llama.cpp/README.md

793 lines
30 KiB
Markdown
Raw Normal View History

2023-09-12 23:02:30 +00:00
# 🦙 Python Bindings for [`llama.cpp`](https://github.com/ggerganov/llama.cpp)
2023-03-24 04:06:24 +00:00
2023-06-26 20:35:38 +00:00
[![Documentation Status](https://readthedocs.org/projects/llama-cpp-python/badge/?version=latest)](https://llama-cpp-python.readthedocs.io/en/latest/?badge=latest)
[![Tests](https://github.com/abetlen/llama-cpp-python/actions/workflows/test.yaml/badge.svg?branch=main)](https://github.com/abetlen/llama-cpp-python/actions/workflows/test.yaml)
2023-03-24 04:06:24 +00:00
[![PyPI](https://img.shields.io/pypi/v/llama-cpp-python)](https://pypi.org/project/llama-cpp-python/)
[![PyPI - Python Version](https://img.shields.io/pypi/pyversions/llama-cpp-python)](https://pypi.org/project/llama-cpp-python/)
[![PyPI - License](https://img.shields.io/pypi/l/llama-cpp-python)](https://pypi.org/project/llama-cpp-python/)
[![PyPI - Downloads](https://img.shields.io/pypi/dm/llama-cpp-python)](https://pypi.org/project/llama-cpp-python/)
[![Github All Releases](https://img.shields.io/github/downloads/abetlen/llama-cpp-python/total.svg?label=Github%20Downloads)]()
2023-03-23 09:33:06 +00:00
2023-03-24 03:55:42 +00:00
Simple Python bindings for **@ggerganov's** [`llama.cpp`](https://github.com/ggerganov/llama.cpp) library.
2023-03-23 20:00:10 +00:00
This package provides:
2023-03-23 09:33:06 +00:00
2023-03-24 03:55:42 +00:00
- Low-level access to C API via `ctypes` interface.
- High-level Python API for text completion
- OpenAI-like API
- [LangChain compatibility](https://python.langchain.com/docs/integrations/llms/llamacpp)
- [LlamaIndex compatibility](https://docs.llamaindex.ai/en/stable/examples/llm/llama_2_llama_cpp.html)
- OpenAI compatible web server
- [Local Copilot replacement](https://llama-cpp-python.readthedocs.io/en/latest/server/#code-completion)
- [Function Calling support](https://llama-cpp-python.readthedocs.io/en/latest/server/#function-calling)
- [Vision API support](https://llama-cpp-python.readthedocs.io/en/latest/server/#multimodal-models)
- [Multiple Models](https://llama-cpp-python.readthedocs.io/en/latest/server/#configuration-and-multi-model-support)
2023-12-22 19:40:44 +00:00
2023-06-26 20:35:38 +00:00
Documentation is available at [https://llama-cpp-python.readthedocs.io/en/latest](https://llama-cpp-python.readthedocs.io/en/latest).
2023-05-17 15:40:12 +00:00
2023-11-28 07:37:34 +00:00
## Installation
2023-03-23 09:33:06 +00:00
2024-02-22 08:25:11 +00:00
Requirements:
- Python 3.8+
- C compiler
- Linux: gcc or clang
- Windows: Visual Studio or MinGW
- MacOS: Xcode
To install the package, run:
2023-03-23 09:33:06 +00:00
```bash
2023-03-23 18:24:34 +00:00
pip install llama-cpp-python
2023-03-23 09:33:06 +00:00
```
2024-02-22 08:25:11 +00:00
This will also build `llama.cpp` from source and install it alongside this python package.
2023-04-28 21:08:18 +00:00
2024-02-22 08:25:11 +00:00
If this fails, add `--verbose` to the `pip install` see the full cmake build log.
2023-11-28 08:15:01 +00:00
**Pre-built Wheel (New)**
It is also possible to install a pre-built wheel with basic CPU support.
```bash
pip install llama-cpp-python \
--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
```
2024-02-22 08:25:11 +00:00
### Installation Configuration
2023-11-28 08:15:01 +00:00
2024-02-22 08:25:11 +00:00
`llama.cpp` supports a number of hardware acceleration backends to speed up inference as well as backend specific options. See the [llama.cpp README](https://github.com/ggerganov/llama.cpp#build) for a full list.
2023-11-28 08:15:01 +00:00
2024-02-22 08:25:11 +00:00
All `llama.cpp` cmake build options can be set via the `CMAKE_ARGS` environment variable or via the `--config-settings / -C` cli flag during installation.
2023-11-28 08:15:01 +00:00
2024-02-22 08:53:52 +00:00
<details open>
2024-02-22 08:25:11 +00:00
<summary>Environment Variables</summary>
2023-05-19 06:20:41 +00:00
```bash
2024-02-22 08:25:11 +00:00
# Linux and Mac
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" \
pip install llama-cpp-python
2023-05-19 06:20:41 +00:00
```
2024-02-22 08:25:11 +00:00
```powershell
# Windows
2023-11-28 08:15:01 +00:00
$env:CMAKE_ARGS = "-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS"
pip install llama-cpp-python
```
2024-02-22 08:25:11 +00:00
</details>
<details>
<summary>CLI / requirements.txt</summary>
They can also be set via `pip install -C / --config-settings` command and saved to a `requirements.txt` file:
```bash
pip install --upgrade pip # ensure pip is up to date
pip install llama-cpp-python \
-C cmake.args="-DLLAMA_BLAS=ON;-DLLAMA_BLAS_VENDOR=OpenBLAS"
```
```txt
# requirements.txt
llama-cpp-python -C cmake.args="-DLLAMA_BLAS=ON;-DLLAMA_BLAS_VENDOR=OpenBLAS"
```
</details>
2023-05-07 09:20:04 +00:00
2024-02-22 08:25:11 +00:00
### Supported Backends
Below are some common backends, their build commands and any additional environment variables required.
2024-02-22 08:53:52 +00:00
<details open>
2024-02-22 08:25:11 +00:00
<summary>OpenBLAS (CPU)</summary>
To install with OpenBLAS, set the `LLAMA_BLAS` and `LLAMA_BLAS_VENDOR` environment variables before installing:
2023-05-07 09:20:04 +00:00
```bash
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
2023-05-07 09:20:04 +00:00
```
2024-02-22 08:25:11 +00:00
</details>
2023-05-07 09:20:04 +00:00
2024-02-22 08:25:11 +00:00
<details>
2024-04-04 07:08:47 +00:00
<summary>CUDA</summary>
2023-11-28 08:15:01 +00:00
2024-04-04 07:08:47 +00:00
To install with CUDA support, set the `LLAMA_CUDA=on` environment variable before installing:
2023-05-07 09:20:04 +00:00
```bash
2024-04-04 06:49:19 +00:00
CMAKE_ARGS="-DLLAMA_CUDA=on" pip install llama-cpp-python
2023-05-07 09:20:04 +00:00
```
**Pre-built Wheel (New)**
It is also possible to install a pre-built wheel with CUDA support. As long as your system meets some requirements:
- CUDA Version is 12.1, 12.2, 12.3, or 12.4
- Python Version is 3.10, 3.11 or 3.12
```bash
pip install llama-cpp-python \
--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/<cuda-version>
```
Where `<cuda-version>` is one of the following:
- `cu121`: CUDA 12.1
- `cu122`: CUDA 12.2
- `cu123`: CUDA 12.3
- `cu124`: CUDA 12.4
For example, to install the CUDA 12.1 wheel:
```bash
pip install llama-cpp-python \
--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121
```
2024-02-22 08:25:11 +00:00
</details>
<details>
<summary>Metal</summary>
2023-05-07 09:20:04 +00:00
2023-06-10 22:59:26 +00:00
To install with Metal (MPS), set the `LLAMA_METAL=on` environment variable before installing:
```bash
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
2023-06-10 22:59:26 +00:00
```
2023-04-28 21:08:18 +00:00
**Pre-built Wheel (New)**
It is also possible to install a pre-built wheel with Metal support. As long as your system meets some requirements:
- MacOS Version is 11.0 or later
- Python Version is 3.10, 3.11 or 3.12
```bash
pip install llama-cpp-python \
--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/metal
```
2024-02-22 08:25:11 +00:00
</details>
<details>
<summary>CLBlast (OpenCL)</summary>
2023-11-28 08:15:01 +00:00
2024-01-29 16:02:25 +00:00
To install with CLBlast, set the `LLAMA_CLBLAST=on` environment variable before installing:
2023-08-25 21:19:23 +00:00
```bash
2023-11-28 08:15:01 +00:00
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
2023-08-25 21:19:23 +00:00
```
2024-02-22 08:25:11 +00:00
</details>
<details>
<summary>hipBLAS (ROCm)</summary>
2023-11-28 08:15:01 +00:00
To install with hipBLAS / ROCm support for AMD cards, set the `LLAMA_HIPBLAS=on` environment variable before installing:
2023-11-28 08:15:01 +00:00
```bash
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
```
2024-02-22 08:25:11 +00:00
</details>
<details>
<summary>Vulkan</summary>
2024-01-29 16:01:26 +00:00
To install with Vulkan support, set the `LLAMA_VULKAN=on` environment variable before installing:
```bash
CMAKE_ARGS="-DLLAMA_VULKAN=on" pip install llama-cpp-python
```
2024-02-22 08:25:11 +00:00
</details>
<details>
<summary>Kompute</summary>
2024-01-30 17:23:07 +00:00
To install with Kompute support, set the `LLAMA_KOMPUTE=on` environment variable before installing:
```bash
CMAKE_ARGS="-DLLAMA_KOMPUTE=on" pip install llama-cpp-python
```
2024-02-22 08:25:11 +00:00
</details>
2024-01-30 17:23:07 +00:00
2024-02-22 08:25:11 +00:00
<details>
<summary>SYCL</summary>
2024-01-30 17:23:07 +00:00
To install with SYCL support, set the `LLAMA_SYCL=on` environment variable before installing:
```bash
source /opt/intel/oneapi/setvars.sh
CMAKE_ARGS="-DLLAMA_SYCL=on -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx" pip install llama-cpp-python
2024-01-30 17:23:07 +00:00
```
2024-02-22 08:25:11 +00:00
</details>
2024-01-30 17:23:07 +00:00
2023-11-28 08:15:01 +00:00
### Windows Notes
2024-02-22 08:25:11 +00:00
<details>
<summary>Error: Can't find 'nmake' or 'CMAKE_C_COMPILER'</summary>
If you run into issues where it complains it can't find `'nmake'` `'?'` or CMAKE_C_COMPILER, you can extract w64devkit as [mentioned in llama.cpp repo](https://github.com/ggerganov/llama.cpp#openblas) and add those manually to CMAKE_ARGS before running `pip` install:
```ps
$env:CMAKE_GENERATOR = "MinGW Makefiles"
2024-01-30 17:23:07 +00:00
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on -DCMAKE_C_COMPILER=C:/w64devkit/bin/gcc.exe -DCMAKE_CXX_COMPILER=C:/w64devkit/bin/g++.exe"
```
See the above instructions and set `CMAKE_ARGS` to the BLAS backend you want to use.
2024-02-22 08:25:11 +00:00
</details>
2023-11-28 08:15:01 +00:00
### MacOS Notes
Detailed MacOS Metal GPU install documentation is available at [docs/install/macos.md](https://llama-cpp-python.readthedocs.io/en/latest/install/macos/)
2024-02-22 08:25:11 +00:00
<details>
<summary>M1 Mac Performance Issue</summary>
2023-11-28 08:15:01 +00:00
Note: If you are using Apple Silicon (M1) Mac, make sure you have installed a version of Python that supports arm64 architecture. For example:
```bash
2023-11-28 08:15:01 +00:00
wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-MacOSX-arm64.sh
bash Miniforge3-MacOSX-arm64.sh
```
2023-11-28 08:15:01 +00:00
Otherwise, while installing it will build the llama.cpp x86 version which will be 10x slower on Apple Silicon (M1) Mac.
2024-02-22 08:25:11 +00:00
</details>
2024-02-22 08:25:11 +00:00
<details>
<summary>M Series Mac Error: `(mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64'))`</summary>
Try installing with
```bash
CMAKE_ARGS="-DCMAKE_OSX_ARCHITECTURES=arm64 -DCMAKE_APPLE_SILICON_PROCESSOR=arm64 -DLLAMA_METAL=on" pip install --upgrade --verbose --force-reinstall --no-cache-dir llama-cpp-python
```
2024-02-22 08:25:11 +00:00
</details>
2023-11-28 08:15:01 +00:00
### Upgrading and Reinstalling
2024-02-22 08:25:11 +00:00
To upgrade and rebuild `llama-cpp-python` add `--upgrade --force-reinstall --no-cache-dir` flags to the `pip install` command to ensure the package is rebuilt from source.
2023-11-28 08:15:01 +00:00
2023-04-05 21:44:25 +00:00
## High-level API
2023-03-23 09:33:06 +00:00
[API Reference](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#high-level-api)
The high-level API provides a simple managed interface through the [`Llama`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama) class.
2023-05-07 05:41:19 +00:00
Below is a short example demonstrating how to use the high-level API to for basic text completion:
2023-05-07 05:41:19 +00:00
2023-03-23 09:33:06 +00:00
```python
from llama_cpp import Llama
llm = Llama(
model_path="./models/7B/llama-model.gguf",
2024-01-30 17:23:07 +00:00
# n_gpu_layers=-1, # Uncomment to use GPU acceleration
# seed=1337, # Uncomment to set a specific seed
# n_ctx=2048, # Uncomment to increase the context window
)
output = llm(
2023-11-27 23:29:13 +00:00
"Q: Name the planets in the solar system? A: ", # Prompt
2024-01-25 15:51:48 +00:00
max_tokens=32, # Generate up to 32 tokens, set to None to generate up to the end of the context window
2023-11-27 23:29:13 +00:00
stop=["Q:", "\n"], # Stop generating just before the model would generate a new question
echo=True # Echo the prompt back in the output
) # Generate a completion, can also call create_completion
print(output)
```
By default `llama-cpp-python` generates completions in an OpenAI compatible format:
```python
2023-03-23 09:33:06 +00:00
{
"id": "cmpl-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
"object": "text_completion",
"created": 1679561337,
"model": "./models/7B/llama-model.gguf",
2023-03-23 09:33:06 +00:00
"choices": [
{
"text": "Q: Name the planets in the solar system? A: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune and Pluto.",
"index": 0,
"logprobs": None,
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 14,
"completion_tokens": 28,
"total_tokens": 42
}
}
```
2023-03-24 04:06:24 +00:00
Text completion is available through the [`__call__`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.__call__) and [`create_completion`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_completion) methods of the [`Llama`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama) class.
2024-02-22 16:31:44 +00:00
### Pulling models from Hugging Face Hub
2024-02-22 16:31:44 +00:00
You can download `Llama` models in `gguf` format directly from Hugging Face using the [`from_pretrained`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.from_pretrained) method.
You'll need to install the `huggingface-hub` package to use this feature (`pip install huggingface-hub`).
```python
2024-02-22 08:25:11 +00:00
llm = Llama.from_pretrained(
repo_id="Qwen/Qwen1.5-0.5B-Chat-GGUF",
filename="*q8_0.gguf",
verbose=False
)
```
2024-02-22 16:31:44 +00:00
By default [`from_pretrained`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.from_pretrained) will download the model to the huggingface cache directory, you can then manage installed model files with the [`huggingface-cli`](https://huggingface.co/docs/huggingface_hub/en/guides/cli) tool.
2024-02-22 08:25:11 +00:00
### Chat Completion
The high-level API also provides a simple interface for chat completion.
Chat completion requires that the model knows how to format the messages into a single prompt.
The `Llama` class does this using pre-registered chat formats (ie. `chatml`, `llama-2`, `gemma`, etc) or by providing a custom chat handler object.
The model will will format the messages into a single prompt using the following order of precedence:
- Use the `chat_handler` if provided
- Use the `chat_format` if provided
- Use the `tokenizer.chat_template` from the `gguf` model's metadata (should work for most new models, older models may not have this)
- else, fallback to the `llama-2` chat format
Set `verbose=True` to see the selected chat format.
```python
from llama_cpp import Llama
llm = Llama(
model_path="path/to/llama-2/llama-model.gguf",
chat_format="llama-2"
)
llm.create_chat_completion(
2023-11-27 23:29:13 +00:00
messages = [
{"role": "system", "content": "You are an assistant who perfectly describes images."},
{
"role": "user",
"content": "Describe this image in detail please."
}
]
)
```
Chat completion is available through the [`create_chat_completion`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_chat_completion) method of the [`Llama`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama) class.
2024-02-22 16:31:44 +00:00
For OpenAI API v1 compatibility, you use the [`create_chat_completion_openai_v1`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_chat_completion_openai_v1) method which will return pydantic models instead of dicts.
### JSON and JSON Schema Mode
To constrain chat responses to only valid JSON or a specific JSON Schema use the `response_format` argument in [`create_chat_completion`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_chat_completion).
2024-01-28 00:37:59 +00:00
#### JSON Mode
2024-02-22 16:31:44 +00:00
The following example will constrain the response to valid JSON strings only.
```python
from llama_cpp import Llama
llm = Llama(model_path="path/to/model.gguf", chat_format="chatml")
llm.create_chat_completion(
messages=[
{
"role": "system",
"content": "You are a helpful assistant that outputs in JSON.",
},
{"role": "user", "content": "Who won the world series in 2020"},
],
response_format={
"type": "json_object",
},
temperature=0.7,
)
```
2024-01-28 00:37:59 +00:00
#### JSON Schema Mode
2024-02-22 16:31:44 +00:00
To constrain the response further to a specific JSON Schema add the schema to the `schema` property of the `response_format` argument.
```python
from llama_cpp import Llama
llm = Llama(model_path="path/to/model.gguf", chat_format="chatml")
llm.create_chat_completion(
messages=[
{
"role": "system",
"content": "You are a helpful assistant that outputs in JSON.",
},
{"role": "user", "content": "Who won the world series in 2020"},
],
response_format={
"type": "json_object",
"schema": {
"type": "object",
"properties": {"team_name": {"type": "string"}},
"required": ["team_name"],
},
},
temperature=0.7,
)
```
### Function Calling
The high-level API supports OpenAI compatible function and tool calling. This is possible through the `functionary` pre-trained models chat format or through the generic `chatml-function-calling` chat format.
```python
from llama_cpp import Llama
llm = Llama(model_path="path/to/chatml/llama-model.gguf", chat_format="chatml-function-calling")
llm.create_chat_completion(
2023-11-27 23:29:13 +00:00
messages = [
{
"role": "system",
"content": "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. The assistant calls functions with appropriate input when necessary"
},
2023-11-27 23:29:13 +00:00
{
"role": "user",
"content": "Extract Jason is 25 years old"
}
],
tools=[{
"type": "function",
"function": {
"name": "UserDetail",
"parameters": {
"type": "object",
2023-11-27 23:29:13 +00:00
"title": "UserDetail",
"properties": {
"name": {
"title": "Name",
"type": "string"
},
"age": {
"title": "Age",
"type": "integer"
}
},
2023-11-27 23:29:13 +00:00
"required": [ "name", "age" ]
}
}
}],
tool_choice={
2023-11-27 23:29:13 +00:00
"type": "function",
"function": {
"name": "UserDetail"
}
}
)
```
<details>
<summary>Functionary v2</summary>
The various gguf-converted files for this set of models can be found [here](https://huggingface.co/meetkai). Functionary is able to intelligently call functions and also analyze any provided function outputs to generate coherent responses. All v2 models of functionary supports **parallel function calling**. You can provide either `functionary-v1` or `functionary-v2` for the `chat_format` when initializing the Llama class.
Due to discrepancies between llama.cpp and HuggingFace's tokenizers, it is required to provide HF Tokenizer for functionary. The `LlamaHFTokenizer` class can be initialized and passed into the Llama class. This will override the default llama.cpp tokenizer used in Llama class. The tokenizer files are already included in the respective HF repositories hosting the gguf files.
```python
from llama_cpp import Llama
from llama_cpp.llama_tokenizer import LlamaHFTokenizer
llm = Llama.from_pretrained(
2024-02-23 17:36:13 +00:00
repo_id="meetkai/functionary-small-v2.2-GGUF",
filename="functionary-small-v2.2.q4_0.gguf",
chat_format="functionary-v2",
2024-02-23 17:36:13 +00:00
tokenizer=LlamaHFTokenizer.from_pretrained("meetkai/functionary-small-v2.2-GGUF")
)
```
**NOTE**: There is no need to provide the default system messages used in Functionary as they are added automatically in the Functionary chat handler. Thus, the messages should contain just the chat messages and/or system messages that provide additional context for the model (e.g.: datetime, etc.).
</details>
### Multi-modal Models
`llama-cpp-python` supports such as llava1.5 which allow the language model to read information from both text and images.
You'll first need to download one of the available multi-modal models in GGUF format:
- [llava-v1.5-7b](https://huggingface.co/mys/ggml_llava-v1.5-7b)
- [llava-v1.5-13b](https://huggingface.co/mys/ggml_llava-v1.5-13b)
- [bakllava-1-7b](https://huggingface.co/mys/ggml_bakllava-1)
- [llava-v1.6-34b](https://huggingface.co/cjpais/llava-v1.6-34B-gguf)
- [moondream2](https://huggingface.co/vikhyatk/moondream2)
Then you'll need to use a custom chat handler to load the clip model and process the chat messages and images.
```python
from llama_cpp import Llama
from llama_cpp.llama_chat_format import Llava15ChatHandler
chat_handler = Llava15ChatHandler(clip_model_path="path/to/llava/mmproj.bin")
llm = Llama(
model_path="./path/to/llava/llama-model.gguf",
chat_handler=chat_handler,
n_ctx=2048, # n_ctx should be increased to accommodate the image embedding
)
llm.create_chat_completion(
messages = [
{"role": "system", "content": "You are an assistant who perfectly describes images."},
{
"role": "user",
"content": [
{"type" : "text", "text": "What's in this image?"},
{"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" } }
]
}
]
)
```
You can also pull the model from the Hugging Face Hub using the `from_pretrained` method.
```python
2024-04-30 14:11:38 +00:00
from llama_cpp import Llama
from llama_cpp.llama_chat_format import MoondreamChatHandler
chat_handler = MoondreamChatHandler.from_pretrained(
repo_id="vikhyatk/moondream2",
filename="*mmproj*",
)
2024-04-30 14:11:38 +00:00
llm = Llama.from_pretrained(
repo_id="vikhyatk/moondream2",
filename="*text-model*",
chat_handler=chat_handler,
n_ctx=2048, # n_ctx should be increased to accommodate the image embedding
)
2024-04-30 14:11:38 +00:00
2024-05-10 14:35:51 +00:00
response = llm.create_chat_completion(
messages = [
{
"role": "user",
"content": [
{"type" : "text", "text": "What's in this image?"},
{"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" } }
]
}
]
)
2024-04-30 14:11:38 +00:00
print(response["choices"][0]["text"])
```
**Note**: Multi-modal models also support tool calling and JSON mode.
<details>
<summary>Loading a Local Image</summary>
Images can be passed as base64 encoded data URIs. The following example demonstrates how to do this.
```python
import base64
def image_to_base64_data_uri(file_path):
with open(file_path, "rb") as img_file:
base64_data = base64.b64encode(img_file.read()).decode('utf-8')
return f"data:image/png;base64,{base64_data}"
# Replace 'file_path.png' with the actual path to your PNG file
file_path = 'file_path.png'
data_uri = image_to_base64_data_uri(file_path)
messages = [
{"role": "system", "content": "You are an assistant who perfectly describes images."},
{
"role": "user",
"content": [
{"type": "image_url", "image_url": {"url": data_uri }},
{"type" : "text", "text": "Describe this image in detail please."}
]
}
]
```
</details>
### Speculative Decoding
`llama-cpp-python` supports speculative decoding which allows the model to generate completions based on a draft model.
The fastest way to use speculative decoding is through the `LlamaPromptLookupDecoding` class.
Just pass this as a draft model to the `Llama` class during initialization.
```python
from llama_cpp import Llama
from llama_cpp.llama_speculative import LlamaPromptLookupDecoding
llama = Llama(
model_path="path/to/model.gguf",
draft_model=LlamaPromptLookupDecoding(num_pred_tokens=10) # num_pred_tokens is the number of tokens to predict 10 is the default and generally good for gpu, 2 performs better for cpu-only machines.
)
```
2024-02-16 04:15:50 +00:00
### Embeddings
To generate text embeddings use [`create_embedding`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_embedding) or [`embed`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.embed). Note that you must pass `embedding=True` to the constructor upon model creation for these to work properly.
2024-02-16 04:15:50 +00:00
```python
import llama_cpp
llm = llama_cpp.Llama(model_path="path/to/model.gguf", embedding=True)
2024-02-16 04:15:50 +00:00
embeddings = llm.create_embedding("Hello, world!")
2024-02-22 16:31:44 +00:00
# or create multiple embeddings at once
2024-02-16 04:15:50 +00:00
embeddings = llm.create_embedding(["Hello, world!", "Goodbye, world!"])
```
There are two primary notions of embeddings in a Transformer-style model: *token level* and *sequence level*. Sequence level embeddings are produced by "pooling" token level embeddings together, usually by averaging them or using the first token.
Models that are explicitly geared towards embeddings will usually return sequence level embeddings by default, one for each input string. Non-embedding models such as those designed for text generation will typically return only token level embeddings, one for each token in each sequence. Thus the dimensionality of the return type will be one higher for token level embeddings.
It is possible to control pooling behavior in some cases using the `pooling_type` flag on model creation. You can ensure token level embeddings from any model using `LLAMA_POOLING_TYPE_NONE`. The reverse, getting a generation oriented model to yield sequence level embeddings is currently not possible, but you can always do the pooling manually.
### Adjusting the Context Window
2023-11-22 23:09:31 +00:00
The context window of the Llama models determines the maximum number of tokens that can be processed at once. By default, this is set to 512 tokens, but can be adjusted based on your requirements.
For instance, if you want to work with larger contexts, you can expand the context window by setting the n_ctx parameter when initializing the Llama object:
```python
llm = Llama(model_path="./models/7B/llama-model.gguf", n_ctx=2048)
```
2023-11-28 07:37:34 +00:00
## OpenAI Compatible Web Server
2023-04-05 21:44:25 +00:00
`llama-cpp-python` offers a web server which aims to act as a drop-in replacement for the OpenAI API.
This allows you to use llama.cpp compatible models with any OpenAI compatible client (language libraries, services, etc).
To install the server package and get started:
```bash
pip install 'llama-cpp-python[server]'
python3 -m llama_cpp.server --model models/7B/llama-model.gguf
```
2023-11-21 05:24:22 +00:00
Similar to Hardware Acceleration section above, you can also install with GPU (cuBLAS) support like this:
```bash
2024-04-04 06:49:19 +00:00
CMAKE_ARGS="-DLLAMA_CUDA=on" FORCE_CMAKE=1 pip install 'llama-cpp-python[server]'
python3 -m llama_cpp.server --model models/7B/llama-model.gguf --n_gpu_layers 35
```
2023-04-05 21:44:25 +00:00
Navigate to [http://localhost:8000/docs](http://localhost:8000/docs) to see the OpenAPI documentation.
2023-11-21 05:24:22 +00:00
To bind to `0.0.0.0` to enable remote connections, use `python3 -m llama_cpp.server --host 0.0.0.0`.
Similarly, to change the port (default is 8000), use `--port`.
You probably also want to set the prompt format. For chatml, use
```bash
python3 -m llama_cpp.server --model models/7B/llama-model.gguf --chat_format chatml
```
That will format the prompt according to how model expects it. You can find the prompt format in the model card.
For possible options, see [llama_cpp/llama_chat_format.py](llama_cpp/llama_chat_format.py) and look for lines starting with "@register_chat_format".
If you have `huggingface-hub` installed, you can also use the `--hf_model_repo_id` flag to load a model from the Hugging Face Hub.
```bash
python3 -m llama_cpp.server --hf_model_repo_id Qwen/Qwen1.5-0.5B-Chat-GGUF --model '*q8_0.gguf'
```
2023-11-28 07:37:34 +00:00
### Web Server Features
- [Local Copilot replacement](https://llama-cpp-python.readthedocs.io/en/latest/server/#code-completion)
- [Function Calling support](https://llama-cpp-python.readthedocs.io/en/latest/server/#function-calling)
- [Vision API support](https://llama-cpp-python.readthedocs.io/en/latest/server/#multimodal-models)
2023-12-22 19:40:13 +00:00
- [Multiple Models](https://llama-cpp-python.readthedocs.io/en/latest/server/#configuration-and-multi-model-support)
2023-04-12 09:53:39 +00:00
## Docker image
A Docker image is available on [GHCR](https://ghcr.io/abetlen/llama-cpp-python). To run the server:
```bash
docker run --rm -it -p 8000:8000 -v /path/to/models:/models -e MODEL=/models/llama-model.gguf ghcr.io/abetlen/llama-cpp-python:latest
2023-04-12 09:53:39 +00:00
```
2024-01-30 17:23:07 +00:00
[Docker on termux (requires root)](https://gist.github.com/FreddieOliveira/efe850df7ff3951cb62d74bd770dce27) is currently the only known way to run this on phones, see [termux support issue](https://github.com/abetlen/llama-cpp-python/issues/389)
2023-04-12 09:53:39 +00:00
2023-04-05 21:44:25 +00:00
## Low-level API
[API Reference](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#low-level-api)
2023-05-07 05:41:19 +00:00
The low-level API is a direct [`ctypes`](https://docs.python.org/3/library/ctypes.html) binding to the C API provided by `llama.cpp`.
The entire low-level API can be found in [llama_cpp/llama_cpp.py](https://github.com/abetlen/llama-cpp-python/blob/master/llama_cpp/llama_cpp.py) and directly mirrors the C API in [llama.h](https://github.com/ggerganov/llama.cpp/blob/master/llama.h).
2023-05-07 05:41:19 +00:00
Below is a short example demonstrating how to use the low-level API to tokenize a prompt:
```python
import llama_cpp
import ctypes
llama_cpp.llama_backend_init(False) # Must be called once at the start of each program
params = llama_cpp.llama_context_default_params()
2023-05-07 05:41:19 +00:00
# use bytes for char * params
model = llama_cpp.llama_load_model_from_file(b"./models/7b/llama-model.gguf", params)
ctx = llama_cpp.llama_new_context_with_model(model, params)
max_tokens = params.n_ctx
2023-05-07 05:41:19 +00:00
# use ctypes arrays for array params
tokens = (llama_cpp.llama_token * int(max_tokens))()
n_tokens = llama_cpp.llama_tokenize(ctx, b"Q: Name the planets in the solar system? A: ", tokens, max_tokens, llama_cpp.c_bool(True))
llama_cpp.llama_free(ctx)
2023-05-07 05:41:19 +00:00
```
Check out the [examples folder](examples/low_level_api) for more examples of using the low-level API.
2023-04-05 21:44:25 +00:00
2023-11-28 07:37:34 +00:00
## Documentation
2023-04-03 01:03:39 +00:00
2023-11-27 23:21:00 +00:00
Documentation is available via [https://llama-cpp-python.readthedocs.io/](https://llama-cpp-python.readthedocs.io/).
2023-04-03 01:03:39 +00:00
If you find any issues with the documentation, please open an issue or submit a PR.
2023-11-28 07:37:34 +00:00
## Development
2023-04-03 01:03:39 +00:00
This package is under active development and I welcome any contributions.
To get started, clone the repository and install the package in editable / development mode:
2023-04-03 01:03:39 +00:00
```bash
git clone --recurse-submodules https://github.com/abetlen/llama-cpp-python.git
cd llama-cpp-python
# Upgrade pip (required for editable mode)
pip install --upgrade pip
# Install with pip
pip install -e .
# if you want to use the fastapi / openapi server
pip install -e .[server]
# to install all optional dependencies
pip install -e .[all]
# to clear the local build cache
make clean
2023-04-03 01:03:39 +00:00
```
2024-01-25 15:51:48 +00:00
You can also test out specific commits of `lama.cpp` by checking out the desired commit in the `vendor/llama.cpp` submodule and then running `make clean` and `pip install -e .` again. Any changes in the `llama.h` API will require
changes to the `llama_cpp/llama_cpp.py` file to match the new API (additional changes may be required elsewhere).
2023-11-28 07:37:34 +00:00
## FAQ
### Are there pre-built binaries / binary wheels available?
The recommended installation method is to install from source as described above.
The reason for this is that `llama.cpp` is built with compiler optimizations that are specific to your system.
Using pre-built binaries would require disabling these optimizations or supporting a large number of pre-built binaries for each platform.
That being said there are some pre-built binaries available through the Releases as well as some community provided wheels.
In the future, I would like to provide pre-built binaries and wheels for common platforms and I'm happy to accept any useful contributions in this area.
2023-11-28 08:15:01 +00:00
This is currently being tracked in [#741](https://github.com/abetlen/llama-cpp-python/issues/741)
2023-11-28 07:37:34 +00:00
### How does this compare to other Python bindings of `llama.cpp`?
2023-04-03 01:03:39 +00:00
2023-04-04 14:57:22 +00:00
I originally wrote this package for my own use with two goals in mind:
2023-04-03 01:03:39 +00:00
- Provide a simple process to install `llama.cpp` and access the full C API in `llama.h` from Python
- Provide a high-level Python API that can be used as a drop-in replacement for the OpenAI API so existing apps can be easily ported to use `llama.cpp`
Any contributions and changes to this package will be made with these goals in mind.
2023-11-28 07:37:34 +00:00
## License
2023-03-24 04:06:24 +00:00
This project is licensed under the terms of the MIT license.