language:
- en
pipeline_tag: text-generation
tags:
- text-generation-inference
- instruct
- conversational
- roleplay
- mistral
- merge
- sillytavern
license: cc-by-4.0
This repo contains GGUF-IQ-Imatrix quantized model files for Erosumika-7B-v3-0.2.
Recommended starting SillyTavern presets here.
Quants:
"Q4_K_M", "Q4_K_S", "IQ4_XS", "Q5_K_M", "Q5_K_S",
"Q6_K", "Q8_0", "IQ3_M", "IQ3_S", "IQ3_XXS"
What does "Imatrix" mean?
It stands for Importance Matrix, a technique used to improve the quality of quantized models. The Imatrix is calculated based on calibration data, and it helps determine the importance of different model activations during the quantization process. The idea is to preserve the most important information during quantization, which can help reduce the loss of model performance, especially when the calibration data is diverse. [1] [2]
For imatrix data generation, kalomaze's groups_merged.txt with added roleplay chats was used, you can find it here. This was just to add a bit more diversity to the data.
Steps:
Base⇢ GGUF(F16)⇢ Imatrix-Data(F16)⇢ GGUF(Imatrix-Quants)
Using the latest llama.cpp at the time.
Original model information:
Erosumika-7B-v3-0.2
~Mistral 0.2 Edition~
Model Details
The Mistral 0.2 version of Erosumika-7B-v3, a DARE TIES merge between Nitral's Kunocchini-7b, Endevor's InfinityRP-v1-7B and my FlatErosAlpha, a flattened(in order to keep the vocab size 32000) version of tavtav's eros-7B-ALPHA. Alpaca and ChatML work best. Slightly smarter and better prompt comprehension than Mistral 0.1 Erosumika-7B-v3. 32k context should work.
Limitations and biases
The intended use-case for this model is fictional writing for entertainment purposes. Any other sort of usage is out of scope. It may produce socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive. Outputs might often be factually wrong or misleading.
merge_method: task_arithmetic
base_model: alpindale/Mistral-7B-v0.2-hf
models:
- model: localfultonextractor/Erosumika-7B-v3
parameters:
weight: 1.0
dtype: float16