- for coding
- for generating stories, writing email, poems etc.
- good overall
- etc.
You must log in or register to comment.
Why does no one talk about the Yi models?
We’ve been fine-tuning models for specific applications like RAG and structured data extraction. Falcon – 7B has been really good for training. It’s both shifting to understand the target domains use of language from the training data, but also picking up instructions really well. Going to try mistral-7B soon for a comparison.
Same here. Switched to Mistral a few weeks ago. The results will blow you away, the difference is remarkable.
Oh god 🤦 But seriously we need a wiki with a leader board with votes😁
For generating full and uncensored stories (I provide a starting paragraph), collectivecognition-v1.1-mistral-7b has been by far the most creative and well written in my testing.
Best ones I tried were zephyr, dolphin, openorca, synthia and naberius.
Although I’m not a fan myself, I think is falcon…
I could just upvote the top comment, but yeah OpenHermes 2.5 is probably the best for now.
In no specific order:
Zephyr B OpenHermes2.5 OpenChat3.5
are these uncensored?
how is neural chat 7b v3 1? i will try out openhermes
Neural Chat 7b works fine with normal instructions for assistant use, but after trying to give it custom instructions for things like summarization, using code blocks or formatting, it completely broke. The same instructions that worked fine with other models I use. YMMV.
Orca2 7B (released just the other day) DEFINITELY competes with OpenHermes 2.5 but its hard to pick a clear winner (tho i would lean toward Orca2 myself)
Synthia Mystral 7B was pretty glorious but OpenHermes 2.5 is just better.
For Coding, DeepSeek coder 6.7b is exceptional
Is it exceptional in any language other than Python?
I’d like to know how it does in Java, Go, Rust and Zig as well as can it handle SQL quite well?
I haven’t had a chance to get hands on with DeepSeek yet. How does it compare to Code Llama?
In my opinion it’s amazing it’s close to gpt4
what hardware are you running it on? cpu/gpu, ram, etc? Trying to figure out what I need. My old gen 1 16 core threadripper with 64GB ram doesnt seem to work very well. Multiple minutes for a simple hello response. No GPU though, but looking to put a 6700XT GPU… not sure if that GPU will help a lot or what.
I was actually today comparing both (codellama 7B) and man codellama just gave crap, deepseek was vey accurate.
Models requiring remote code without any explanation are shady imo
elaborate
I for one just don’t trust these Chinese models at all. Not saying there’s anything wrong with this but it’s clear it’s aligned with the Chinese agenda when I try to ask it anything about Taiwan. But for coding it works good and you can run it offline
AFAIK models used to be just plain code, when you load one, for example, it would do so by calling a method pickled inside the model file. Uploader could set up this method to do practically anything they want, and it doesn’t need to be obviously malicious since code runs just like a normal python script. For example, it could simply load/render a webp image that is designed to use the recent libwebp vulnerability.
They changed this a while back, so now you need to pass an argument when loading the model to allow this behavior, and this model requires it.
Im a little new here, does DeepSeek coder 6.7b somehow phone home?
shady maybe, but it can code decent without depending on the internet. So theres that
For coding probably DeepSeek Coder Instruct 6.7B
For more general stuff I’ve found Zephyr 7B to be really good, but I still need to try OpenHermes 2.5
openchat 3.5 by pretty much all metrics
Mistral Instruct for overall