MaziyarPanahi commited on
Commit
100e2e7
1 Parent(s): ffa5bf3

18cf6dc8de518b3a132b763a815a6dda135572764d778f0376f9143971d55f61

Browse files
Files changed (1) hide show
  1. README.md +226 -0
README.md ADDED
@@ -0,0 +1,226 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - quantized
4
+ - 2-bit
5
+ - 3-bit
6
+ - 4-bit
7
+ - 5-bit
8
+ - 6-bit
9
+ - 8-bit
10
+ - GGUF
11
+ - transformers
12
+ - safetensors
13
+ - mistral
14
+ - text-generation
15
+ - arxiv:2304.12244
16
+ - arxiv:2306.08568
17
+ - arxiv:2308.09583
18
+ - license:apache-2.0
19
+ - autotrain_compatible
20
+ - endpoints_compatible
21
+ - text-generation-inference
22
+ - region:us
23
+ - text-generation
24
+ model_name: WizardLM-2-7B-GGUF
25
+ base_model: microsoft/WizardLM-2-7B
26
+ inference: false
27
+ model_creator: microsoft
28
+ pipeline_tag: text-generation
29
+ quantized_by: MaziyarPanahi
30
+ ---
31
+ # [MaziyarPanahi/WizardLM-2-7B-GGUF](https://huggingface.co/MaziyarPanahi/WizardLM-2-7B-GGUF)
32
+ - Model creator: [microsoft](https://huggingface.co/microsoft)
33
+ - Original model: [microsoft/WizardLM-2-7B](https://huggingface.co/microsoft/WizardLM-2-7B)
34
+
35
+ ## Description
36
+ [MaziyarPanahi/WizardLM-2-7B-GGUF](https://huggingface.co/MaziyarPanahi/WizardLM-2-7B-GGUF) contains GGUF format model files for [microsoft/WizardLM-2-7B](https://huggingface.co/microsoft/WizardLM-2-7B).
37
+
38
+ ## How to use
39
+ Thanks to [TheBloke](https://huggingface.co/TheBloke) for preparing an amazing README on how to use GGUF models:
40
+
41
+ ### About GGUF
42
+
43
+ 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.
44
+
45
+ Here is an incomplete list of clients and libraries that are known to support GGUF:
46
+
47
+ * [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
48
+ * [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
49
+ * [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
50
+ * [GPT4All](https://gpt4all.io/index.html), a free and open source local running GUI, supporting Windows, Linux and macOS with full GPU accel.
51
+ * [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration. Linux available, in beta as of 27/11/2023.
52
+ * [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with many interesting and unique features, including a full model library for easy model selection.
53
+ * [Faraday.dev](https://faraday.dev/), an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
54
+ * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
55
+ * [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
56
+ * [ctransformers](https://github.com/marella/ctransformers), a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server. Note, as of time of writing (November 27th 2023), ctransformers has not been updated in a long time and does not support many recent models.
57
+
58
+ ### Explanation of quantisation methods
59
+
60
+ <details>
61
+ <summary>Click to see details</summary>
62
+
63
+ The new methods available are:
64
+
65
+ * 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)
66
+ * 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.
67
+ * 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.
68
+ * GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
69
+ * 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
70
+
71
+ ## How to download GGUF files
72
+
73
+ **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.
74
+
75
+ The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
76
+
77
+ * LM Studio
78
+ * LoLLMS Web UI
79
+ * Faraday.dev
80
+
81
+ ### In `text-generation-webui`
82
+
83
+ Under Download Model, you can enter the model repo: [MaziyarPanahi/WizardLM-2-7B-GGUF](https://huggingface.co/MaziyarPanahi/WizardLM-2-7B-GGUF) and below it, a specific filename to download, such as: WizardLM-2-7B-GGUF.Q4_K_M.gguf.
84
+
85
+ Then click Download.
86
+
87
+ ### On the command line, including multiple files at once
88
+
89
+ I recommend using the `huggingface-hub` Python library:
90
+
91
+ ```shell
92
+ pip3 install huggingface-hub
93
+ ```
94
+
95
+ Then you can download any individual model file to the current directory, at high speed, with a command like this:
96
+
97
+ ```shell
98
+ huggingface-cli download MaziyarPanahi/WizardLM-2-7B-GGUF WizardLM-2-7B.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
99
+ ```
100
+ </details>
101
+ <details>
102
+ <summary>More advanced huggingface-cli download usage (click to read)</summary>
103
+
104
+ You can also download multiple files at once with a pattern:
105
+
106
+ ```shell
107
+ huggingface-cli download [MaziyarPanahi/WizardLM-2-7B-GGUF](https://huggingface.co/MaziyarPanahi/WizardLM-2-7B-GGUF) --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
108
+ ```
109
+
110
+ For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
111
+
112
+ To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
113
+
114
+ ```shell
115
+ pip3 install hf_transfer
116
+ ```
117
+
118
+ And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
119
+
120
+ ```shell
121
+ HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download MaziyarPanahi/WizardLM-2-7B-GGUF WizardLM-2-7B.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
122
+ ```
123
+
124
+ Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
125
+ </details>
126
+
127
+ ## Example `llama.cpp` command
128
+
129
+ Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
130
+
131
+ ```shell
132
+ ./main -ngl 35 -m WizardLM-2-7B.Q4_K_M.gguf --color -c 32768 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system
133
+ {system_message}<|im_end|>
134
+ <|im_start|>user
135
+ {prompt}<|im_end|>
136
+ <|im_start|>assistant"
137
+ ```
138
+
139
+ Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
140
+
141
+ Change `-c 32768` 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.
142
+
143
+ If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
144
+
145
+ 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)
146
+
147
+ ## How to run in `text-generation-webui`
148
+
149
+ Further instructions can be found in the text-generation-webui documentation, here: [text-generation-webui/docs/04 ‐ Model Tab.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/04%20-%20Model%20Tab.md#llamacpp).
150
+
151
+ ## How to run from Python code
152
+
153
+ You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/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.
154
+
155
+ ### How to load this model in Python code, using llama-cpp-python
156
+
157
+ For full documentation, please see: [llama-cpp-python docs](https://github.com/abetlen/llama-cpp-python/).
158
+
159
+ #### First install the package
160
+
161
+ Run one of the following commands, according to your system:
162
+
163
+ ```shell
164
+ # Base ctransformers with no GPU acceleration
165
+ pip install llama-cpp-python
166
+ # With NVidia CUDA acceleration
167
+ CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
168
+ # Or with OpenBLAS acceleration
169
+ CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
170
+ # Or with CLBLast acceleration
171
+ CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
172
+ # Or with AMD ROCm GPU acceleration (Linux only)
173
+ CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
174
+ # Or with Metal GPU acceleration for macOS systems only
175
+ CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
176
+
177
+ # In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
178
+ $env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
179
+ pip install llama-cpp-python
180
+ ```
181
+
182
+ #### Simple llama-cpp-python example code
183
+
184
+ ```python
185
+ from llama_cpp import Llama
186
+
187
+ # Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
188
+ llm = Llama(
189
+ model_path="./WizardLM-2-7B.Q4_K_M.gguf", # Download the model file first
190
+ n_ctx=32768, # The max sequence length to use - note that longer sequence lengths require much more resources
191
+ n_threads=8, # The number of CPU threads to use, tailor to your system and the resulting performance
192
+ n_gpu_layers=35 # The number of layers to offload to GPU, if you have GPU acceleration available
193
+ )
194
+
195
+ # Simple inference example
196
+ output = llm(
197
+ "<|im_start|>system
198
+ {system_message}<|im_end|>
199
+ <|im_start|>user
200
+ {prompt}<|im_end|>
201
+ <|im_start|>assistant", # Prompt
202
+ max_tokens=512, # Generate up to 512 tokens
203
+ stop=["</s>"], # Example stop token - not necessarily correct for this specific model! Please check before using.
204
+ echo=True # Whether to echo the prompt
205
+ )
206
+
207
+ # Chat Completion API
208
+
209
+ llm = Llama(model_path="./WizardLM-2-7B.Q4_K_M.gguf", chat_format="llama-2") # Set chat_format according to the model you are using
210
+ llm.create_chat_completion(
211
+ messages = [
212
+ {"role": "system", "content": "You are a story writing assistant."},
213
+ {
214
+ "role": "user",
215
+ "content": "Write a story about llamas."
216
+ }
217
+ ]
218
+ )
219
+ ```
220
+
221
+ ## How to use with LangChain
222
+
223
+ Here are guides on using llama-cpp-python and ctransformers with LangChain:
224
+
225
+ * [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
226
+ * [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)