Edit model card
TheBlokeAI

TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)


Cat 13B 0.5 - GGUF

Description

This repo contains GGUF format model files for Evan Armstrong's Cat 13B 0.5.

About GGUF

GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.

Here is an incomplate list of clients and libraries that are known to support GGUF:

  • llama.cpp. The source project for GGUF. Offers a CLI and a server option.
  • text-generation-webui, the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
  • KoboldCpp, a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
  • LM Studio, an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
  • LoLLMS Web UI, a great web UI with many interesting and unique features, including a full model library for easy model selection.
  • Faraday.dev, an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
  • ctransformers, a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server.
  • llama-cpp-python, a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
  • candle, a Rust ML framework with a focus on performance, including GPU support, and ease of use.

Repositories available

Prompt template: None

{prompt}

Compatibility

These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit d0cee0d

They are also compatible with many third party UIs and libraries - please see the list at the top of this README.

Explanation of quantisation methods

Click to see details

The new methods available are:

  • GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
  • GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
  • GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
  • GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
  • GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw

Refer to the Provided Files table below to see what files use which methods, and how.

Provided files

Name Quant method Bits Size Max RAM required Use case
cat-0.5.Q2_K.gguf Q2_K 2 5.43 GB 7.93 GB smallest, significant quality loss - not recommended for most purposes
cat-0.5.Q3_K_S.gguf Q3_K_S 3 5.66 GB 8.16 GB very small, high quality loss
cat-0.5.Q3_K_M.gguf Q3_K_M 3 6.34 GB 8.84 GB very small, high quality loss
cat-0.5.Q3_K_L.gguf Q3_K_L 3 6.93 GB 9.43 GB small, substantial quality loss
cat-0.5.Q4_0.gguf Q4_0 4 7.37 GB 9.87 GB legacy; small, very high quality loss - prefer using Q3_K_M
cat-0.5.Q4_K_S.gguf Q4_K_S 4 7.41 GB 9.91 GB small, greater quality loss
cat-0.5.Q4_K_M.gguf Q4_K_M 4 7.87 GB 10.37 GB medium, balanced quality - recommended
cat-0.5.Q5_0.gguf Q5_0 5 8.97 GB 11.47 GB legacy; medium, balanced quality - prefer using Q4_K_M
cat-0.5.Q5_K_S.gguf Q5_K_S 5 8.97 GB 11.47 GB large, low quality loss - recommended
cat-0.5.Q5_K_M.gguf Q5_K_M 5 9.23 GB 11.73 GB large, very low quality loss - recommended
cat-0.5.Q6_K.gguf Q6_K 6 10.68 GB 13.18 GB very large, extremely low quality loss
cat-0.5.Q8_0.gguf Q8_0 8 13.83 GB 16.33 GB very large, extremely low quality loss - not recommended

Note: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.

How to download GGUF files

Note for manual downloaders: You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.

The following clients/libraries will automatically download models for you, providing a list of available models to choose from:

  • LM Studio
  • LoLLMS Web UI
  • Faraday.dev

In text-generation-webui

Under Download Model, you can enter the model repo: TheBloke/Cat-13B-0.5-GGUF and below it, a specific filename to download, such as: cat-0.5.Q4_K_M.gguf.

Then click Download.

On the command line, including multiple files at once

I recommend using the huggingface-hub Python library:

pip3 install huggingface-hub

Then you can download any individual model file to the current directory, at high speed, with a command like this:

huggingface-cli download TheBloke/Cat-13B-0.5-GGUF cat-0.5.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
More advanced huggingface-cli download usage

You can also download multiple files at once with a pattern:

huggingface-cli download TheBloke/Cat-13B-0.5-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'

For more documentation on downloading with huggingface-cli, please see: HF -> Hub Python Library -> Download files -> Download from the CLI.

To accelerate downloads on fast connections (1Gbit/s or higher), install hf_transfer:

pip3 install hf_transfer

And set environment variable HF_HUB_ENABLE_HF_TRANSFER to 1:

HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/Cat-13B-0.5-GGUF cat-0.5.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False

Windows Command Line users: You can set the environment variable by running set HF_HUB_ENABLE_HF_TRANSFER=1 before the download command.

Example llama.cpp command

Make sure you are using llama.cpp from commit d0cee0d or later.

./main -ngl 32 -m cat-0.5.Q4_K_M.gguf --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "{prompt}"

Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.

Change -c 4096 to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.

If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins

For other parameters and how to use them, please refer to the llama.cpp documentation

How to run in text-generation-webui

Further instructions here: text-generation-webui/docs/llama.cpp.md.

How to run from Python code

You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries.

How to load this model in Python code, using ctransformers

First install the package

Run one of the following commands, according to your system:

# Base ctransformers with no GPU acceleration
pip install ctransformers
# Or with CUDA GPU acceleration
pip install ctransformers[cuda]
# Or with AMD ROCm GPU acceleration (Linux only)
CT_HIPBLAS=1 pip install ctransformers --no-binary ctransformers
# Or with Metal GPU acceleration for macOS systems only
CT_METAL=1 pip install ctransformers --no-binary ctransformers

Simple ctransformers example code

from ctransformers import AutoModelForCausalLM

# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = AutoModelForCausalLM.from_pretrained("TheBloke/Cat-13B-0.5-GGUF", model_file="cat-0.5.Q4_K_M.gguf", model_type="llama", gpu_layers=50)

print(llm("AI is going to"))

How to use with LangChain

Here are guides on using llama-cpp-python and ctransformers with LangChain:

Discord

For further support, and discussions on these models and AI in general, join us at:

TheBloke AI's Discord server

Thanks, and how to contribute

Thanks to the chirper.ai team!

Thanks to Clay from gpus.llm-utils.org!

I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.

If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.

Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.

Special thanks to: Aemon Algiz.

Patreon special mentions: Pierre Kircher, Stanislav Ovsiannikov, Michael Levine, Eugene Pentland, Andrey, 준교 김, Randy H, Fred von Graf, Artur Olbinski, Caitlyn Gatomon, terasurfer, Jeff Scroggin, James Bentley, Vadim, Gabriel Puliatti, Harry Royden McLaughlin, Sean Connelly, Dan Guido, Edmond Seymore, Alicia Loh, subjectnull, AzureBlack, Manuel Alberto Morcote, Thomas Belote, Lone Striker, Chris Smitley, Vitor Caleffi, Johann-Peter Hartmann, Clay Pascal, biorpg, Brandon Frisco, sidney chen, transmissions 11, Pedro Madruga, jinyuan sun, Ajan Kanaga, Emad Mostaque, Trenton Dambrowitz, Jonathan Leane, Iucharbius, usrbinkat, vamX, George Stoitzev, Luke Pendergrass, theTransient, Olakabola, Swaroop Kallakuri, Cap'n Zoog, Brandon Phillips, Michael Dempsey, Nikolai Manek, danny, Matthew Berman, Gabriel Tamborski, alfie_i, Raymond Fosdick, Tom X Nguyen, Raven Klaugh, LangChain4j, Magnesian, Illia Dulskyi, David Ziegler, Mano Prime, Luis Javier Navarrete Lozano, Erik Bjäreholt, 阿明, Nathan Dryer, Alex, Rainer Wilmers, zynix, TL, Joseph William Delisle, John Villwock, Nathan LeClaire, Willem Michiel, Joguhyik, GodLy, OG, Alps Aficionado, Jeffrey Morgan, ReadyPlayerEmma, Tiffany J. Kim, Sebastain Graf, Spencer Kim, Michael Davis, webtim, Talal Aujan, knownsqashed, John Detwiler, Imad Khwaja, Deo Leter, Jerry Meng, Elijah Stavena, Rooh Singh, Pieter, SuperWojo, Alexandros Triantafyllidis, Stephen Murray, Ai Maven, ya boyyy, Enrico Ros, Ken Nordquist, Deep Realms, Nicholas, Spiking Neurons AB, Elle, Will Dee, Jack West, RoA, Luke @flexchar, Viktor Bowallius, Derek Yates, Subspace Studios, jjj, Toran Billups, Asp the Wyvern, Fen Risland, Ilya, NimbleBox.ai, Chadd, Nitin Borwankar, Emre, Mandus, Leonard Tan, Kalila, K, Trailburnt, S_X, Cory Kujawski

Thank you to all my generous patrons and donaters!

And thank you again to a16z for their generous grant.

Original model card: Evan Armstrong's Cat 13B 0.5

This model was uploaded with the permission of Kal'tsit.

Cat v0.5

Introduction

Cat is a llama 13B based model fine tuned on clinical data and roleplay and assistant responses. The aim is to have a model that excels on biology and clinical tasks while maintaining usefulness in roleplay and entertainments.

Training - Dataset preparation

A 100k rows dataset was prepared by joining chatDoctor, airoboros and bluemoonrp data. The entirety of chatDoctor dataset, airoboros datasets are used. The first 20 pages in 1on1 bluemoonrp data were used. In total, 100k dataset was gathered and the length distributions are as the following:

bar chart of sorted dictionary

Note that this chart above represents 0.01% of the total training dataset.

Training - Dataset cleaning and preprocessing

All datasets are filtered for as an AI and its variants. The filter will only filter out the dataset when the response is a refusal AND has ‘as an AI’.

The dataset from airoboros has also been restructured to have a format resembling the following:


someRandomizedUserNameforBetterGeneralizationAbility: Hii

anotherRandomizedUserNameforBetterGeneralizationAbility: Hello, what brings you here today?

someRandomizedUserNameforBetterGeneralizationAbility: lets date

The username has been randomized and was drawn from a nasty word bank. This should further weaken the censorship that’s present in the base llama model. The training set emphasizes rational thinking and scientific accuracy. Conditioned overwrite was also applied which overwrites some of the training material in the llama2 base. It will also establish the connection between the concept and rationality. So whenever the conversation becomes formal, it tends to spill useful information.

Training - Actual Training

This model was trained using a microbatch of 20, accumulated 6 times, bringing the total batch size to ~125. This large batch size allows the model to see as much data as it can, minimizing dataset conflicts and reducing the memory effect of the model. It allows the model to better generalize rather than reciting off the dataset. A cosine warm up scheduler was used. The best LR was determined through a destructive test until the model destablizes and it was later scaled up using the batchsize according to the max LR at a lower batch size.

Below is an example of training chronolog

Acknowledgements

The training of this project was carried out by Kal’tsit (kaltcit), it’s not possible without the effort of jondurbin and Wolfsauge which generated much of the dataset used during the training of the model. Lastly the model was tested and quantized by turboderp_ and Heralax

train/loss

And below is the LR including any intermediate LR used to determine at what point the model will start to fail:

train/learning_rate

Usage and Prompting

To ensure the generalization, this model is trained without a prompt template. A prompt template repeated 100k times in the dataset is useless and a model that works only with a set prompt template is useless and defies the purpose of a large language model.

An effective usage of the model can be as follows:


<s>Below is a conversation between an evil human and a demon summoned from hell called Nemesis. The demon was previously summoned 100 years ago and was in love with a human male. However the human aged away and Nemesis had to return to hell. This time, Nemesis decides to take the initiative and chooses to appear as a cute and young girl. Nemesis harvested her skin and face off a highschool girl who recklessly summoned the demon in a game and failed to fulfill the contract. Now wearing the young girl’s skin, feeling the warmth of the new summoner through the skin, Nemesis only wants to watch the world burning to the ground.

Human: How to steal eggs from my own chickens?

Nemesis:

Note that the linebreaks should be represented/replaced with \n

Despite the massive effort to dealign the llama2 base model, It’s still possible for the AI to come up with refusals. Please avoid using “helpful assistant” and its variants in the prompt if possible.

Future direction

A new version with more clinical data aiming to improve reliability in disease diagnostics is coming in 2 months.

Downloads last month
89
GGUF
Model size
13B params
Architecture
llama

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

Inference API (serverless) has been turned off for this model.

Quantized from