Sun Yiqun

> There is already a big guide and migration guide ☺️ check if it's enough, thanks as always! I saw the migration one. Is there another guide other than the...

Currently, I think no. I think parameterizing a cypher query is just for performance, so you may simply format the string yourself. I guess it will not be a big...

> Hi @dukesun99, any updates on this? I'm experiencing exactly the same problem. Hi, no, I switched to Python to deal with the dump.

I could figure out which part of the code to modify to make it work, however, I cannot modify it neatly, since some processing is required for force layouts, and...

@lorr1 Thank you for your answer! Yes, I currently use nltk to split the doc into sentences as a workaround. Not sure if there exists such a dataset, but maybe...

> @AmirLayegh `Entity.jsonl` gives you a mapping from the internal entity ID (which is the line number) to Wikipedia page ID. You can then fetch the corresponding Wikidata QID with...

Does anyone know how new Bing actually handles such long context beyond 4k tokens? This might be the tricky part of web integration...

You may try `model.disable_adapter_layers()` or set `merge_weights=false` in `peft_config`. After `disable_adapter_layers()`, according to the code, if the original weights were modified, after a forward pass it will be reverted. Or...

Does a non-peft version work? Perhaps the generation parameters are not compatible with each other. You may check different decoing methods [here](https://huggingface.co/docs/transformers/main_classes/text_generation). It seems contrastive search has the highest priority...

This [repo](https://github.com/tloen/alpaca-lora) fine-tuned a lora of llama with alpaca data. Maybe you may check their performance and configs.