Chroma pipeline docstrings state defaults that differ from the signature

[iridescentWen]

Describe the bug

The three Chroma pipelines' __call__ docstrings state defaults to X for seven parameters where
the signature uses a different value. Reading the docstring and passing the documented number
explicitly therefore produces different behavior than omitting the argument.

All signature values below were read per-function via AST (not grep — encode_prompt in these files
declares some of the same parameter names with different defaults):

File	__call__	Parameter	Docstring says	Signature
pipeline_chroma.py	:589	num_inference_steps	50	35
pipeline_chroma.py	:589	guidance_scale	3.5	5.0
pipeline_chroma_img2img.py	:648	guidance_scale	3.5	5.0
pipeline_chroma_inpainting.py	:766	num_inference_steps	35	28
pipeline_chroma_inpainting.py	:766	guidance_scale	3.5	7.0
pipeline_chroma_inpainting.py	:766	strength	0.9	0.6
pipeline_chroma_inpainting.py	:766	max_sequence_length	512	256

Evidence that defaults to is meant to mirror the literal signature default in these docstrings: in
that same pipeline_chroma.py docstring, num_images_per_prompt (defaults to 1) and
max_sequence_length (defaults to 512) both match their signature exactly — only these seven drifted,
which is consistent with the docstrings having been carried between pipelines while each signature was
retuned.

Reproduction

import ast, inspect
from diffusers import ChromaPipeline

sig = inspect.signature(ChromaPipeline.__call__)
print(sig.parameters["guidance_scale"].default)   # 5.0
# docstring for the same parameter says "defaults to 3.5"

Logs

No response

System Info

Documentation-only issue; reproduced by reading main (@ 4e0466f3e when filed).

Who can help?

@stevhliu

Note

A fix is already open as #14578 (docstrings only, +7/-7 across the three files) — filing this so
that PR has a linked issue, per the pr-link-issue reminder. Not looking for anyone else to pick it
up; happy to adjust the PR instead if you would rather handle the wider pattern differently.

The same check finds ~190 more defaults to mismatches elsewhere in src/diffusers/, clustered by
pipeline family (num_inference_steps 50 → 28 in 13 files, guidance_scale 7.5 → 5.0 in 12, …). I
deliberately scoped #14578 to one family rather than sending a repo-wide sweep. If it would be more
useful as a mechanical utils/ check, I can propose that separately.