QuantFactory/Halu-8B-Llama3-v0.35-GGUF

QuantFactory/Halu-8B-Llama3-v0.35-GGUF

This is quantized version of Hastagaras/Halu-8B-Llama3-v0.35 created using llama.cpp

Model Description

This model has a similar steps to the Halu 0.3 model, but it utilizes a different base model.

The structure is as follows:

Base Model: [Daredevil](https://huggingface.co/mlabonne/Daredevil-8B)
|
Adapter: [HaluStory]
|
Adapter: [HaluConversation]

The overall structure is essentially the same, with the primary difference being the base model used. While the 0.3 model employed the Iterative-DPO model, this particular model is built upon the Daredevil model.

Despite the change in the base model, the adapters remain the same. After some testing, the 0.3 version is better at following instructions.

NOTES: You can see the difference between the 0.3 and 0.35 when you use 0 temperature and ask, "Who developed you?"

This model's answer will be something like "Developed by Meta AI," while the 0.3 version will answer like this: "Developed by OpenAI based on GPT-3."

So, from now on, all the Halu arc that end with 0.x0 will be based on the SFR Iterative DPO, while the 0.x5 versions will be based on any Llama3 Instruct-based model (in this case, it is Daredevil).

On the other hand... the Anjir Model responds like this: "Developed by Meta AI based on GPT-3." Hehe, the baukit slerp surgery is a success, I guess.