Original concept and code for increasing context length: kaiokendev
Updated Llama modelling code that includes this automatically via trust_remote_code: emozilla.

Please read carefully below to see how to use it.

GGML versions are not yet provided, as there is not yet support for SuperHOT in llama.cpp. This is being investigated and will hopefully come soon.

Repositories available

How to easily download and use this model in text-generation-webui with ExLlama

Please make sure you're using the latest version of text-generation-webui

Click the Model tab.
Under Download custom model or LoRA, enter TheBloke/Minotaur-13B-fixed-SuperHOT-8K-GPTQ.
Click Download.
The model will start downloading. Once it's finished it will say "Done"
Untick Autoload the model
In the top left, click the refresh icon next to Model.
In the Model dropdown, choose the model you just downloaded: Minotaur-13B-fixed-SuperHOT-8K-GPTQ
To use the increased context, set the Loader to ExLlama, set max_seq_len to 8192 or 4096, and set compress_pos_emb to 4 for 8192 context, or to 2 for 4096 context.
Now click Save Settings followed by Reload
The model will automatically load, and is now ready for use!
Once you're ready, click the Text Generation tab and enter a prompt to get started!

How to use this GPTQ model from Python code with AutoGPTQ

First make sure you have AutoGPTQ and Einops installed:

pip3 install einops auto-gptq

Then run the following code. Note that in order to get this to work, config.json has been hardcoded to a sequence length of 8192.

If you want to try 4096 instead to reduce VRAM usage, please manually edit config.json to set max_position_embeddings to the value you want.

python

from transformers import AutoTokenizer, pipeline, logging
from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
import argparse

model_name_or_path = "TheBloke/Minotaur-13B-fixed-SuperHOT-8K-GPTQ"
model_basename = "minotaur-13b-fixed-superhot-8k-GPTQ-4bit-128g.no-act.order"

use_triton = False

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)

model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
        model_basename=model_basename,
        use_safetensors=True,
        trust_remote_code=True,
        device_map='auto',
        use_triton=use_triton,
        quantize_config=None)

model.seqlen = 8192

# Note: check the prompt template is correct for this model.
prompt = "Tell me about AI"
prompt_template=f'''USER: {prompt}
ASSISTANT:'''

print("\n\n*** Generate:")

input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
output = model.generate(inputs=input_ids, temperature=0.7, max_new_tokens=512)
print(tokenizer.decode(output[0]))

# Inference can also be done using transformers' pipeline

# Prevent printing spurious transformers error when using pipeline with AutoGPTQ
logging.set_verbosity(logging.CRITICAL)

print("*** Pipeline:")
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.95,
    repetition_penalty=1.15
)

print(pipe(prompt_template)[0]['generated_text'])

Using other UIs: monkey patch

Provided in the repo is llama_rope_scaled_monkey_patch.py, written by @kaiokendev.

It can be theoretically be added to any Python UI or custom code to enable the same result as trust_remote_code=True. I have not tested this, and it should be superseded by using trust_remote_code=True, but I include it for completeness and for interest.

Provided files

minotaur-13b-fixed-superhot-8k-GPTQ-4bit-128g.no-act.order.safetensors

This will work with AutoGPTQ, ExLlama, and CUDA versions of GPTQ-for-LLaMa. There are reports of issues with Triton mode of recent GPTQ-for-LLaMa. If you have issues, please use AutoGPTQ instead.

It was created with group_size 128 to increase inference accuracy, but without --act-order (desc_act) to increase compatibility and improve inference speed.

minotaur-13b-fixed-superhot-8k-GPTQ-4bit-128g.no-act.order.safetensors
- Works for use with ExLlama with increased context (4096 or 8192)
- Works with AutoGPTQ in Python code, including with increased context, if trust_remote_code=True is set.
- Should work with GPTQ-for-LLaMa in CUDA mode, but unknown if increased context works - TBC. May have issues with GPTQ-for-LLaMa Triton mode.
- Works with text-generation-webui, including one-click-installers.
- Parameters: Groupsize = 128. Act Order / desc_act = False.

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!

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.

Patreon: https://patreon.com/TheBlokeAI
Ko-Fi: https://ko-fi.com/TheBlokeAI

Special thanks to: Aemon Algiz.

Patreon special mentions: Sam, theTransient, Jonathan Leane, Steven Wood, webtim, Johann-Peter Hartmann, Geoffrey Montalvo, Gabriel Tamborski, Willem Michiel, John Villwock, Derek Yates, Mesiah Bishop, Eugene Pentland, Pieter, Chadd, Stephen Murray, Daniel P. Andersen, terasurfer, Brandon Frisco, Thomas Belote, Sid, Nathan LeClaire, Magnesian, Alps Aficionado, Stanislav Ovsiannikov, Alex, Joseph William Delisle, Nikolai Manek, Michael Davis, Junyu Yang, K, J, Spencer Kim, Stefan Sabev, Olusegun Samson, transmissions 11, Michael Levine, Cory Kujawski, Rainer Wilmers, zynix, Kalila, Luke @flexchar, Ajan Kanaga, Mandus, vamX, Ai Maven, Mano Prime, Matthew Berman, subjectnull, Vitor Caleffi, Clay Pascal, biorpg, alfie_i, 阿明, Jeffrey Morgan, ya boyyy, Raymond Fosdick, knownsqashed, Olakabola, Leonard Tan, ReadyPlayerEmma, Enrico Ros, Dave, Talal Aujan, Illia Dulskyi, Sean Connelly, senxiiz, Artur Olbinski, Elle, Raven Klaugh, Fen Risland, Deep Realms, Imad Khwaja, Fred von Graf, Will Dee, usrbinkat, SuperWojo, Alexandros Triantafyllidis, Swaroop Kallakuri, Dan Guido, John Detwiler, Pedro Madruga, Iucharbius, Viktor Bowallius, Asp the Wyvern, Edmond Seymore, Trenton Dambrowitz, Space Cruiser, Spiking Neurons AB, Pyrater, LangChain4j, Tony Hughes, Kacper Wikieł, Rishabh Srivastava, David Ziegler, Luke Pendergrass, Andrey, Gabriel Puliatti, Lone Striker, Sebastain Graf, Pierre Kircher, Randy H, NimbleBox.ai, Vadim, danny, Deo Leter

Thank you to all my generous patrons and donaters!

And thank you again to a16z for their generous grant.

Original model card: Kaio Ken's SuperHOT 8K

SuperHOT Prototype 2 w/ 8K Context

This is a second prototype of SuperHOT, this time 30B with 8K context and no RLHF, using the same technique described in the github blog. Tests have shown that the model does indeed leverage the extended context at 8K.

You will need to use either the monkeypatch or, if you are already using the monkeypatch, change the scaling factor to 0.25 and the maximum sequence length to 8192

Looking for Merged & Quantized Models?

30B 4-bit CUDA: tmpupload/superhot-30b-8k-4bit-safetensors
30B 4-bit CUDA 128g: tmpupload/superhot-30b-8k-4bit-128g-safetensors

Training Details

I trained the LoRA with the following configuration:

1200 samples (~400 samples over 2048 sequence length)
learning rate of 3e-4
3 epochs
The exported modules are:
- q_proj
- k_proj
- v_proj
- o_proj
- no bias
Rank = 4
Alpha = 8
no dropout
weight decay of 0.1
AdamW beta1 of 0.9 and beta2 0.99, epsilon of 1e-5
Trained on 4-bit base model

Original model card: OpenAccess AI Collective's Minotaur 13B Fixed

💵 Donate to OpenAccess AI Collective to help us keep building great tools and models!

Due to a bug, the initial release of Minotaur 13B dropped a few datasets during training. We have corrected the issue and this is the retrained model

The affected datasets include:

prose generation
classification
coding

Minotaur 13B (FIXED)

Minotaur 13B is an instruct fine-tuned model on top of LlaMA-13B. Minotaur 13B is fine-tuned on only completely open datasets making this model reproducible by anyone.

Questions, comments, feedback, looking to donate, or want to help? Reach out on our Discord or email wing@openaccessaicollective.org

Prompts

Chat only style prompts using USER:,ASSISTANT:.

Training Datasets

Minotaur 13B model is fine-tuned on the following openly available datasets:

WizardLM
subset of QingyiSi/Alpaca-CoT for roleplay and CoT
GPTeacher-General-Instruct
metaeval/ScienceQA_text_only - instruct for concise responses
openai/summarize_from_feedback - instruct augmented tl;dr summarization
camel-ai/math
camel-ai/physics
camel-ai/chemistry
camel-ai/biology
winglian/evals - instruct augmented datasets
- custom sysnthetic datasets around misconceptions, in-context qa, jokes, N-tasks problems, and context-insensitivity
- ARC-Easy & ARC-Challenge - instruct augmented for detailed responses, derived from the train split
- hellaswag - 30K+ rows of instruct augmented for detailed explanations w 30K+ rows, derived from the train split
- riddle_sense - instruct augmented, derived from the train split
- gsm8k - instruct augmented, derived from the train split
- prose generation

Shoutouts

Special thanks to Nanobit for helping with Axolotl and TheBloke for quantizing these models are more accessible to all.

Demo

HF Demo in Spaces available in the Community ChatBot Arena under the OAAIC Chatbots tab.

Release Notes

https://wandb.ai/wing-lian/minotaur-13b/runs/5ystr7w6/workspace

Build

Minotaur was built with Axolotl on 6XA100 80GB

1 epochs taking approximately 7.5 hours

Bias, Risks, and Limitations

Minotaur has not been aligned to human preferences with techniques like RLHF or deployed with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so). Minotaur was fine-tuned from the base model LLaMA-13B, please refer to its model card's Limitations Section for relevant information. (included below)

Benchmarks

hf-causal-experimental (pretrained=openaccess-ai-collective/minotaur-13b-fixed), limit: None, provide_description: False, num_fewshot: 0, batch_size: None

Task	Version	Metric	Value		Stderr
arc_challenge	0	acc	0.4778	_	0.0146
		acc_norm	0.4932	_	0.0146
arc_easy	0	acc	0.7786	_	0.0085
		acc_norm	0.7487	_	0.0089
boolq	1	acc	0.8141	_	0.0068
hellaswag	0	acc	0.6072	_	0.0049
		acc_norm	0.7929	_	0.0040
openbookqa	0	acc	0.3420	_	0.0212
		acc_norm	0.4580	_	0.0223
piqa	0	acc	0.8025	_	0.0093
		acc_norm	0.8107	_	0.0091
winogrande	0	acc	0.7285	_	0.0125

Examples - results may vary based on temperature (0.7 for this) and other settings

The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: Solve for x: 5x = 3x + 10
ASSISTANT:A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
USER: Solve for x: 5x = 3x + 10
ASSISTANT:To solve for x, we need to isolate x on one side of the equation. We can do this by following these steps:

1. Subtract 3x from both sides of the equation:
5x - 3x = 3x + 10 - 3x
2x = 10

2. Divide both sides of the equation by 2:
2x/2 = 10/2
x = 5

The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: You have 3 apples. You eat 2 pears. How many apples do you have left?
ASSISTANT:You still have 3 apples. Eating pears does not affect the number of apples you have.

The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: List three words that start with "ex". What is 33 + 77?
ASSISTANT:Three words that start with "ex" are "express", "exact", and "excite". 33 + 77 = 110.

The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: Write a haiku about dinosaurs.
ASSISTANT:Titanic creatures roam,
Thunderous footsteps echo,
Extinct but not forgotten.

meh, 6/7/7 rather than 5/7/5

A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
USER: Create an original short story in the style of J.K Rowling about the Weesley family going on a vacation in Australia whith Harry Potter and Hermione Grainger tagging along.
ASSISTANT:Once upon a time, there was a magical family named the Weesleys. They lived in a cozy little cottage at the edge of Hogsmeade village, where they ran a charming bed and breakfast called "The Flying Ford Anglia." A
s any other magical family, they enjoyed adventures that brought them close to nature, so they decided to go on a trip to Australia. They invited their dear friends Harry Potter and Hermione Grainger to join them on this ex
citing journey.

The Weesleys had been planning for months and had everything ready for their trip. Their children, Freddie and Roxanne were thrilled to visit the land down under, and their parents, George and Angelina, could not wait to ex
plore the beautiful country with their friends. They would be leaving soon, but before departing, there was one small problem - the luggage.

"We can't possibly take all these trunks!" exclaimed Freddie as he stared at the mountain of bags stacked up in front of him.

His father looked at him seriously. "You see, son," he began, "our luggage is crucial because we have to pack all our magical essentials like wand holsters, extra wand chargers, spell books, potions ingredients, and much mo
re."

Roxanne interrupted, "But daddy, we don't even use magic in everyday life! And what if it rains? We'll look weird carrying umbrellas around!"

Her mother joined in. "My dear, you know how unpredictable the weather can be. We must always be prepared for unexpected circumstances. Don't forget that even Muggles carry umbrellas."

After hours of debating, they finally managed to narrow down their luggage list and fit all the necessary supplies into several large trunks. The day had come; they were ready to leave for their grand adventure!

As the Weesleys boarded the Portkey that would transport them across the world, their wands began to glow softly, indicating that they had enough energy to make the journey. The Portkey dropped them off in Sydney, right in
front of the magnificent Opera House.

They spent the first few days exploring the city, marveling at the iconic architecture and tasting local cuisine. Then, as planned, they headed north to visit the Great Barrier Reef, one of the most famous natural wonders o
f the world.

Harry and Hermione joined them during this leg of the trip, which made it even more enjoyable. Harry regaled them with tales of his own travels while Hermione shared her extensive knowledge of plants, animals, and the envir
onment.

Soon, they arrived at a quaint town nestled among vibrant green hills and surrounded by vast cattle farms. It was here that they would learn about Aboriginal culture and see some truly unique sights.

One morning, after enjoying a hearty breakfast, they set out to explore the local area. They visited a nearby art gallery that showcased amazing Indigenous works of art, including traditional paintings, sculptures, and text
iles. Afterward, they attended a didgeridoo concert given by a talented young musician who captivated everyone with his soulful tunes.

The following day, they embarked on a bushwalk through the rainforest trails. The air was cool and fresh, and the towering trees seemed to reach for the sky. Hermione took great pleasure in identifying various species of pl
ants and explaining their medicinal properties.

As they walked deeper into the forest, they heard faint chattering noises coming from high above. To their surprise, they spotted a group of small, furry creatures perched on a tree branch. They turned out to be native Aust
ralian possums, which delighted the children, especially Freddie, who could communicate with them using sign language.

Later that afternoon, they visited Uluru, also known as Ayers Rock, a massive sandstone monolith located in the heart of the Outback. The sight of its magnificent red hues against the blue sky left them all speechless. They
 spent the rest of the day exploring the rock formation, taking in the breathtaking views and learning about its significance to the Anangu people.

Their last destination was Queensland, home to the world-famous Great Barrier Reef. They spent two days snorkeling, diving, and swimming among the colorful coral reefs, watching schools of fish, sea turtles, and dolphins sw
im past them.

As the sun began to set over the crystal-clear waters, Harry spoke up. "I think it's time we return to reality, don't you agree?" he said, looking at his wife. "This trip has been fantastic, but we can't stay forever."

Everyone nodded in agreement as they prepared themselves for their final journey back to Sydney. When they got there, they gathered on the balcony of their hotel room, reminiscing about the incredible memories they had crea
ted together.

"Thank you for inviting us along on your adventure, Weesleys," said Harry. "It was an honor and a privilege to share this experience with such kind and generous friends. I hope we can do it again sometime."

With tears in their eyes and heavy hearts, they bid each other farewell, promising to keep in touch and meet again soon. And so, their epic journey came to an end. But the memories would remain with them forever, reminding
them of the power of friendship, the beauty of nature, and the importance of discovering new worlds.