llama-bigasp-prompt-enhancer_fancyfeast_gguf

GGUF versions of Fancyfeast's Llama Bigasp Prompt Enhancer model

All credits (and then some) go to them, they do an amazing and very much needed job in the community.

The models' precision is f16 and I've created only three quantization variants: Q8_0, Q6_0 and Q4_K_M (because I wanted to test them, and in 99% of the cases I end up using one of them, depending on model size)

The conversion was done with version b4575 of llama-cpp.

---

Extra info:

User khronex has written a mini guide on how to run this with Ollama. If you end up using my quants with their guide you'll have to change some steps (as the guide downloads the original model files and quantizes those).

Here are the default parameters for the model at the moment (using khronex's info from the guide). But yeah, something something experimentation:

system_prompt = "You are a helpful image generation prompt writing AI. You write image generation prompts based on user requests. The prompt you write should be 150 words or longer."
temperature = 0.6
top_p = 0.9
stop_strings = [
    "<|start_header_id|>",
    "<|end_header_id|>",
    "<|eot_id|>",
]
max_tokens = 512
num_ctx = 4096