85 lines
3.5 KiB
Markdown
85 lines
3.5 KiB
Markdown
|
---
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
tags:
|
||
|
|
- text-generation-inference
|
||
|
|
- instruct
|
||
|
|
- conversational
|
||
|
|
- roleplay
|
||
|
|
- sillytavern
|
||
|
|
- gguf
|
||
|
|
- anime
|
||
|
|
- quantized
|
||
|
|
- mistral
|
||
|
|
license: cc-by-4.0
|
||
|
|
---
|
||
|
|
|
||
|
|
# **THIS VERSION IS NOW DEPRECATED. USE V3-0.2. V2 HAS PROBLEMS WITH ALIGNMENT AND THE NEW VERSION IS A SUBSTANTIAL IMPROVMENT!**
|
||
|
|
|
||
|
|
This repository hosts deprecated GGUF-IQ-Imatrix quants for [localfultonextractor/Erosumika-7B-v2](https://huggingface.co/localfultonextractor/Erosumika-7B-v2).
|
||
|
|
|
||
|
|
*"Better, smarter erosexika!!"*
|
||
|
|
|
||
|
|
[Quantized as per user request.](https://huggingface.co/Lewdiculous/Model-Requests/discussions/19)
|
||
|
|
|
||
|
|
Quants:
|
||
|
|
```python
|
||
|
|
quantization_options = [
|
||
|
|
"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]](https://github.com/ggerganov/llama.cpp/discussions/5006) [[2]](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384)
|
||
|
|
|
||
|
|
For imatrix data generation, kalomaze's `groups_merged.txt` with added roleplay chats was used, you can find it [here](https://huggingface.co/Lewdiculous/Datura_7B-GGUF-Imatrix/blob/main/imatrix-with-rp-format-data.txt). 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:
|
||
|
|
|
||
|
|
<h1 style="text-align: center">Erosumika-7B-v2</h1>
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
## Model Details
|
||
|
|
A DARE TIES merge between Nitral's [Kunocchini-7b](https://huggingface.co/Nitral-AI/Kunocchini-7b-128k-test), Epiculous' [Mika-7B](https://huggingface.co/Epiculous/Mika-7B) and my [FlatErosAlpha](https://huggingface.co/localfultonextractor/FlatErosAlpha), a flattened(in order to keep the vocab size 32000) version of tavtav's [eros-7B-ALPHA](https://huggingface.co/tavtav/eros-7B-ALPHA). In my brief testing, v2 is a significant improvement over the original Erosumika; I guess it won the DARE TIES lottery. Alpaca and Mistral seem to work best. Chat-ML might also work but I expect it to never end generations. Anything goes!
|
||
|
|
|
||
|
|
Due to it being an experimental model, there are some quirks...
|
||
|
|
|
||
|
|
- Rare occasion to misspell words
|
||
|
|
- Very rare occasion to have random formatting artifact at the end of generations
|
||
|
|
|
||
|
|
[GGUF quants](https://huggingface.co/localfultonextractor/Erosumika-7B-v2-GGUF)
|
||
|
|
|
||
|
|
## 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.
|
||
|
|
|
||
|
|
|
||
|
|
```yaml
|
||
|
|
base_model: localfultonextractor/FlatErosAlpha
|
||
|
|
models:
|
||
|
|
- model: localfultonextractor/FlatErosAlpha
|
||
|
|
- model: Epiculous/Mika-7B
|
||
|
|
parameters:
|
||
|
|
density: 0.5
|
||
|
|
weight: 0.25
|
||
|
|
- model: Nitral-AI/Kunocchini-7b
|
||
|
|
parameters:
|
||
|
|
density: 0.5
|
||
|
|
weight: 0.75
|
||
|
|
merge_method: dare_ties
|
||
|
|
dtype: bfloat16
|
||
|
|
```
|