[LLM:Bugfix] Fix WikiText calibration dataset loading

[LudovicoYIN]

Description

Fix LLM calibration dataset loading with recent datasets / huggingface_hub versions.

Recent HuggingFace Hub clients require dataset repo IDs to include a namespace. The old load_dataset("wikitext", ...) path now fails with an invalid HF URI error. This updates built-in WikiText calibration loading to Salesforce/wikitext.

Also updates download_data.py to accept dataset specs like Salesforce/wikitext/wikitext-2-raw-v1 by treating the final path component as the dataset config and the preceding components as the repo ID.

Module

LLM

Type

• Feature
• Bugfix
• Perf
• Refact
• Style
• Doc
• Test
• Chore

Checklist

• Commit message follows [Module:Type] Description format
• Code compiles without errors
• Tested on relevant platform(s)
• No unrelated format or style changes included

[wangzhaode]

Thanks for the fix! The root cause is confirmed — recent HuggingFace Hub versions require a namespaced repo ID, and the old load_dataset('wikitext', ...) call fails with HfUriError. The updated Salesforce/wikitext identifier and the new three-part spec parsing are both correct.

[LudovicoYIN]

Thanks your review.