TheBloke commited on
Commit
84cb7c4
1 Parent(s): da2279d

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +399 -0
README.md ADDED
@@ -0,0 +1,399 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: berkeley-nest/Starling-LM-7B-alpha
3
+ datasets:
4
+ - berkeley-nest/Nectar
5
+ inference: false
6
+ language:
7
+ - en
8
+ library_name: transformers
9
+ license: cc-by-nc-4.0
10
+ model_creator: Berkeley-Nest
11
+ model_name: Starling LM 7B Alpha
12
+ model_type: mistral
13
+ prompt_template: 'GPT4 User: {prompt}<|end_of_turn|>GPT4 Assistant:
14
+
15
+ '
16
+ quantized_by: TheBloke
17
+ tags:
18
+ - reward model
19
+ - RLHF
20
+ - RLAIF
21
+ ---
22
+ <!-- markdownlint-disable MD041 -->
23
+
24
+ <!-- header start -->
25
+ <!-- 200823 -->
26
+ <div style="width: auto; margin-left: auto; margin-right: auto">
27
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
28
+ </div>
29
+ <div style="display: flex; justify-content: space-between; width: 100%;">
30
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
31
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
32
+ </div>
33
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
34
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
35
+ </div>
36
+ </div>
37
+ <div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
38
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
39
+ <!-- header end -->
40
+
41
+ # Starling LM 7B Alpha - GGUF
42
+ - Model creator: [Berkeley-Nest](https://huggingface.co/berkeley-nest)
43
+ - Original model: [Starling LM 7B Alpha](https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha)
44
+
45
+ <!-- description start -->
46
+ ## Description
47
+
48
+ This repo contains GGUF format model files for [Berkeley-Nest's Starling LM 7B Alpha](https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha).
49
+
50
+ These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
51
+
52
+ <!-- description end -->
53
+ <!-- README_GGUF.md-about-gguf start -->
54
+ ### About GGUF
55
+
56
+ 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.
57
+
58
+ Here is an incomplete list of clients and libraries that are known to support GGUF:
59
+
60
+ * [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
61
+ * [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.
62
+ * [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.
63
+ * [GPT4All](https://gpt4all.io/index.html), a free and open source local running GUI, supporting Windows, Linux and macOS with full GPU accel.
64
+ * [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.
65
+ * [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.
66
+ * [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.
67
+ * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
68
+ * [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
69
+ * [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.
70
+
71
+ <!-- README_GGUF.md-about-gguf end -->
72
+ <!-- repositories-available start -->
73
+ ## Repositories available
74
+
75
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-AWQ)
76
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GPTQ)
77
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF)
78
+ * [Berkeley-Nest's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha)
79
+ <!-- repositories-available end -->
80
+
81
+ <!-- prompt-template start -->
82
+ ## Prompt template: OpenChat
83
+
84
+ ```
85
+ GPT4 User: {prompt}<|end_of_turn|>GPT4 Assistant:
86
+
87
+ ```
88
+
89
+ <!-- prompt-template end -->
90
+
91
+
92
+ <!-- compatibility_gguf start -->
93
+ ## Compatibility
94
+
95
+ These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
96
+
97
+ They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
98
+
99
+ ## Explanation of quantisation methods
100
+
101
+ <details>
102
+ <summary>Click to see details</summary>
103
+
104
+ The new methods available are:
105
+
106
+ * 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)
107
+ * 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.
108
+ * 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.
109
+ * GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
110
+ * 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
111
+
112
+ Refer to the Provided Files table below to see what files use which methods, and how.
113
+ </details>
114
+ <!-- compatibility_gguf end -->
115
+
116
+ <!-- README_GGUF.md-provided-files start -->
117
+ ## Provided files
118
+
119
+ | Name | Quant method | Bits | Size | Max RAM required | Use case |
120
+ | ---- | ---- | ---- | ---- | ---- | ----- |
121
+ | [starling-lm-7b-alpha.Q2_K.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q2_K.gguf) | Q2_K | 2 | 3.08 GB| 5.58 GB | smallest, significant quality loss - not recommended for most purposes |
122
+ | [starling-lm-7b-alpha.Q3_K_S.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q3_K_S.gguf) | Q3_K_S | 3 | 3.17 GB| 5.67 GB | very small, high quality loss |
123
+ | [starling-lm-7b-alpha.Q3_K_M.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q3_K_M.gguf) | Q3_K_M | 3 | 3.52 GB| 6.02 GB | very small, high quality loss |
124
+ | [starling-lm-7b-alpha.Q3_K_L.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q3_K_L.gguf) | Q3_K_L | 3 | 3.82 GB| 6.32 GB | small, substantial quality loss |
125
+ | [starling-lm-7b-alpha.Q4_0.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q4_0.gguf) | Q4_0 | 4 | 4.11 GB| 6.61 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
126
+ | [starling-lm-7b-alpha.Q4_K_S.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q4_K_S.gguf) | Q4_K_S | 4 | 4.14 GB| 6.64 GB | small, greater quality loss |
127
+ | [starling-lm-7b-alpha.Q4_K_M.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q4_K_M.gguf) | Q4_K_M | 4 | 4.37 GB| 6.87 GB | medium, balanced quality - recommended |
128
+ | [starling-lm-7b-alpha.Q5_0.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q5_0.gguf) | Q5_0 | 5 | 5.00 GB| 7.50 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
129
+ | [starling-lm-7b-alpha.Q5_K_S.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q5_K_S.gguf) | Q5_K_S | 5 | 5.00 GB| 7.50 GB | large, low quality loss - recommended |
130
+ | [starling-lm-7b-alpha.Q5_K_M.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q5_K_M.gguf) | Q5_K_M | 5 | 5.13 GB| 7.63 GB | large, very low quality loss - recommended |
131
+ | [starling-lm-7b-alpha.Q6_K.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q6_K.gguf) | Q6_K | 6 | 5.94 GB| 8.44 GB | very large, extremely low quality loss |
132
+ | [starling-lm-7b-alpha.Q8_0.gguf](https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/blob/main/starling-lm-7b-alpha.Q8_0.gguf) | Q8_0 | 8 | 7.70 GB| 10.20 GB | very large, extremely low quality loss - not recommended |
133
+
134
+ **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.
135
+
136
+
137
+
138
+ <!-- README_GGUF.md-provided-files end -->
139
+
140
+ <!-- README_GGUF.md-how-to-download start -->
141
+ ## How to download GGUF files
142
+
143
+ **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.
144
+
145
+ The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
146
+
147
+ * LM Studio
148
+ * LoLLMS Web UI
149
+ * Faraday.dev
150
+
151
+ ### In `text-generation-webui`
152
+
153
+ Under Download Model, you can enter the model repo: TheBloke/Starling-LM-7B-alpha-GGUF and below it, a specific filename to download, such as: starling-lm-7b-alpha.Q4_K_M.gguf.
154
+
155
+ Then click Download.
156
+
157
+ ### On the command line, including multiple files at once
158
+
159
+ I recommend using the `huggingface-hub` Python library:
160
+
161
+ ```shell
162
+ pip3 install huggingface-hub
163
+ ```
164
+
165
+ Then you can download any individual model file to the current directory, at high speed, with a command like this:
166
+
167
+ ```shell
168
+ huggingface-cli download TheBloke/Starling-LM-7B-alpha-GGUF starling-lm-7b-alpha.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
169
+ ```
170
+
171
+ <details>
172
+ <summary>More advanced huggingface-cli download usage (click to read)</summary>
173
+
174
+ You can also download multiple files at once with a pattern:
175
+
176
+ ```shell
177
+ huggingface-cli download TheBloke/Starling-LM-7B-alpha-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
178
+ ```
179
+
180
+ 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).
181
+
182
+ To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
183
+
184
+ ```shell
185
+ pip3 install hf_transfer
186
+ ```
187
+
188
+ And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
189
+
190
+ ```shell
191
+ HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/Starling-LM-7B-alpha-GGUF starling-lm-7b-alpha.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
192
+ ```
193
+
194
+ Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
195
+ </details>
196
+ <!-- README_GGUF.md-how-to-download end -->
197
+
198
+ <!-- README_GGUF.md-how-to-run start -->
199
+ ## Example `llama.cpp` command
200
+
201
+ Make sure you are using `llama.cpp` from commit [d0cee0d](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
202
+
203
+ ```shell
204
+ ./main -ngl 35 -m starling-lm-7b-alpha.Q4_K_M.gguf --color -c 8192 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "GPT4 User: {prompt}<|end_of_turn|>GPT4 Assistant:"
205
+ ```
206
+
207
+ Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
208
+
209
+ 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.
210
+
211
+ If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
212
+
213
+ 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)
214
+
215
+ ## How to run in `text-generation-webui`
216
+
217
+ 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%E2%80%90%20Model%20Tab.md#llamacpp).
218
+
219
+ ## How to run from Python code
220
+
221
+ 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.
222
+
223
+ ### How to load this model in Python code, using llama-cpp-python
224
+
225
+ For full documentation, please see: [llama-cpp-python docs](https://abetlen.github.io/llama-cpp-python/).
226
+
227
+ #### First install the package
228
+
229
+ Run one of the following commands, according to your system:
230
+
231
+ ```shell
232
+ # Base ctransformers with no GPU acceleration
233
+ pip install llama-cpp-python
234
+ # With NVidia CUDA acceleration
235
+ CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
236
+ # Or with OpenBLAS acceleration
237
+ CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
238
+ # Or with CLBLast acceleration
239
+ CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
240
+ # Or with AMD ROCm GPU acceleration (Linux only)
241
+ CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
242
+ # Or with Metal GPU acceleration for macOS systems only
243
+ CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
244
+
245
+ # In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
246
+ $env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
247
+ pip install llama-cpp-python
248
+ ```
249
+
250
+ #### Simple llama-cpp-python example code
251
+
252
+ ```python
253
+ from llama_cpp import Llama
254
+
255
+ # Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
256
+ llm = Llama(
257
+ model_path="./starling-lm-7b-alpha.Q4_K_M.gguf", # Download the model file first
258
+ n_ctx=8192, # The max sequence length to use - note that longer sequence lengths require much more resources
259
+ n_threads=8, # The number of CPU threads to use, tailor to your system and the resulting performance
260
+ n_gpu_layers=35 # The number of layers to offload to GPU, if you have GPU acceleration available
261
+ )
262
+
263
+ # Simple inference example
264
+ output = llm(
265
+ "GPT4 User: {prompt}<|end_of_turn|>GPT4 Assistant:", # Prompt
266
+ max_tokens=512, # Generate up to 512 tokens
267
+ stop=["</s>"], # Example stop token - not necessarily correct for this specific model! Please check before using.
268
+ echo=True # Whether to echo the prompt
269
+ )
270
+
271
+ # Chat Completion API
272
+
273
+ llm = Llama(model_path="./starling-lm-7b-alpha.Q4_K_M.gguf", chat_format="llama-2") # Set chat_format according to the model you are using
274
+ llm.create_chat_completion(
275
+ messages = [
276
+ {"role": "system", "content": "You are a story writing assistant."},
277
+ {
278
+ "role": "user",
279
+ "content": "Write a story about llamas."
280
+ }
281
+ ]
282
+ )
283
+ ```
284
+
285
+ ## How to use with LangChain
286
+
287
+ Here are guides on using llama-cpp-python and ctransformers with LangChain:
288
+
289
+ * [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
290
+ * [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
291
+
292
+ <!-- README_GGUF.md-how-to-run end -->
293
+
294
+ <!-- footer start -->
295
+ <!-- 200823 -->
296
+ ## Discord
297
+
298
+ For further support, and discussions on these models and AI in general, join us at:
299
+
300
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
301
+
302
+ ## Thanks, and how to contribute
303
+
304
+ Thanks to the [chirper.ai](https://chirper.ai) team!
305
+
306
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
307
+
308
+ 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.
309
+
310
+ 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.
311
+
312
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
313
+
314
+ * Patreon: https://patreon.com/TheBlokeAI
315
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
316
+
317
+ **Special thanks to**: Aemon Algiz.
318
+
319
+ **Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
320
+
321
+
322
+ Thank you to all my generous patrons and donaters!
323
+
324
+ And thank you again to a16z for their generous grant.
325
+
326
+ <!-- footer end -->
327
+
328
+ <!-- original-model-card start -->
329
+ # Original model card: Berkeley-Nest's Starling LM 7B Alpha
330
+
331
+ # Starling-RM-7B-alpha
332
+
333
+ <!-- Provide a quick summary of what the model is/does. -->
334
+
335
+ - **Developed by:** Banghua Zhu * , Evan Frick * , Tianhao Wu * , Hanlin Zhu and Jiantao Jiao.
336
+ - **Model type:** Language Model finetuned with RLHF / RLAIF
337
+ - **License:** Non commercial license
338
+ - **Finetuned from model:** [Openchat 3.5](https://huggingface.co/openchat/openchat_3.5) (based on [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1))
339
+
340
+
341
+
342
+ We introduce Starling-7B, an open large language model (LLM) trained by Reinforcement Learning from AI Feedback (RLAIF). The model harnesses the power of our new GPT-4 labeled ranking dataset, [berkeley-nest/Nectar](https://huggingface.co/datasets/berkeley-nest/Nectar), and our new reward training and policy tuning pipeline. Starling-7B-alpha scores 8.09 in MT Bench with GPT-4 as a judge, outperforming every model to date on MT-Bench except for OpenAI's GPT-4 and GPT-4 Turbo. We release the ranking dataset [Nectar](https://huggingface.co/datasets/berkeley-nest/Nectar), the reward model [Starling-RM-7B-alpha](https://huggingface.co/berkeley-nest/Starling-RM-7B-alpha) and the language model [Starling-LM-7B-alpha](https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha) on HuggingFace, and an online demo in LMSYS [Chatbot Arena](https://chat.lmsys.org). Stay tuned for our forthcoming code and paper, which will provide more details on the whole process.
343
+
344
+ Starling-LM-7B-alpha is a language model trained from [Openchat 3.5](https://huggingface.co/openchat/openchat_3.5) with reward model [berkeley-nest/Starling-RM-7B-alpha](https://huggingface.co/berkeley-nest/Starling-RM-7B-alpha) and policy optimization method [advantage-induced policy alignment (APA)](https://arxiv.org/abs/2306.02231). The evaluation results are listed below.
345
+
346
+
347
+ | Model | Tuning Method | MT Bench | AlpacaEval | MMLU |
348
+ |-----------------------|------------------|----------|------------|------|
349
+ | GPT-4-Turbo | ? | 9.32 | 97.70 | |
350
+ | GPT-4 | SFT + PPO | 8.99 | 95.28 | 86.4 |
351
+ | **Starling-7B** | C-RLFT + APA | 8.09 | 91.99 | 63.9 |
352
+ | Claude-2 | ? | 8.06 | 91.36 | 78.5 |
353
+ | GPT-3.5-Turbo | ? | 7.94 | 89.37 | 70 |
354
+ | Claude-1 | ? | 7.9 | 88.39 | 77 |
355
+ | Tulu-2-dpo-70b | SFT + DPO | 7.89 | 95.1 | |
356
+ | Openchat-3.5 | C-RLFT | 7.81 | 88.51 | 64.3 |
357
+ | Zephyr-7B-beta | SFT + DPO | 7.34 | 90.60 | 61.4 |
358
+ | Llama-2-70b-chat-hf | SFT + PPO | 6.86 | 92.66 | 63 |
359
+ | Neural-chat-7b-v3-1 | SFT + DPO | 6.84 | 84.53 | 62.4 |
360
+ | Tulu-2-dpo-7b | SFT + DPO | 6.29 | 85.1 | |
361
+
362
+
363
+
364
+ For more detailed discussions, please check out our [blog post](https://starling.cs.berkeley.edu), and stay tuned for our upcoming code and paper!
365
+ <!-- Provide the basic links for the model. -->
366
+
367
+ - **Blog:** https://starling.cs.berkeley.edu/
368
+ - **Paper:** Coming soon!
369
+ - **Code:** Coming soon!
370
+
371
+
372
+
373
+ ## Uses
374
+
375
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
376
+ Our model follows the exact chat template and usage as [Openchat 3.5](https://huggingface.co/openchat/openchat_3.5). Please refer to their model card for more details.
377
+ In addition, our model is hosted on LMSYS [Chatbot Arena](https://chat.lmsys.org) for free test.
378
+
379
+
380
+
381
+ ## License
382
+ The dataset, model and online demo is a research preview intended for non-commercial use only, subject to the data distillation [License](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) of LLaMA, [Terms of Use](https://openai.com/policies/terms-of-use) of the data generated by OpenAI, and [Privacy Practices](https://chrome.google.com/webstore/detail/sharegpt-share-your-chatg/daiacboceoaocpibfodeljbdfacokfjb) of ShareGPT. Please contact us if you find any potential violation.
383
+
384
+
385
+ ## Acknowledgment
386
+ We would like to thank Wei-Lin Chiang from Berkeley for detailed feedback of the blog and the projects. We would like to thank the [LMSYS Organization](https://lmsys.org/) for their support of [lmsys-chat-1M](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) dataset, evaluation and online demo. We would like to thank the open source community for their efforts in providing the datasets and base models we used to develope the project, including but not limited to Anthropic, Llama, Mistral, Hugging Face H4, LMSYS, OpenChat, OpenBMB, Flan and ShareGPT.
387
+
388
+ ## Citation
389
+ ```
390
+ @misc{starling2023,
391
+ title = {Starling-7B: Improving LLM Helpfulness & Harmlessness with RLAIF},
392
+ url = {},
393
+ author = {Zhu, Banghua and Frick, Evan and Wu, Tianhao and Zhu, Hanlin and Jiao, Jiantao},
394
+ month = {November},
395
+ year = {2023}
396
+ }
397
+ ```
398
+
399
+ <!-- original-model-card end -->