh123]

• update first example to point to Llama 3
• make it more explicit you can always check out a models API page (with link)
• provide an async example for an official (versionless) model call

[mattrothenberg]

I think I'm against using async here, given that there's an actual (and different asyncio) way of running predictions async in Python land! WDYT?

[cbh123]

Ah yes, that is a good point. I want some example of replicate.models.predictions.create on this page and wasn't sure how to refer to it — async felt like the best option? Open to other ideas on how to word it though

[mattrothenberg]

@cbh123 Here's how we word the difference on the playground:

The run() function returns the output directly, which you can then use or pass as the input to another model. If you want to access the full prediction object (not just the output), use the replicate.predictions.create() method instead. This will include the prediction id, status, logs, etc.

[bfirsh]

Look a bit below for the words we already use: https://github.com/replicate/replicate-python?tab=readme-ov-file#run-a-model-in-the-background

"Run a model in the background"

It also shows rather than tells you what happens so you can clearly understand what it does.

[bfirsh]

This might just want to be a part of that section, thinking about it, otherwise this is rather disjointed.

But, gosh, let's fix these APIs as soon as possible. How horrible. I'm tempted to say let's just use this energy to fix the API. Come grab me in Gather if you want to hash out a design.

[mattt]

replicate.predictions.create should have an overload that takes a model kwarg.

prediction = replicate.predictions.create(model="meta/meta-llama-3-8b-instruct", input={...})

The original models namespace indirection was a hedge, and I didn't have a chance to circle back to clean things up.

[mattt]

PR with that change here: #290

Please take a look and let me know what you think!