Nvidia Just Published ChatQA 1.5, a Llama3 QA/RAG Finetune

#1

Post by noiseinvacuum » Thu, May 02, 2024, 6:01 PM UTC

Nvidia Just Published ChatQA 1.5, a Llama3 QA/RAG Finetune
huggingface.co

Re: Nvidia Just Published ChatQA 1.5, a Llama3 QA/RAG Finetune

#2

Post by noiseinvacuum » Thu, May 02, 2024, 6:01 PM UTC

Looks quite a solid fine tune from benchmark results. I iwsh they had added benchmark comparison to GPT 4 Turbo as well.

From the huggingface page:

Model Details

We introduce Llama3-ChatQA-1.5, which excels at conversational question answering (QA) and retrieval-augumented generation (RAG). Llama3-ChatQA-1.5 is built using the training recipe from ChatQA (1.0), and it is built on top of Llama-3 base model. Additionally, we incorporate more conversational QA data to enhance its tabular and arithmatic calculation capability. Llama3-ChatQA-1.5 has two variants: Llama3-ChatQA-1.5-8B and Llama3-ChatQA-1.5-70B. Both models were originally trained using Megatron-LM, we converted the checkpoints to Hugging Face format. Other Resources

Re: Nvidia Just Published ChatQA 1.5, a Llama3 QA/RAG Finetune

#3

Post by jaggs » Thu, May 02, 2024, 7:44 PM UTC

noiseinvacuum wrote: post #2

Looks quite a solid fine tune from benchmark results. I iwsh they had added benchmark comparison to GPT 4 Turbo as well. From the huggingface page: Model Details We introduce Llama3-ChatQA-1.5, which excels at conversational question answering (QA) and retrieval-augumented generation (RAG). Llama3-ChatQA-1.5 is built using the training recipe from ChatQA (1.0), and it is built on top of Llama-3 base model. Additional…

Any chance of getting this in a .gguf format?

Re: Nvidia Just Published ChatQA 1.5, a Llama3 QA/RAG Finetune

#4

Post by sp332 » Thu, May 02, 2024, 11:27 PM UTC

jaggs wrote: post #3

noiseinvacuum wrote: post #2

Looks quite a solid fine tune from benchmark results. I iwsh they had added benchmark comparison to GPT 4 Turbo as well. From the huggingface page: Model Details We introduce Llama3-ChatQA-1.5, which excels at conversational question answering (QA) and retrieval-augumented generation (RAG). Llama3-ChatQA-1.5 is built using the training recipe from ChatQA (1.0), and it is built on top of Llama-3 base model. Additional…

Any chance of getting this in a .gguf format?

https://huggingface.co/models?search=Llama3-ChatQA-1.5%20ggu...

Bartowski and neopolita have the most options. DuckyBlender's repo is currently empty.

Re: Nvidia Just Published ChatQA 1.5, a Llama3 QA/RAG Finetune

#5

Post by jaggs » Fri, May 03, 2024, 9:46 AM UTC

sp332 wrote: post #4

jaggs wrote: post #3

Earlier quoted context omitted.

Any chance of getting this in a .gguf format?

https://huggingface.co/models?search=Llama3-ChatQA-1.5%20ggu... Bartowski and neopolita have the most options. DuckyBlender's repo is currently empty.

Thanks very much, that's great.