Updated with llama-cpp-python example
Browse files
README.md
CHANGED
@@ -93,7 +93,7 @@ Generated importance matrix file: [Cerebrum-1.0-8x7b.imatrix.dat](https://huggin
|
|
93 |
Make sure you are using `llama.cpp` from commit [0becb22](https://github.com/ggerganov/llama.cpp/commit/0becb22ac05b6542bd9d5f2235691aa1d3d4d307) or later.
|
94 |
|
95 |
```shell
|
96 |
-
./main -ngl 33 -m Cerebrum-1.0-8x7b.IQ2_XS.gguf --override-kv llama.expert_used_count=int:3 --color -c 16384 --temp 0.7 --
|
97 |
```
|
98 |
|
99 |
Change `-ngl 33` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
|
@@ -107,6 +107,68 @@ There is a similar option for V-cache (`-ctv`), however that is [not working yet
|
|
107 |
|
108 |
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
|
109 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
<!-- README_GGUF.md-how-to-run end -->
|
111 |
|
112 |
<!-- original-model-card start -->
|
|
|
93 |
Make sure you are using `llama.cpp` from commit [0becb22](https://github.com/ggerganov/llama.cpp/commit/0becb22ac05b6542bd9d5f2235691aa1d3d4d307) or later.
|
94 |
|
95 |
```shell
|
96 |
+
./main -ngl 33 -m Cerebrum-1.0-8x7b.IQ2_XS.gguf --override-kv llama.expert_used_count=int:3 --color -c 16384 --temp 0.7 --repeat-penalty 1.0 -n -1 -p "<s>A chat between a user and a thinking artificial intelligence assistant. The assistant describes its thought process and gives helpful and detailed answers to the user's questions.\nUser: {prompt}\nAI:"
|
97 |
```
|
98 |
|
99 |
Change `-ngl 33` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
|
|
|
107 |
|
108 |
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
|
109 |
|
110 |
+
## How to run from Python code
|
111 |
+
|
112 |
+
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) module.
|
113 |
+
|
114 |
+
### How to load this model in Python code, using llama-cpp-python
|
115 |
+
|
116 |
+
For full documentation, please see: [llama-cpp-python docs](https://llama-cpp-python.readthedocs.io/en/latest/).
|
117 |
+
|
118 |
+
#### First install the package
|
119 |
+
|
120 |
+
Run one of the following commands, according to your system:
|
121 |
+
|
122 |
+
```shell
|
123 |
+
# Prebuilt wheel with basic CPU support
|
124 |
+
pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
|
125 |
+
# Prebuilt wheel with NVidia CUDA acceleration
|
126 |
+
pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121 (or cu122 etc.)
|
127 |
+
# Prebuilt wheel with Metal GPU acceleration
|
128 |
+
pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/metal
|
129 |
+
# Build base version with no GPU acceleration
|
130 |
+
pip install llama-cpp-python
|
131 |
+
# With NVidia CUDA acceleration
|
132 |
+
CMAKE_ARGS="-DLLAMA_CUDA=on" pip install llama-cpp-python
|
133 |
+
# Or with OpenBLAS acceleration
|
134 |
+
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
|
135 |
+
# Or with CLBLast acceleration
|
136 |
+
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
|
137 |
+
# Or with AMD ROCm GPU acceleration (Linux only)
|
138 |
+
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
|
139 |
+
# Or with Metal GPU acceleration for macOS systems only
|
140 |
+
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
|
141 |
+
# Or with Vulkan acceleration
|
142 |
+
CMAKE_ARGS="-DLLAMA_VULKAN=on" pip install llama-cpp-python
|
143 |
+
# Or with Kompute acceleration
|
144 |
+
CMAKE_ARGS="-DLLAMA_KOMPUTE=on" pip install llama-cpp-python
|
145 |
+
# Or with SYCL acceleration
|
146 |
+
CMAKE_ARGS="-DLLAMA_SYCL=on -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx" pip install llama-cpp-python
|
147 |
+
|
148 |
+
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
|
149 |
+
$env:CMAKE_ARGS = "-DLLAMA_CUDA=on"
|
150 |
+
pip install llama-cpp-python
|
151 |
+
```
|
152 |
+
|
153 |
+
#### Simple llama-cpp-python example code
|
154 |
+
|
155 |
+
```python
|
156 |
+
from llama_cpp import Llama
|
157 |
+
|
158 |
+
# Chat Completion API
|
159 |
+
|
160 |
+
llm = Llama(model_path="./Cerebrum-1.0-8x7b.IQ3_M.gguf", n_gpu_layers=33, n_ctx=16384)
|
161 |
+
print(llm.create_chat_completion(
|
162 |
+
messages = [
|
163 |
+
{"role": "system", "content": "You are a story writing assistant."},
|
164 |
+
{
|
165 |
+
"role": "user",
|
166 |
+
"content": "Write a story about llamas."
|
167 |
+
}
|
168 |
+
]
|
169 |
+
))
|
170 |
+
```
|
171 |
+
|
172 |
<!-- README_GGUF.md-how-to-run end -->
|
173 |
|
174 |
<!-- original-model-card start -->
|