this post was submitted on 14 Nov 2023
1 points (100.0% liked)

LocalLLaMA

1 readers
1 users here now

Community to discuss about Llama, the family of large language models created by Meta AI.

founded 10 months ago
MODERATORS
communick@poweruser.forum
1
Neural-chat-7b-v3-1 GGUF. New Mistral finetune (alien.top)
submitted 10 months ago by fakezeta@alien.top to c/localllama@poweruser.forum
6 comments fedilink hide all child comments
 

Couldn't wait for the great TheBloke to release it so I've uploaded a Q5_K_M GGUF of Intel/neural-chat-7b-v3-1.

From some preliminary test on PISA sample questions it seems at least on par with OpenHermers-2.5-Mistral-7B

https://preview.redd.it/bkaezfb51c0c1.png?width=1414&format=png&auto=webp&s=735d0f03109488e01d65c1cf8ec676fa7e18c1d5

you are viewing a single comment's thread
view the rest of the comments
[โ€“] AdamDhahabi@alien.top 1 points 10 months ago (1 children)

Interested to know how it scores for RAG use cases, there is a benchmark for that https://github.com/vectara/hallucination-leaderboard

Up to now, Mistral underperforms Llama2.

[โ€“] fakezeta@alien.top 1 points 10 months ago

Currently all the finetuned version of Mistral I've tested have a high rate of hallucination: this one also seems to have this tendency.