[Bugfix] Pass trust_remote_code_model=True for deepseek examples

[dsikka]

SUMMARY:

• Pass trust_remote_code_model=True for deepseek models
• Needed due to the slight differences in how tokenizer and processor pull down their relevant configs

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

[horheynm]

why do we need this now? what change caused this?

Needed due to the slight differences in how tokenizer and processor pull down their relevant configs

Sorry missed this

[kylesayrs]

@horheynm Both AutoConfig and tokenizers share the same cache for AutoConfig, meaning that if you pass trust_remote_code to the processor, you do not have the pass it into oneshot (specifically, the instantiation of AutoConfig in oneshot).

However, the AutoConfig cache is not shared between AutoConfig and processors. Previously we were getting away with not passing trust_remote_code_model because it was sharing from the tokenizer cache, but now we do since processor and AutoConfig no longer shared from the same cache.