Quantization / TST: Fix remaining quantization tests

[younesbelkada]

What does this PR do?

This PR fixes the remaining quantization tests( quanto quantized cache) + makes sure the GGUF tests are run (they all pass on the T4 VM) by removing it from the transformers-all-latest-gpu docker image to the quantization docker image

========================================================================================================= 11 passed, 3 warnings in 506.17s (0:08:26) ==========================================================================================================

cc @SunMarc @ydshieh @zucchini-nlp

[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.

[zucchini-nlp]

Oh, for quanto it's hardware differences I guess 🥲
Thanks for fixing!

[ydshieh]

Thanks. BTW, what's the role of gguf here?

I mean shouldn't we have some extra require_xxx for class QuantoKVCacheQuantizationTest(unittest.TestCase): ?

[younesbelkada]

For GGUF it is just that the tests were currently compltely skipped, these tests are independent from the quanto cache tests !

[ydshieh]

OK. But my question is still there: we don't need any require_xxx for QuantoKVCacheQuantizationTest? It looks like it requires some quantize libraries being installed, no?

[SunMarc]

LGTM ! +1 for @ydshieh comments. Let's add a @require_quanto for the kv-cache quantization tests