TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)
This repo contains GGUF format model files for Code Llama's Codellama 70B Instruct.
These files were quantised using hardware kindly provided by Massed Compute.
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:
Source: system
{system_message}<step> Source: user
{prompt} <step> Source: assistant
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.
The new methods available are:
Refer to the Provided Files table below to see what files use which methods, and how.
| Name | Quant method | Bits | Size | Max RAM required | Use case |
|---|---|---|---|---|---|
| codellama-70b-instruct.Q2_K.gguf | Q2_K | 2 | 25.46 GB | 27.96 GB | significant quality loss - not recommended for most purposes |
| codellama-70b-instruct.Q3_K_S.gguf | Q3_K_S | 3 | 29.92 GB | 32.42 GB | very small, high quality loss |
| codellama-70b-instruct.Q3_K_M.gguf | Q3_K_M | 3 | 33.27 GB | 35.77 GB | very small, high quality loss |
| codellama-70b-instruct.Q3_K_L.gguf | Q3_K_L | 3 | 36.15 GB | 38.65 GB | small, substantial quality loss |
| codellama-70b-instruct.Q4_0.gguf | Q4_0 | 4 | 38.87 GB | 41.37 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| codellama-70b-instruct.Q4_K_S.gguf | Q4_K_S | 4 | 39.25 GB | 41.75 GB | small, greater quality loss |
| codellama-70b-instruct.Q4_K_M.gguf | Q4_K_M | 4 | 41.42 GB | 43.92 GB | medium, balanced quality - recommended |
| codellama-70b-instruct.Q5_0.gguf | Q5_0 | 5 | 47.46 GB | 49.96 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| codellama-70b-instruct.Q5_K_S.gguf | Q5_K_S | 5 | 47.46 GB | 49.96 GB | large, low quality loss - recommended |
| codellama-70b-instruct.Q5_K_M.gguf | Q5_K_M | 5 | 48.75 GB | 51.25 GB | large, very low quality loss - recommended |
| codellama-70b-instruct.Q6_K.gguf | Q6_K | 6 | 56.59 GB | 59.09 GB | very large, extremely low quality loss |
| codellama-70b-instruct.Q8_0.gguf | Q8_0 | 8 | 73.29 GB | 75.79 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.
Note: HF does not support uploading files larger than 50GB. Therefore I have uploaded the Q6_K and Q8_0 files as split files.
Please download:
codellama-70b-instruct.Q6_K.gguf-split-acodellama-70b-instruct.Q6_K.gguf-split-bPlease download:
codellama-70b-instruct.Q8_0.gguf-split-acodellama-70b-instruct.Q8_0.gguf-split-bTo join the files, do the following:
Linux and macOS:
cat codellama-70b-instruct.Q6_K.gguf-split-* > codellama-70b-instruct.Q6_K.gguf && rm codellama-70b-instruct.Q6_K.gguf-split-*
cat codellama-70b-instruct.Q8_0.gguf-split-* > codellama-70b-instruct.Q8_0.gguf && rm codellama-70b-instruct.Q8_0.gguf-split-*
Windows command line:
COPY /B codellama-70b-instruct.Q6_K.gguf-split-a + codellama-70b-instruct.Q6_K.gguf-split-b codellama-70b-instruct.Q6_K.gguf
del codellama-70b-instruct.Q6_K.gguf-split-a codellama-70b-instruct.Q6_K.gguf-split-b
COPY /B codellama-70b-instruct.Q8_0.gguf-split-a + codellama-70b-instruct.Q8_0.gguf-split-b codellama-70b-instruct.Q8_0.gguf
del codellama-70b-instruct.Q8_0.gguf-split-a codellama-70b-instruct.Q8_0.gguf-split-b
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:
text-generation-webui
Under Download Model, you can enter the model repo: TheBloke/CodeLlama-70B-Instruct-GGUF and below it, a specific filename to download, such as: codellama-70b-instruct.Q4_K_M.gguf.
Then click Download.
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/CodeLlama-70B-Instruct-GGUF codellama-70b-instruct.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
You can also download multiple files at once with a pattern:
huggingface-cli download TheBloke/CodeLlama-70B-Instruct-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/CodeLlama-70B-Instruct-GGUF codellama-70b-instruct.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.
llama.cpp command
Make sure you are using llama.cpp from commit d0cee0d or later.
./main -ngl 35 -m codellama-70b-instruct.Q4_K_M.gguf --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "Source: system\n\n {system_message}<step> Source: user\n\n {prompt} <step> Source: assistant"
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. 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
For other parameters and how to use them, please refer to the llama.cpp documentation
text-generation-webui
Further instructions can be found in the text-generation-webui documentation, here: text-generation-webui/docs/04 ‐ Model Tab.md.
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.
For full documentation, please see: llama-cpp-python docs.
Run one of the following commands, according to your system:
# Base ctransformers with no GPU acceleration
pip install llama-cpp-python
# With NVidia CUDA acceleration
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
# Or with OpenBLAS acceleration
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
# Or with CLBLast acceleration
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
# Or with AMD ROCm GPU acceleration (Linux only)
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
# Or with Metal GPU acceleration for macOS systems only
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
pip install llama-cpp-python
from llama_cpp import Llama
# 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 = Llama(
model_path="./codellama-70b-instruct.Q4_K_M.gguf", # Download the model file first
n_ctx=4096, # The max sequence length to use - note that longer sequence lengths require much more resources
n_threads=8, # The number of CPU threads to use, tailor to your system and the resulting performance
n_gpu_layers=35 # The number of layers to offload to GPU, if you have GPU acceleration available
)
# Simple inference example
output = llm(
"Source: system\n\n {system_message}<step> Source: user\n\n {prompt} <step> Source: assistant", # Prompt
max_tokens=512, # Generate up to 512 tokens
stop=["</s>"], # Example stop token - not necessarily correct for this specific model! Please check before using.
echo=True # Whether to echo the prompt
)
# Chat Completion API
llm = Llama(model_path="./codellama-70b-instruct.Q4_K_M.gguf", chat_format="llama-2") # Set chat_format according to the model you are using
llm.create_chat_completion(
messages = [
{"role": "system", "content": "You are a story writing assistant."},
{
"role": "user",
"content": "Write a story about llamas."
}
]
)
Here are guides on using llama-cpp-python and ctransformers with LangChain:
For further support, and discussions on these models and AI in general, join us at:
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: Michael Levine, 阿明, Trailburnt, Nikolai Manek, John Detwiler, Randy H, Will Dee, Sebastain Graf, NimbleBox.ai, Eugene Pentland, Emad Mostaque, Ai Maven, Jim Angel, Jeff Scroggin, Michael Davis, Manuel Alberto Morcote, Stephen Murray, Robert, Justin Joy, Luke @flexchar, Brandon Frisco, Elijah Stavena, S_X, Dan Guido, Undi ., Komninos Chatzipapas, Shadi, theTransient, Lone Striker, Raven Klaugh, jjj, Cap'n Zoog, Michel-Marie MAUDET (LINAGORA), Matthew Berman, David, Fen Risland, Omer Bin Jawed, Luke Pendergrass, Kalila, OG, Erik Bjäreholt, Rooh Singh, Joseph William Delisle, Dan Lewis, TL, John Villwock, AzureBlack, Brad, Pedro Madruga, Caitlyn Gatomon, K, jinyuan sun, Mano Prime, Alex, Jeffrey Morgan, Alicia Loh, Illia Dulskyi, Chadd, transmissions 11, fincy, Rainer Wilmers, ReadyPlayerEmma, knownsqashed, Mandus, biorpg, Deo Leter, Brandon Phillips, SuperWojo, Sean Connelly, Iucharbius, Jack West, Harry Royden McLaughlin, Nicholas, terasurfer, Vitor Caleffi, Duane Dunston, Johann-Peter Hartmann, David Ziegler, Olakabola, Ken Nordquist, Trenton Dambrowitz, Tom X Nguyen, Vadim, Ajan Kanaga, Leonard Tan, Clay Pascal, Alexandros Triantafyllidis, JM33133, Xule, vamX, ya boyyy, subjectnull, Talal Aujan, Alps Aficionado, wassieverse, Ari Malik, James Bentley, Woland, Spencer Kim, Michael Dempsey, Fred von Graf, Elle, zynix, William Richards, Stanislav Ovsiannikov, Edmond Seymore, Jonathan Leane, Martin Kemka, usrbinkat, Enrico Ros
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
Code Llama is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 70B instruct-tuned version in the Hugging Face Transformers format. This model is designed for general code synthesis and understanding. Links to other models can be found in the index at the bottom.
Model capabilities:
Install transformers
pip install transformers accelerate
Chat use: The 70B Instruct model uses a different prompt template than the smaller versions. To use it with transformers, we recommend you use the built-in chat template:
from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch
model_id = "codellama/CodeLlama-70b-Instruct-hf"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float16,
device_map="auto",
)
chat = [
{"role": "system", "content": "You are a helpful and honest code assistant expert in JavaScript. Please, provide all answers to programming questions in JavaScript"},
{"role": "user", "content": "Write a function that computes the set of sums of all contiguous sublists of a given list."},
]
inputs = tokenizer.apply_chat_template(chat, return_tensors="pt").to("cuda")
output = model.generate(input_ids=inputs, max_new_tokens=200)
output = output[0].to("cpu")
print(tokenizer.decode(output))
You can also use the model for text or code completion. This examples uses transformers' pipeline interface:
from transformers import AutoTokenizer
import transformers
import torch
model_id = "codellama/CodeLlama-70b-hf"
tokenizer = AutoTokenizer.from_pretrained(model_id)
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
torch_dtype=torch.float16,
device_map="auto",
)
sequences = pipeline(
'def fibonacci(',
do_sample=True,
temperature=0.2,
top_p=0.9,
num_return_sequences=1,
eos_token_id=tokenizer.eos_token_id,
max_length=100,
)
for seq in sequences:
print(f"Result: {seq['generated_text']}")
CodeLlama 70B Instruct uses a different format for the chat prompt than previous Llama 2 or CodeLlama models. As mentioned above, the easiest way to use it is with the help of the tokenizer's chat template. If you need to build the string or tokens, manually, here's how to do it.
We'll do our tests with the following made-up dialog:
chat = [
{"role": "system", "content": "System prompt "},
{"role": "user", "content": "First user query"},
{"role": "assistant", "content": "Model response to first query"},
{"role": "user", "content": "Second user query"},
]
First, let's see what the prompt looks like if we use the chat template:
tokenizer.apply_chat_template(chat, tokenize=False)
'<s>Source: system\n\n System prompt <step> Source: user\n\n First user query <step> Source: assistant\n\n Model response to first query <step> Source: user\n\n Second user query <step> Source: assistant\nDestination: user\n\n '
So each turn of the conversation has a Source (system, user, or assistant), and then the content appears after two newlines and a space. Turns are separated with the special token <step>. After the last turn (which must necessarily come from the user), we invite the model to respond by using the special syntax Source: assistant\nDestination: user\n\n . Let's see how we can build the same string ourselves:
output = "<s>"
for m in chat:
output += f"Source: {m['role']}\n\n {m['content'].strip()}"
output += " <step> "
output += "Source: assistant\nDestination: user\n\n "
output
'<s>Source: system\n\n System prompt <step> Source: user\n\n First user query <step> Source: assistant\n\n Model response to first query <step> Source: user\n\n Second user query <step> Source: assistant\nDestination: user\n\n '
To verify that we got it right, we'll compare against the reference code in the original GitHub repo. We used the same dialog and tokenized it with the dialog_prompt_tokens function and got the following tokens:
reference_tokens = [1, 7562, 29901, 1788, 13, 13, 2184, 9508, 32015, 7562, 29901, 1404, 13, 13, 3824, 1404, 2346, 32015, 7562, 29901, 20255, 13, 13, 8125, 2933, 304, 937, 2346, 32015, 7562, 29901, 1404, 13, 13, 6440, 1404, 2346, 32015, 7562, 29901, 20255, 13, 14994, 3381, 29901, 1404, 13, 13, 29871]
Let's see what we get with the string we built using our Python loop. Note that we don't add "special tokens" because the string already starts with <s>, the beginning of sentence token:
tokens = tokenizer.encode(output, add_special_tokens=False)
assert reference_tokens == tokens
Similarly, let's verify that the chat template produces the same token sequence:
assert reference_tokens == tokenizer.apply_chat_template(chat)
As a final detail, please note that if the dialog does not start with a system turn, the original code will insert one with an empty content string.
*Note: Use of this model is governed by the Meta license. Meta developed and publicly released the Code Llama family of large language models (LLMs).
Model Developers Meta
Variations Code Llama comes in four model sizes, and three variants:
All variants are available in sizes of 7B, 13B, 34B, and 70B parameters.
This repository contains the Instruct version of the 70B parameters model.
Input Models input text only.
Output Models generate text only.
Model Architecture Code Llama is an auto-regressive language model that uses an optimized transformer architecture. It was fine-tuned with up to 16k tokens. This variant does not support long context of up to 100k tokens.
Model Dates Code Llama and its variants have been trained between January 2023 and January 2024.
Status This is a static model trained on an offline dataset. Future versions of Code Llama - Instruct will be released as we improve model safety with community feedback.
License A custom commercial license is available at: https://ai.meta.com/resources/models-and-libraries/llama-downloads/
Research Paper More information can be found in the paper "Code Llama: Open Foundation Models for Code" or its arXiv page.
Intended Use Cases Code Llama and its variants are intended for commercial and research use in English and relevant programming languages. The base model Code Llama can be adapted for a variety of code synthesis and understanding tasks, Code Llama - Python is designed specifically to handle the Python programming language, and Code Llama - Instruct is intended to be safer to use for code assistant and generation applications.
Out-of-Scope Uses Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English. Use in any other way that is prohibited by the Acceptable Use Policy and Licensing Agreement for Code Llama and its variants.
Training Factors We used custom training libraries. The training and fine-tuning of the released models have been performed Meta’s Research Super Cluster. Carbon Footprint In aggregate, training all 12 Code Llama models required 1400K GPU hours of computation on hardware of type A100-80GB (TDP of 350-400W). Estimated total emissions were 228.55 tCO2eq, 100% of which were offset by Meta’s sustainability program.
See evaluations for the main models and detailed ablations in Section 3 and safety evaluations in Section 4 of the research paper.
Code Llama and its variants are a new technology that carries risks with use. Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, Code Llama’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate or objectionable responses to user prompts. Therefore, before deploying any applications of Code Llama, developers should perform safety testing and tuning tailored to their specific applications of the model.
Please see the Responsible Use Guide available available at https://ai.meta.com/llama/responsible-use-guide.