support two more calib datasets and fix embedding layer bug

[wenhuach21]

No description provided.

[Copilot]

Pull Request Overview

This PR introduces support for the ultrachat_200k dataset, extends dataset registration to multiple aliases, and refines the embedding quantization logic.

• Extend register_dataset decorator to accept multiple dataset names and integrate ultrachat_200k
• Import and standardize load_dataset usage across existing dataset functions
• Modify quantize_embedding_layer to return whether any layers were actually quantized

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 6 comments.

File	Description
auto_round/utils.py	Sort GGUF_CONFIG keys for deterministic ordering in _gguf_format
auto_round/calib_dataset.py	Added load_dataset import, multi-name registration, ultrachat_200k support, and hardcoded dataset fixes
auto_round/autoround.py	Introduce to_quantize flag and change return value of quantize_embedding_layer

Comments suppressed due to low confidence (1)

auto_round/autoround.py:783

• Changing the return value to to_quantize alters the previous always-True behavior. Downstream callers expecting True on completion may now misinterpret False as failure. Either update callers or restore the original return semantics and expose to_quantize via a separate API.

        return to_quantize

[wenhuach21]

The lambada_openai evaluation appears to have some issues maybe due to the update of datasets library and couldn't be reproduced locally. Merging for now; will address the problem in a future fix.