[Feature] Support nvidia_Llama3-ChatQA-1.5-8B

[ThiloteE]

Feature Request

Consider supporting https://huggingface.co/nvidia/Llama3-ChatQA-1.5-8B. According to its model card it is pretty good at conversational question answering (QA) and retrieval-augmented generation (RAG), beats the original llama-3-8b-instruct on selected benchmarks by a large margin and should play well with LocalDocs. I initially was looking at https://huggingface.co/cerebras/Llama3-DocChat-1.0-8B, but then realized it has a non-commercial license. The one from Nvidia is slightly worse on the benchmarks, but has the meta license for llama-3.

Unfortunately, the model only supports a context of 8k.