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. GGUF offers numerous advantages over GGML, such as better tokenisation, and support for special tokens. It is also supports metadata, and is designed to be extensible.
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.
[INST] <<SYS>>
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
<</SYS>>
{prompt}[/INST]
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.
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/Trurl-2-13B-GGUF and below it, a specific filename to download, such as: trurl-2-13b.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>=0.17.1
Then you can download any individual model file to the current directory, at high speed, with a command like this:
./main -ngl 32 -m trurl-2-13b.q4_K_M.gguf --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\n<</SYS>>\n{prompt}[/INST]"
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
How to load this model from Python using ctransformers
First install the package
bash
1# Base ctransformers with no GPU acceleration2pip install ctransformers>=0.2.24
3# Or with CUDA GPU acceleration4pip install ctransformers[cuda]>=0.2.24
5# Or with ROCm GPU acceleration6CT_HIPBLAS=1 pip install ctransformers>=0.2.24 --no-binary ctransformers
7# Or with Metal GPU acceleration for macOS systems8CT_METAL=1 pip install ctransformers>=0.2.24 --no-binary ctransformers
Simple example code to load one of these GGUF models
python
1from ctransformers import AutoModelForCausalLM
23# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.4llm = AutoModelForCausalLM.from_pretrained("TheBloke/Trurl-2-13B-GGUF", model_file="trurl-2-13b.q4_K_M.gguf", model_type="llama", gpu_layers=50)56print(llm("AI is going to"))
How to use with LangChain
Here's guides on using llama-cpp-python or ctransformers with LangChain:
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 special mentions: Alicia Loh, Stephen Murray, K, Ajan Kanaga, RoA, Magnesian, Deo Leter, Olakabola, Eugene Pentland, zynix, Deep Realms, Raymond Fosdick, Elijah Stavena, Iucharbius, Erik Bjäreholt, Luis Javier Navarrete Lozano, Nicholas, theTransient, John Detwiler, alfie_i, knownsqashed, Mano Prime, Willem Michiel, Enrico Ros, LangChain4j, OG, Michael Dempsey, Pierre Kircher, Pedro Madruga, James Bentley, Thomas Belote, Luke @flexchar, Leonard Tan, Johann-Peter Hartmann, Illia Dulskyi, Fen Risland, Chadd, S_X, Jeff Scroggin, Ken Nordquist, Sean Connelly, Artur Olbinski, Swaroop Kallakuri, Jack West, Ai Maven, David Ziegler, Russ Johnson, transmissions 11, John Villwock, Alps Aficionado, Clay Pascal, Viktor Bowallius, Subspace Studios, Rainer Wilmers, Trenton Dambrowitz, vamX, Michael Levine, 준교 김, Brandon Frisco, Kalila, Trailburnt, Randy H, Talal Aujan, Nathan Dryer, Vadim, 阿明, ReadyPlayerEmma, Tiffany J. Kim, George Stoitzev, Spencer Kim, Jerry Meng, Gabriel Tamborski, Cory Kujawski, Jeffrey Morgan, Spiking Neurons AB, Edmond Seymore, Alexandros Triantafyllidis, Lone Striker, Cap'n Zoog, Nikolai Manek, danny, ya boyyy, Derek Yates, usrbinkat, Mandus, TL, Nathan LeClaire, subjectnull, Imad Khwaja, webtim, Raven Klaugh, Asp the Wyvern, Gabriel Puliatti, Caitlyn Gatomon, Joseph William Delisle, Jonathan Leane, Luke Pendergrass, SuperWojo, Sebastain Graf, Will Dee, Fred von Graf, Andrey, Dan Guido, Daniel P. Andersen, Nitin Borwankar, Elle, Vitor Caleffi, biorpg, jjj, NimbleBox.ai, Pieter, Matthew Berman, terasurfer, Michael Davis, Alex, Stanislav Ovsiannikov
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
Original model card: Voicelab's Trurl 2 13B
logo voicelab nlp
Trurl 2 -- Polish Llama 2
The new OPEN TRURL is a finetuned Llama 2, trained on over 1.7b tokens (970k conversational Polish and English samples) with a large context of 4096 tokens.
TRURL was trained on a large number of Polish data.
TRURL 2 is a collection of fine-tuned generative text models with 7 billion and 13 billion parameters.
This is the repository for the 13B fine-tuned model, optimized for dialogue use cases.
Overview
TRURL developers Voicelab.AI
Variations Trurl 2 comes in 7B and 13B versions.
Input Models input text only.
Output Models generate text only.
Model Architecture Trurl is an auto-regressive language model that uses an optimized transformer architecture.
Training Data
Params
Content Length
Num. Samples
Num. Tokens
start LR
Trurl 2
A new mix of private and publicly available online data without MMLU
7B
4k
855k
1.19b
2.0 x 10-5
Trurl 2
A new mix of private and publicly available online data with MMLU
13B
4k
970k
1.7b
2.0 x 10-5
Trurl 2 Academic
A new mix of private and publicly available online data without MMLU
13B
4k
855k
1.19b
2.0 x 10-5
Training data
The training data includes Q&A pairs from various sources including Alpaca comparison data with GPT, Falcon comparison data, Dolly 15k, Oasst1, Phu saferlfhf, ShareGPT version 2023.05.08v0 filtered and cleaned, Voicelab private datasets for JSON data extraction, modification, and analysis, CURLICAT dataset containing journal entries, dataset from Polish wiki with Q&A pairs grouped into conversations, MMLU data in textual format, Voicelab private dataset with sales conversations, arguments and objections, paraphrases, contact reason detection, and corrected dialogues.
Intended Use
Trurl 2 is intended for commercial and research use in Polish and English. Tuned models are intended for assistant-like chat, but also adapted for a variety of natural language generation tasks.
Evaluation Results
Model
Size
hellaswag
arc_challenge
MMLU
Llama-2-chat
7B
78.55%
52.9%
48.32%
Llama-2-chat
13B
81.94%
59.04%
54.64%
Trurl 2.0 (with MMLU)
13B
80.09%
59.30%
78.35%
Trurl 2.0 (no MMLU)
13B
TO-DO
TO-DO
TO-DO
Trurl 2.0 (no MMLU)
7b
75.29%
53.41%
50.0%
trurl graphic
Examples
Polish language
HUMAN: What’s the point of training you on Polish?
TRURL: The point of training me on Polish, or any other language, is to enable me to understand and respond to queries and questions in that language. This allows people who speak that language to use me as a helpful tool for finding information, answering questions, and completing tasks. By training me on a wide variety of texts in Polish, I can learn about the culture, history, and language of Poland, and be able to provide more accurate and relevant responses to users. Additionally, training me on multiple languages allows me to be useful to a wider range of people, regardless of their native language.
Stanisław Lem
HUMAN: who is Trurl from Stanisław Lem book?
TRURL: Trurl is a character in a series of short stories by Polish science fiction writer Stanisław Lem. The stories feature Trurl and another character, Klapaucius, who are both inventors and explorers of the universe. They use their advanced technology and knowledge to solve various problems and encounter strange and fantastical creatures along the way. The stories often explore themes of science, technology, and the nature of intelligence and consciousness.
The description of original Trurl generated by TRURL 2.0. Trurl was a humanoid robot and a genius constructor who created AI-like machines.
HUMAN: why are you named Trurl?
TRURL: I was named after the character Trurl from the short stories by Stanisław Lem. My creators, the VoiceLab AI team, were inspired by the character’s intelligence and creativity, and they wanted to create an AI assistant that embodied those qualities. They also wanted to give me a unique and memorable name, and Trurl seemed like the perfect fit.
Example use
LLM
Simply pass a prompt to a model and decode an output. Model will continue writing text based on sample you provided.
import torch
from transformers import LlamaForCausalLM, LlamaTokenizer
tokenizer = LlamaTokenizer.from_pretrained("Voicelab/trurl-2-13b")
model = LlamaForCausalLM.from_pretrained("Voicelab/trurl-2-13b")
prompt = "Yesterday, when I was"
tokenized_prompt = tokenizer(prompt, return_tensors="pt")
model.eval()
with torch.no_grad():
print(tokenizer.decode(
model.generate(**tokenized_prompt, max_new_tokens=200)[0],
skip_special_tokens=True))
Chat
When using TRURL in a chat mode you should remember to use Llama 2 conversation template like in the example below.
import torch
from transformers import LlamaForCausalLM, LlamaTokenizer
tokenizer = LlamaTokenizer.from_pretrained("Voicelab/trurl-2-13b")
model = LlamaForCausalLM.from_pretrained("Voicelab/trurl-2-13b")
prompt = """
<s>[INST] <<SYS>> You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe.
Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content.
Please ensure that your responses are socially unbiased and positive in nature.\n\n
If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct.
If you don't know the answer to a question, please don't share false information. <</SYS>>
What was the reason for calling in the conversation below? \n\n
AGENT: Hello, Bank of Albion, this is Mata Hari. How can I help you?
CLIENT: Hi. I've been locked out from my Internet account. I need your help.
AGENT: (yy) Yes, of course, I'll do my best to help you. But I need to find out why the locking-out happened. (yy) In order to ascertain that, I'll ask you a couple of questions to confirm your identity. I'm going to need your full name.
CLIENT: Lizz Truss.
AGENT: Thank you. Now I need your personal identification number.
CLIENT: Fourteen, two hundred thirty-one, thirty-eight, twenty-nine, sixty-five.
AGENT: Thank you. Now I need your client ID number. The client ID number is the eight digits we assigned to you at the very beginning, on conclusion of the contract.
CLIENT: OK. Give me a moment. I have to find it.
AGENT: (mhm) You'll find… You'll find it in the contract.
CLIENT: Yes, yes. I can see it. Sixty-five, twenty-nine, thirty-eight, thirty-one.
AGENT: Thank you. One final security question. Do you have any deposits in our bank?
CLIENT: No, no. I don't have any deposits in this bank.
AGENT: Thank you. Your identity has been (yy) confirmed. (yy) I can see that the account has been blocked, indeed, and you won't be able to log in via the Internet (yy) because (yy) the identity document which is listed for reference has expired. (yy) From what I can see, your identity document expired some time ago. Have you been issued a new one?
CLIENT: Well, no. I think my ID is still valid, you know. I didn't even know.
AGENT: Well, no... Your ID expired at the end of March. Well, almost at the end. Your old ID had been valid until 26 March. (yy) For that reason, your accout has been blocked, because you haven't notified us about the ID change for a few months. We are not interested if the ID document has been officialy reissued. (...) On our end, what matters is whether the document listed for our reference is valid (yy) so without a valid document I can't unlock your accout.
CLIENT: But I have to carry out an operation right now, so this is sort of problematic.
AGENT: I understand. But (yy) you are obligated, as an account holder, to notify the bank about any changes pending (yy), regrding, for example, your home address or phone number. Now, one of such safeguards protecting your… (yy) money, your sensitive data, is precisely about having a valid identification document. Since this is missing in your case, the account has been blocked. Now, I don't think this would have caught you off guard, because we always remind our customers that their ID is about to expire. When the ID is nearing expiration, we display relevant messages at least sixty days in advance. They appear once you've logged in, at the very top of the screen, there is a notification that (yy) the ID is about to expire (yy), so, well... The bank did notify you about this issue. Now, how you chose to act on this information was your choice, right? In any case, at this point, in order to unlock your accout, our protocols require that you produce a new identification document at one of our branches. You shall provide information concerning the new document number, new valid-thru date, and only then will you be able to use your account again. I can schedule an appointment with a consultant at our branch for you. What locality would you prefer?
CLIENT: Well, I'm not sure if I should share such information with you.
AGENT: And may I ask why exactly you are unsure? After all, you're calling a bank that runs your account, right?
CLIENT: Right, you know what, I need to go now. Good bye.
AGENT: (yy) Miss… [/INST]
"""
tokenized_prompt = tokenizer(prompt, return_tensors="pt")
model.eval()
with torch.no_grad():
print(tokenizer.decode(
model.generate(**tokenized_prompt, max_new_tokens=200)[0],
skip_special_tokens=True))
To get the expected features and performance for the chat versions, a specific Llama 2 formatting needs to be followed, including the INST and <<SYS>> tags, BOS and EOS tokens, and the whitespaces and breaklines in between (we recommend calling strip() on inputs to avoid double-spaces). See reference code in github for details: chat_completion.
<s>[INST] <<SYS>> system prompt <</SYS>>
human prompt [/INST]
gpt response </s>
<s>[INST] human prompt [/INST]
gpt response </s>
Ethical Considerations and Limitations
Trurl 2, same as a Llama 2, is a new technology that carries risks with use. Testing conducted to date has been in Polish and English, and has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, Trurl 2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Trurl 2, developers should perform safety testing and tuning tailored to their specific applications of the model.
