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 incomplete list of clients and libraries that are known to support GGUF:
llama.cpp. This is the source project for GGUF, providing both a Command Line Interface (CLI) and a server option.
text-generation-webui, Known as the most widely used web UI, this project boasts numerous features and powerful extensions, and supports GPU acceleration.
Ollama Ollama is a lightweight and extensible framework designed for building and running language models locally. It features a simple API for creating, managing, and executing models, along with a library of pre-built models for use in various applications
KoboldCpp, A comprehensive web UI offering GPU acceleration across all platforms and architectures, particularly renowned for storytelling.
GPT4All, This is a free and open source GUI that runs locally, supporting Windows, Linux, and macOS with full GPU acceleration.
LM Studio An intuitive and powerful local GUI for Windows and macOS (Silicon), featuring GPU acceleration.
LoLLMS Web UI. A notable web UI with a variety of unique features, including a comprehensive model library for easy model selection.
Faraday.dev, An attractive, user-friendly character-based chat GUI for Windows and macOS (both Silicon and Intel), also offering GPU acceleration.
llama-cpp-python, A Python library equipped with GPU acceleration, LangChain support, and an OpenAI-compatible API server.
candle, A Rust-based ML framework focusing on performance, including GPU support, and designed for ease of use.
ctransformers, A Python library featuring GPU acceleration, LangChain support, and an OpenAI-compatible AI server.
localGPT An open-source initiative enabling private conversations with documents.
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.
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 folder.
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: LiteLLMs/Llama3-OpenBioLLM-70B-GGUF and below it, a specific filename to download, such as: Q4_0/Q4_0-00001-of-00009.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:
Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change -c 8192 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. Note that longer sequence lengths require much more resources, so you may need to reduce this value.
If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins
You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.
How to load this model in Python code, using llama-cpp-python
Run one of the following commands, according to your system:
shell
1# Base ctransformers with no GPU acceleration2pip install llama-cpp-python
3# With NVidia CUDA acceleration4CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
5# Or with OpenBLAS acceleration6CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
7# Or with CLBLast acceleration8CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
9# Or with AMD ROCm GPU acceleration (Linux only)10CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
11# Or with Metal GPU acceleration for macOS systems only12CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
13# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:14$env:CMAKE_ARGS ="-DLLAMA_OPENBLAS=on"15pip install llama-cpp-python
Simple llama-cpp-python example code
python
1from llama_cpp import Llama
2# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.3llm = Llama(4 model_path="./Q4_0/Q4_0-00001-of-00009.gguf",# Download the model file first5 n_ctx=32768,# The max sequence length to use - note that longer sequence lengths require much more resources6 n_threads=8,# The number of CPU threads to use, tailor to your system and the resulting performance7 n_gpu_layers=35# The number of layers to offload to GPU, if you have GPU acceleration available8)9# Simple inference example10output = llm(11"<PROMPT>",# Prompt12 max_tokens=512,# Generate up to 512 tokens13 stop=["</s>"],# Example stop token - not necessarily correct for this specific model! Please check before using.14 echo=True# Whether to echo the prompt15)16# Chat Completion API17llm = Llama(model_path="./Q4_0/Q4_0-00001-of-00009.gguf", chat_format="llama-2")# Set chat_format according to the model you are using18llm.create_chat_completion(19 messages =[20{"role":"system","content":"You are a story writing assistant."},21{22"role":"user",23"content":"Write a story about llamas."24}25]26)
How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
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 incomplete list of clients and libraries that are known to support GGUF:
llama.cpp. This is the source project for GGUF, providing both a Command Line Interface (CLI) and a server option.
text-generation-webui, Known as the most widely used web UI, this project boasts numerous features and powerful extensions, and supports GPU acceleration.
Ollama Ollama is a lightweight and extensible framework designed for building and running language models locally. It features a simple API for creating, managing, and executing models, along with a library of pre-built models for use in various applications
KoboldCpp, A comprehensive web UI offering GPU acceleration across all platforms and architectures, particularly renowned for storytelling.
GPT4All, This is a free and open source GUI that runs locally, supporting Windows, Linux, and macOS with full GPU acceleration.
LM Studio An intuitive and powerful local GUI for Windows and macOS (Silicon), featuring GPU acceleration.
LoLLMS Web UI. A notable web UI with a variety of unique features, including a comprehensive model library for easy model selection.
Faraday.dev, An attractive, user-friendly character-based chat GUI for Windows and macOS (both Silicon and Intel), also offering GPU acceleration.
llama-cpp-python, A Python library equipped with GPU acceleration, LangChain support, and an OpenAI-compatible API server.
candle, A Rust-based ML framework focusing on performance, including GPU support, and designed for ease of use.
ctransformers, A Python library featuring GPU acceleration, LangChain support, and an OpenAI-compatible AI server.
localGPT An open-source initiative enabling private conversations with documents.
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.
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 folder.
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: LiteLLMs/Llama3-OpenBioLLM-70B-GGUF and below it, a specific filename to download, such as: Q4_0/Q4_0-00001-of-00009.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:
Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change -c 8192 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. Note that longer sequence lengths require much more resources, so you may need to reduce this value.
If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins
You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.
How to load this model in Python code, using llama-cpp-python
Run one of the following commands, according to your system:
shell
1# Base ctransformers with no GPU acceleration2pip install llama-cpp-python
3# With NVidia CUDA acceleration4CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
5# Or with OpenBLAS acceleration6CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
7# Or with CLBLast acceleration8CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
9# Or with AMD ROCm GPU acceleration (Linux only)10CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
11# Or with Metal GPU acceleration for macOS systems only12CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
13# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:14$env:CMAKE_ARGS ="-DLLAMA_OPENBLAS=on"15pip install llama-cpp-python
Simple llama-cpp-python example code
python
1from llama_cpp import Llama
2# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.3llm = Llama(4 model_path="./Q4_0/Q4_0-00001-of-00009.gguf",# Download the model file first5 n_ctx=32768,# The max sequence length to use - note that longer sequence lengths require much more resources6 n_threads=8,# The number of CPU threads to use, tailor to your system and the resulting performance7 n_gpu_layers=35# The number of layers to offload to GPU, if you have GPU acceleration available8)9# Simple inference example10output = llm(11"<PROMPT>",# Prompt12 max_tokens=512,# Generate up to 512 tokens13 stop=["</s>"],# Example stop token - not necessarily correct for this specific model! Please check before using.14 echo=True# Whether to echo the prompt15)16# Chat Completion API17llm = Llama(model_path="./Q4_0/Q4_0-00001-of-00009.gguf", chat_format="llama-2")# Set chat_format according to the model you are using18llm.create_chat_completion(19 messages =[20{"role":"system","content":"You are a story writing assistant."},21{22"role":"user",23"content":"Write a story about llamas."24}25]26)
How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
Introducing OpenBioLLM-70B: A State-of-the-Art Open Source Biomedical Large Language Model
OpenBioLLM-70B is an advanced open source language model designed specifically for the biomedical domain. Developed by Saama AI Labs, this model leverages cutting-edge techniques to achieve state-of-the-art performance on a wide range of biomedical tasks.
🏥 Biomedical Specialization: OpenBioLLM-70B is tailored for the unique language and knowledge requirements of the medical and life sciences fields. It was fine-tuned on a vast corpus of high-quality biomedical data, enabling it to understand and generate text with domain-specific accuracy and fluency.
🎓 Superior Performance: With 70 billion parameters, OpenBioLLM-70B outperforms other open source biomedical language models of similar scale. It has also demonstrated better results compared to larger proprietary & open-source models like GPT-4, Gemini, Meditron-70B, Med-PaLM-1 & Med-PaLM-2 on biomedical benchmarks.
🧠 Advanced Training Techniques: OpenBioLLM-70B builds upon the powerful foundations of the Meta-Llama-3-70B-Instruct and Meta-Llama-3-70B-Instruct models. It incorporates the DPO dataset and fine-tuning recipe along with a custom diverse medical instruction dataset. Key components of the training pipeline include:
The model can be fine-tuned for more specialized tasks and datasets as needed.
OpenBioLLM-70B represents an important step forward in democratizing advanced language AI for the biomedical community. By leveraging state-of-the-art architectures and training techniques from leading open source efforts like Llama-3, we have created a powerful tool to accelerate innovation and discovery in healthcare and the life sciences.
We are excited to share OpenBioLLM-70B with researchers and developers around the world.
Use with transformers
Important: Please use the exact chat template provided by Llama-3 instruct version. Otherwise there will be a degradation in the performance. The model output can be verbose in rare cases. Please consider setting temperature = 0 to make this happen less.
See the snippet below for usage with Transformers:
python
1import transformers
2import torch
34model_id ="aaditya/OpenBioLLM-Llama3-70B"56pipeline = transformers.pipeline(7"text-generation",8 model=model_id,9 model_kwargs={"torch_dtype": torch.bfloat16},10 device="auto",11)1213messages =[14{"role":"system","content":"You are an expert and experienced from the healthcare and biomedical domain with extensive medical knowledge and practical experience. Your name is OpenBioLLM, and you were developed by Saama AI Labs. who's willing to help answer the user's query with explanation. In your explanation, leverage your deep medical expertise such as relevant anatomical structures, physiological processes, diagnostic criteria, treatment guidelines, or other pertinent medical concepts. Use precise medical terminology while still aiming to make the explanation clear and accessible to a general audience."},15{"role":"user","content":"How can i split a 3mg or 4mg waefin pill so i can get a 2.5mg pill?"},16]1718prompt = pipeline.tokenizer.apply_chat_template(19 messages,20 tokenize=False,21 add_generation_prompt=True22)2324terminators =[25 pipeline.tokenizer.eos_token_id,26 pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")27]2829outputs = pipeline(30 prompt,31 max_new_tokens=256,32 eos_token_id=terminators,33 do_sample=True,34 temperature=0.0,35 top_p=0.9,36)37print(outputs[0]["generated_text"][len(prompt):])
Training procedure
Training hyperparameters
Click to see details
learning_rate: 0.0002
lr_scheduler: cosine
train_batch_size: 12
eval_batch_size: 8
GPU: H100 80GB SXM5
num_devices: 8
optimizer: adamw_bnb_8bit
lr_scheduler_warmup_steps: 100
num_epochs: 4
Peft hyperparameters
Click to see details
adapter: qlora
lora_r: 128
lora_alpha: 256
lora_dropout: 0.05
lora_target_linear: true
-lora_target_modules:
q_proj
v_proj
k_proj
o_proj
gate_proj
down_proj
up_proj
Training results
Framework versions
Transformers 4.39.3
Pytorch 2.1.2+cu121
Datasets 2.18.0
Tokenizers 0.15.1
Axolotl
Lm harness for evaluation
Benchmark Results
🔥 OpenBioLLM-70B demonstrates superior performance compared to larger models, such as GPT-4, Gemini, Meditron-70B, Med-PaLM-1 & Med-PaLM-2 across 9 diverse biomedical datasets, achieving state-of-the-art results with an average score of 86.06%, despite having a significantly smaller parameter count. The model's strong performance in domain-specific tasks, such as Clinical KG, Medical Genetics, and PubMedQA, highlights its ability to effectively capture and apply biomedical knowledge.
🚨 The GPT-4, Med-PaLM-1, and Med-PaLM-2 results are taken from their official papers. Since Med-PaLM doesn't provide zero-shot accuracy, we are using 5-shot accuracy from their paper for comparison. All results presented are in the zero-shot setting, except for Med-PaLM-2 and Med-PaLM-1, which use 5-shot accuracy.
🚨 **Below results are from the quantized version of OpenBioLLM-70B
Summarize Clinical Notes
OpenBioLLM-70B can efficiently analyze and summarize complex clinical notes, EHR data, and discharge summaries, extracting key information and generating concise, structured summaries
image/png
Answer Medical Questions
OpenBioLLM-70B can provide answers to a wide range of medical questions.
image/png
image/png
Click to see details
image/png
image/png
image/png
Clinical Entity Recognition
OpenBioLLM-70B can perform advanced clinical entity recognition by identifying and extracting key medical concepts, such as diseases, symptoms, medications, procedures, and anatomical structures, from unstructured clinical text. By leveraging its deep understanding of medical terminology and context, the model can accurately annotate and categorize clinical entities, enabling more efficient information retrieval, data analysis, and knowledge discovery from electronic health records, research articles, and other biomedical text sources. This capability can support various downstream applications, such as clinical decision support, pharmacovigilance, and medical research.
image/png
image/png
image/png
Biomarkers Extraction
image/png
Classification
OpenBioLLM-70B can perform various biomedical classification tasks, such as disease prediction, sentiment analysis, medical document categorization
image/png
De-Identification
OpenBioLLM-70B can detect and remove personally identifiable information (PII) from medical records, ensuring patient privacy and compliance with data protection regulations like HIPAA.
image/png
Advisory Notice!
While OpenBioLLM-70B leverages high-quality data sources, its outputs may still contain inaccuracies, biases, or misalignments that could pose risks if relied upon for medical decision-making without further testing and refinement. The model's performance has not yet been rigorously evaluated in randomized controlled trials or real-world healthcare environments.
Therefore, we strongly advise against using OpenBioLLM-70B for any direct patient care, clinical decision support, or other professional medical purposes at this time. Its use should be limited to research, development, and exploratory applications by qualified individuals who understand its limitations.
OpenBioLLM-70B is intended solely as a research tool to assist healthcare professionals and should never be considered a replacement for the professional judgment and expertise of a qualified medical doctor.
Appropriately adapting and validating OpenBioLLM-70B for specific medical use cases would require significant additional work, potentially including:
Thorough testing and evaluation in relevant clinical scenarios
Alignment with evidence-based guidelines and best practices
Mitigation of potential biases and failure modes
Integration with human oversight and interpretation
Compliance with regulatory and ethical standards
Always consult a qualified healthcare provider for personal medical needs.
Citation
If you find OpenBioLLM-70B & 8B useful in your work, please cite the model as follows:
@misc{OpenBioLLMs,
author = {Ankit Pal, Malaikannan Sankarasubbu},
title = {OpenBioLLMs: Advancing Open-Source Large Language Models for Healthcare and Life Sciences},
year = {2024},
publisher = {Hugging Face},
journal = {Hugging Face repository},
howpublished = {\url{https://huggingface.co/aaditya/OpenBioLLM-Llama3-70B}}
}
The accompanying paper is currently in progress and will be released soon.
💌 Contact
We look forward to hearing you and collaborating on this exciting project!
Note: I am looking for a funded PhD opportunity, especially if it fits my Responsible Generative AI, Multimodal LLMs, Geometric Deep Learning, and Healthcare AI skillset.
