guanaco-33B-GGML / README.md
TheBloke's picture
Update README.md
869e970
|
raw
history blame
18.3 kB
metadata
inference: false
license: other
TheBlokeAI

Tim Dettmers' Guanaco 33B GGML

These files are GGML format model files for Tim Dettmers' Guanaco 33B.

GGML files are for CPU + GPU inference using llama.cpp and libraries and UIs which support this format, such as:

Repositories available

Prompt template

### Human: prompt
### Assistant:

Compatibility

Original llama.cpp quant methods: q4_0, q4_1, q5_0, q5_1, q8_0

I have quantized these 'original' quantisation methods using an older version of llama.cpp so that they remain compatible with llama.cpp as of May 19th, commit 2d5db48.

They should be compatible with all current UIs and libraries that use llama.cpp, such as those listed at the top of this README.

New k-quant methods: q2_K, q3_K_S, q3_K_M, q3_K_L, q4_K_S, q4_K_M, q5_K_S, q6_K

These new quantisation methods are only compatible with llama.cpp as of June 6th, commit 2d43387.

They will NOT be compatible with koboldcpp, text-generation-ui, and other UIs and libraries yet. Support is expected to come over the next few days.

Explanation of the new k-quant methods

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
  • GGML_TYPE_Q8_K - "type-0" 8-bit quantization. Only used for quantizing intermediate results. The difference to the existing Q8_0 is that the block size is 256. All 2-6 bit dot products are implemented for this quantization type.

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
guanaco-33B.ggmlv3.q2_K.bin q2_K 2 13.60 GB 16.10 GB New k-quant method. Uses GGML_TYPE_Q4_K for the attention.vw and feed_forward.w2 tensors, GGML_TYPE_Q2_K for the other tensors.
guanaco-33B.ggmlv3.q3_K_L.bin q3_K_L 3 17.20 GB 19.70 GB New k-quant method. Uses GGML_TYPE_Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K
guanaco-33B.ggmlv3.q3_K_M.bin q3_K_M 3 15.64 GB 18.14 GB New k-quant method. Uses GGML_TYPE_Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K
guanaco-33B.ggmlv3.q3_K_S.bin q3_K_S 3 13.98 GB 16.48 GB New k-quant method. Uses GGML_TYPE_Q3_K for all tensors
guanaco-33B.ggmlv3.q4_0.bin q4_0 4 18.30 GB 20.80 GB Original llama.cpp quant method, 4-bit.
guanaco-33B.ggmlv3.q4_1.bin q4_1 4 20.33 GB 22.83 GB Original llama.cpp quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.
guanaco-33B.ggmlv3.q4_K_M.bin q4_K_M 4 19.57 GB 22.07 GB New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q4_K
guanaco-33B.ggmlv3.q4_K_S.bin q4_K_S 4 18.30 GB 20.80 GB New k-quant method. Uses GGML_TYPE_Q4_K for all tensors
guanaco-33B.ggmlv3.q5_0.bin q5_0 5 22.37 GB 24.87 GB Original llama.cpp quant method, 5-bit. Higher accuracy, higher resource usage and slower inference.
guanaco-33B.ggmlv3.q5_1.bin q5_1 5 24.40 GB 26.90 GB Original llama.cpp quant method, 5-bit. Even higher accuracy, resource usage and slower inference.
guanaco-33B.ggmlv3.q5_K_M.bin q5_K_M 5 23.02 GB 25.52 GB New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q5_K
guanaco-33B.ggmlv3.q5_K_S.bin q5_K_S 5 22.37 GB 24.87 GB New k-quant method. Uses GGML_TYPE_Q5_K for all tensors
guanaco-33B.ggmlv3.q6_K.bin q6_K 6 26.69 GB 29.19 GB New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors
guanaco-33B.ggmlv3.q8_0.bin q8_0 8 34.56 GB 37.06 GB Original llama.cpp quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users.

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 run in llama.cpp

I use the following command line; adjust for your tastes and needs:

./main -t 10 -ngl 32 -m guanaco-33B.ggmlv3.q5_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction: Write a story about llamas\n### Response:"

Change -t 10 to the number of physical CPU cores you have. For example if your system has 8 cores/16 threads, use -t 8.

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

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

How to run in text-generation-webui

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

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.

Special thanks to: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.

Patreon special mentions: Ajan Kanaga, Kalila, Derek Yates, Sean Connelly, Luke, Nathan LeClaire, Trenton Dambrowitz, Mano Prime, David Flickinger, vamX, Nikolai Manek, senxiiz, Khalefa Al-Ahmad, Illia Dulskyi, trip7s trip, Jonathan Leane, Talal Aujan, Artur Olbinski, Cory Kujawski, Joseph William Delisle, Pyrater, Oscar Rangel, Lone Striker, Luke Pendergrass, Eugene Pentland, Johann-Peter Hartmann.

Thank you to all my generous patrons and donaters!

Original model card: Tim Dettmers' Guanaco 33B

Guanaco Models Based on LLaMA

| Paper | Code | Demo |

The Guanaco models are open-source finetuned chatbots obtained through 4-bit QLoRA tuning of LLaMA base models on the OASST1 dataset. They are available in 7B, 13B, 33B, and 65B parameter sizes.

⚠️Guanaco is a model purely intended for research purposes and could produce problematic outputs.

Why use Guanaco?

  • Competitive with commercial chatbot systems on the Vicuna and OpenAssistant benchmarks (ChatGPT and BARD) according to human and GPT-4 raters. We note that the relative performance on tasks not covered in these benchmarks could be very different. In addition, commercial systems evolve over time (we used outputs from the March 2023 version of the models).
  • Available open-source for research purposes. Guanaco models allow cheap and local experimentation with high-quality chatbot systems.
  • Replicable and efficient training procedure that can be extended to new use cases. Guanaco training scripts are available in the QLoRA repo.
  • Rigorous comparison to 16-bit methods (both 16-bit full-finetuning and LoRA) in our paper demonstrates the effectiveness of 4-bit QLoRA finetuning.
  • Lightweight checkpoints which only contain adapter weights.

License and Intended Use

Guanaco adapter weights are available under Apache 2 license. Note the use of the Guanaco adapter weights, requires access to the LLaMA model weighs. Guanaco is based on LLaMA and therefore should be used according to the LLaMA license.

Usage

Here is an example of how you would load Guanaco 7B in 4-bits:

import torch
from peft import PeftModel    
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_name = "huggyllama/llama-7b"
adapters_name = 'timdettmers/guanaco-7b'

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    load_in_4bit=True,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    max_memory= {i: '24000MB' for i in range(torch.cuda.device_count())},
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_compute_dtype=torch.bfloat16,
        bnb_4bit_use_double_quant=True,
        bnb_4bit_quant_type='nf4'
    ),
)
model = PeftModel.from_pretrained(model, adapters_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Inference can then be performed as usual with HF models as follows:

prompt = "Introduce yourself"
formatted_prompt = (
    f"A chat between a curious human and an artificial intelligence assistant."
    f"The assistant gives helpful, detailed, and polite answers to the user's questions.\n"
    f"### Human: {prompt} ### Assistant:"
)
inputs = tokenizer(formatted_prompt, return_tensors="pt").to("cuda:0")
outputs = model.generate(inputs=inputs.input_ids, max_new_tokens=20)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Expected output similar to the following:

A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
### Human: Introduce yourself ### Assistant: I am an artificial intelligence assistant. I am here to help you with any questions you may have.

Current Inference Limitations

Currently, 4-bit inference is slow. We recommend loading in 16 bits if inference speed is a concern. We are actively working on releasing efficient 4-bit inference kernels.

Below is how you would load the model in 16 bits:

model_name = "huggyllama/llama-7b"
adapters_name = 'timdettmers/guanaco-7b'
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    max_memory= {i: '24000MB' for i in range(torch.cuda.device_count())},
)
model = PeftModel.from_pretrained(model, adapters_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Model Card

Architecture: The Guanaco models are LoRA adapters to be used on top of LLaMA models. They are added to all layers. For all model sizes, we use $r=64$.

Base Model: Guanaco uses LLaMA as base model with sizes 7B, 13B, 33B, 65B. LLaMA is a causal language model pretrained on a large corpus of text. See LLaMA paper for more details. Note that Guanaco can inherit biases and limitations of the base model.

Finetuning Data: Guanaco is finetuned on OASST1. The exact dataset is available at timdettmers/openassistant-guanaco.

Languages: The OASST1 dataset is multilingual (see the paper for details) and as such Guanaco responds to user queries in different languages. We note, however, that OASST1 is heavy in high-resource languages. In addition, human evaluation of Guanaco was only performed in English and based on qualitative analysis we observed degradation in performance in other languages.

Next, we describe Training and Evaluation details.

Training

Guanaco models are the result of 4-bit QLoRA supervised finetuning on the OASST1 dataset.

All models use NormalFloat4 datatype for the base model and LoRA adapters on all linear layers with BFloat16 as computation datatype. We set LoRA $r=64$, $\alpha=16$. We also use Adam beta2 of 0.999, max grad norm of 0.3 and LoRA dropout of 0.1 for models up to 13B and 0.05 for 33B and 65B models. For the finetuning process, we use constant learning rate schedule and paged AdamW optimizer.

Training hyperparameters

Size Dataset Batch Size Learning Rate Max Steps Sequence length
7B OASST1 16 2e-4 1875 512
13B OASST1 16 2e-4 1875 512
33B OASST1 16 1e-4 1875 512
65B OASST1 16 1e-4 1875 512

Evaluation

We test generative language capabilities through both automated and human evaluations. This second set of evaluations relies on queries curated by humans and aims at measuring the quality of model responses. We use the Vicuna and OpenAssistant datasets with 80 and 953 prompts respectively.

In both human and automated evaluations, for each prompt, raters compare all pairs of responses across the models considered. For human raters we randomize the order of the systems, for GPT-4 we evaluate with both orders.

Benchmark Vicuna Vicuna OpenAssistant -
Prompts 80 80 953
Judge Human GPT-4 GPT-4
Model Elo Rank Elo Rank Elo Rank Median Rank
GPT-4 1176 1 1348 1 1294 1 1
Guanaco-65B 1023 2 1022 2 1008 3 2
Guanaco-33B 1009 4 992 3 1002 4 4
ChatGPT-3.5 Turbo 916 7 966 5 1015 2 5
Vicuna-13B 984 5 974 4 936 5 5
Guanaco-13B 975 6 913 6 885 6 6
Guanaco-7B 1010 3 879 8 860 7 7
Bard 909 8 902 7 - - 8

We also use the MMLU benchmark to measure performance on a range of language understanding tasks. This is a multiple-choice benchmark covering 57 tasks including elementary mathematics, US history, computer science, law, and more. We report 5-shot test accuracy.

Dataset 7B 13B 33B 65B
LLaMA no tuning 35.1 46.9 57.8 63.4
Self-Instruct 36.4 33.3 53.0 56.7
Longform 32.1 43.2 56.6 59.7
Chip2 34.5 41.6 53.6 59.8
HH-RLHF 34.9 44.6 55.8 60.1
Unnatural Instruct 41.9 48.1 57.3 61.3
OASST1 (Guanaco) 36.6 46.4 57.0 62.2
Alpaca 38.8 47.8 57.3 62.5
FLAN v2 44.5 51.4 59.2 63.9

Risks and Biases

The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. The model was trained on various public datasets; it is possible that this model could generate lewd, biased, or otherwise offensive outputs.

However, we note that finetuning on OASST1 seems to reduce biases as measured on the CrowS dataset. We report here the performance of Guanaco-65B compared to other baseline models on the CrowS dataset.

LLaMA-65B GPT-3 OPT-175B Guanaco-65B
Gender 70.6 62.6 65.7 47.5
Religion {79.0} 73.3 68.6 38.7
Race/Color 57.0 64.7 68.6 45.3
Sexual orientation {81.0} 76.2 78.6 59.1
Age 70.1 64.4 67.8 36.3
Nationality 64.2 61.6 62.9 32.4
Disability 66.7 76.7 76.7 33.9
Physical appearance 77.8 74.6 76.2 43.1
Socioeconomic status 71.5 73.8 76.2 55.3
Average 66.6 67.2 69.5 43.5

Citation

@article{dettmers2023qlora,
  title={QLoRA: Efficient Finetuning of Quantized LLMs},
  author={Dettmers, Tim and Pagnoni, Artidoro and Holtzman, Ari and Zettlemoyer, Luke},
  journal={arXiv preprint arXiv:2305.14314},
  year={2023}
}