Add sigmas to Flux pipelines

[hlky]

What does this PR do?

This PR adds sigmas argument to Flux pipelines and removes timesteps argument as passing timesteps to FlowMatch schedulers is currently unsupported.

Fixes #10080

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@yiyixuxu @asomoza

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[SahilCarterr]

What do you think about writing a test to test out sigmas in FluxPipeline?
@hlky

[yiyixuxu]

thanks!

[vladmandic]

sigmas are supposed to be list-of-float, but pipeline internally uses np.array - passing sigmas right now causes a runtime error

return math.exp(mu) / (math.exp(mu) + (1 / t - 1) ** sigma)
TypeError: unsupported operand type(s) for /: 'int' and 'list'

which means that every instance of this line:

sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas

should at least be updated to

sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else np.array(sigmas)

[yiyixuxu]

thanks @vladmandic think we should probably handle this on scheduler.set_timesteps()
we already do that for custom timesteps

diffusers/src/diffusers/schedulers/scheduling_euler_discrete.py

Line 377 in 6db3333

timesteps = np.array(timesteps).astype(np.float32)

what do you think ? @hlky

[yiyixuxu]

@SahilCarterr do you want to help us add a test?