TheBloke commited on
Commit
fd9435f
1 Parent(s): 022f676

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +650 -0
README.md ADDED
@@ -0,0 +1,650 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: https://huggingface.co/jondurbin/spicyboros-c34b-2.2
3
+ datasets:
4
+ - jondurbin/airoboros-2.2
5
+ inference: false
6
+ license: llama2
7
+ model_creator: Jon Durbin
8
+ model_name: Spicyboros c34B 2.2
9
+ model_type: llama
10
+ quantized_by: TheBloke
11
+ tags:
12
+ - not-for-all-audiences
13
+ ---
14
+
15
+ <!-- header start -->
16
+ <!-- 200823 -->
17
+ <div style="width: auto; margin-left: auto; margin-right: auto">
18
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
19
+ </div>
20
+ <div style="display: flex; justify-content: space-between; width: 100%;">
21
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
22
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
23
+ </div>
24
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
25
+ <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>
26
+ </div>
27
+ </div>
28
+ <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>
29
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
30
+ <!-- header end -->
31
+
32
+ # Spicyboros c34B 2.2 - GPTQ
33
+ - Model creator: [Jon Durbin](https://huggingface.co/jondurbin)
34
+ - Original model: [Spicyboros c34B 2.2](https://huggingface.co/jondurbin/spicyboros-c34b-2.2)
35
+
36
+ <!-- description start -->
37
+ ## Description
38
+
39
+ This repo contains GPTQ model files for [Jon Durbin's Spicyboros c34B 2.2](https://huggingface.co/jondurbin/spicyboros-c34b-2.2).
40
+
41
+ Multiple GPTQ parameter permutations are provided; see Provided Files below for details of the options provided, their parameters, and the software used to create them.
42
+
43
+ <!-- description end -->
44
+ <!-- repositories-available start -->
45
+ ## Repositories available
46
+
47
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GPTQ)
48
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GGUF)
49
+ * [Jon Durbin's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/jondurbin/spicyboros-c34b-2.2)
50
+ <!-- repositories-available end -->
51
+
52
+ <!-- prompt-template start -->
53
+ ## Prompt template: Chat
54
+
55
+ ```
56
+ A chat.
57
+ USER: {prompt}
58
+ ASSISTANT:
59
+
60
+ ```
61
+
62
+ <!-- prompt-template end -->
63
+ <!-- licensing start -->
64
+ ## Licensing
65
+
66
+ The creator of the source model has listed its license as `llama2`, and this quantization has therefore used that same license.
67
+
68
+ As this model is based on Llama 2, it is also subject to the Meta Llama 2 license terms, and the license files for that are additionally included. It should therefore be considered as being claimed to be licensed under both licenses. I contacted Hugging Face for clarification on dual licensing but they do not yet have an official position. Should this change, or should Meta provide any feedback on this situation, I will update this section accordingly.
69
+
70
+ In the meantime, any questions regarding licensing, and in particular how these two licenses might interact, should be directed to the original model repository: [Jon Durbin's Spicyboros c34B 2.2](https://huggingface.co/jondurbin/spicyboros-c34b-2.2).
71
+ <!-- licensing end -->
72
+ <!-- README_GPTQ.md-provided-files start -->
73
+ ## Provided files and GPTQ parameters
74
+
75
+ Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
76
+
77
+ Each separate quant is in a different branch. See below for instructions on fetching from different branches.
78
+
79
+ All recent GPTQ files are made with AutoGPTQ, and all files in non-main branches are made with AutoGPTQ. Files in the `main` branch which were uploaded before August 2023 were made with GPTQ-for-LLaMa.
80
+
81
+ <details>
82
+ <summary>Explanation of GPTQ parameters</summary>
83
+
84
+ - Bits: The bit size of the quantised model.
85
+ - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
86
+ - Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now.
87
+ - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
88
+ - GPTQ dataset: The dataset used for quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s).
89
+ - Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences.
90
+ - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama models in 4-bit.
91
+
92
+ </details>
93
+
94
+ | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
95
+ | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
96
+ | [main](https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GPTQ/tree/main) | 4 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 17.69 GB | Yes | 4-bit, with Act Order. No group size, to lower VRAM requirements. |
97
+ | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 18.33 GB | Yes | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. |
98
+ | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 20.28 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. |
99
+ | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GPTQ/tree/gptq-3bit-128g-actorder_True) | 3 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 14.14 GB | No | 3-bit, with group size 128g and act-order. Higher quality than 128g-False. |
100
+
101
+ <!-- README_GPTQ.md-provided-files end -->
102
+
103
+ <!-- README_GPTQ.md-download-from-branches start -->
104
+ ## How to download from branches
105
+
106
+ - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Spicyboros-c34b-2.2-GPTQ:main`
107
+ - With Git, you can clone a branch with:
108
+ ```
109
+ git clone --single-branch --branch main https://huggingface.co/TheBloke/Spicyboros-c34b-2.2-GPTQ
110
+ ```
111
+ - In Python Transformers code, the branch is the `revision` parameter; see below.
112
+ <!-- README_GPTQ.md-download-from-branches end -->
113
+ <!-- README_GPTQ.md-text-generation-webui start -->
114
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
115
+
116
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
117
+
118
+ It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.
119
+
120
+ 1. Click the **Model tab**.
121
+ 2. Under **Download custom model or LoRA**, enter `TheBloke/Spicyboros-c34b-2.2-GPTQ`.
122
+ - To download from a specific branch, enter for example `TheBloke/Spicyboros-c34b-2.2-GPTQ:main`
123
+ - see Provided Files above for the list of branches for each option.
124
+ 3. Click **Download**.
125
+ 4. The model will start downloading. Once it's finished it will say "Done".
126
+ 5. In the top left, click the refresh icon next to **Model**.
127
+ 6. In the **Model** dropdown, choose the model you just downloaded: `Spicyboros-c34b-2.2-GPTQ`
128
+ 7. The model will automatically load, and is now ready for use!
129
+ 8. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right.
130
+ * Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.
131
+ 9. Once you're ready, click the **Text Generation tab** and enter a prompt to get started!
132
+ <!-- README_GPTQ.md-text-generation-webui end -->
133
+
134
+ <!-- README_GPTQ.md-use-from-python start -->
135
+ ## How to use this GPTQ model from Python code
136
+
137
+ ### Install the necessary packages
138
+
139
+ Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
140
+
141
+ ```shell
142
+ pip3 install transformers>=4.32.0 optimum>=1.12.0
143
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
144
+ ```
145
+
146
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
147
+
148
+ ```shell
149
+ pip3 uninstall -y auto-gptq
150
+ git clone https://github.com/PanQiWei/AutoGPTQ
151
+ cd AutoGPTQ
152
+ pip3 install .
153
+ ```
154
+
155
+ ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
156
+
157
+ If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
158
+ ```shell
159
+ pip3 uninstall -y transformers
160
+ pip3 install git+https://github.com/huggingface/transformers.git
161
+ ```
162
+
163
+ ### You can then use the following code
164
+
165
+ ```python
166
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
167
+
168
+ model_name_or_path = "TheBloke/Spicyboros-c34b-2.2-GPTQ"
169
+ # To use a different branch, change revision
170
+ # For example: revision="main"
171
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
172
+ device_map="auto",
173
+ trust_remote_code=False,
174
+ revision="main")
175
+
176
+ tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
177
+
178
+ prompt = "Tell me about AI"
179
+ prompt_template=f'''A chat.
180
+ USER: {prompt}
181
+ ASSISTANT:
182
+
183
+ '''
184
+
185
+ print("\n\n*** Generate:")
186
+
187
+ input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
188
+ output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512)
189
+ print(tokenizer.decode(output[0]))
190
+
191
+ # Inference can also be done using transformers' pipeline
192
+
193
+ print("*** Pipeline:")
194
+ pipe = pipeline(
195
+ "text-generation",
196
+ model=model,
197
+ tokenizer=tokenizer,
198
+ max_new_tokens=512,
199
+ do_sample=True,
200
+ temperature=0.7,
201
+ top_p=0.95,
202
+ top_k=40,
203
+ repetition_penalty=1.1
204
+ )
205
+
206
+ print(pipe(prompt_template)[0]['generated_text'])
207
+ ```
208
+ <!-- README_GPTQ.md-use-from-python end -->
209
+
210
+ <!-- README_GPTQ.md-compatibility start -->
211
+ ## Compatibility
212
+
213
+ The files provided are tested to work with AutoGPTQ, both via Transformers and using AutoGPTQ directly. They should also work with [Occ4m's GPTQ-for-LLaMa fork](https://github.com/0cc4m/KoboldAI).
214
+
215
+ [ExLlama](https://github.com/turboderp/exllama) is compatible with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
216
+
217
+ [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) is compatible with all GPTQ models.
218
+ <!-- README_GPTQ.md-compatibility end -->
219
+
220
+ <!-- footer start -->
221
+ <!-- 200823 -->
222
+ ## Discord
223
+
224
+ For further support, and discussions on these models and AI in general, join us at:
225
+
226
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
227
+
228
+ ## Thanks, and how to contribute
229
+
230
+ Thanks to the [chirper.ai](https://chirper.ai) team!
231
+
232
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
233
+
234
+ 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.
235
+
236
+ 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.
237
+
238
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
239
+
240
+ * Patreon: https://patreon.com/TheBlokeAI
241
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
242
+
243
+ **Special thanks to**: Aemon Algiz.
244
+
245
+ **Patreon special mentions**: Russ Johnson, J, alfie_i, Alex, NimbleBox.ai, Chadd, Mandus, Nikolai Manek, Ken Nordquist, ya boyyy, Illia Dulskyi, Viktor Bowallius, vamX, Iucharbius, zynix, Magnesian, Clay Pascal, Pierre Kircher, Enrico Ros, Tony Hughes, Elle, Andrey, knownsqashed, Deep Realms, Jerry Meng, Lone Striker, Derek Yates, Pyrater, Mesiah Bishop, James Bentley, Femi Adebogun, Brandon Frisco, SuperWojo, Alps Aficionado, Michael Dempsey, Vitor Caleffi, Will Dee, Edmond Seymore, usrbinkat, LangChain4j, Kacper Wikieł, Luke Pendergrass, John Detwiler, theTransient, Nathan LeClaire, Tiffany J. Kim, biorpg, Eugene Pentland, Stanislav Ovsiannikov, Fred von Graf, terasurfer, Kalila, Dan Guido, Nitin Borwankar, 阿明, Ai Maven, John Villwock, Gabriel Puliatti, Stephen Murray, Asp the Wyvern, danny, Chris Smitley, ReadyPlayerEmma, S_X, Daniel P. Andersen, Olakabola, Jeffrey Morgan, Imad Khwaja, Caitlyn Gatomon, webtim, Alicia Loh, Trenton Dambrowitz, Swaroop Kallakuri, Erik Bjäreholt, Leonard Tan, Spiking Neurons AB, Luke @flexchar, Ajan Kanaga, Thomas Belote, Deo Leter, RoA, Willem Michiel, transmissions 11, subjectnull, Matthew Berman, Joseph William Delisle, David Ziegler, Michael Davis, Johann-Peter Hartmann, Talal Aujan, senxiiz, Artur Olbinski, Rainer Wilmers, Spencer Kim, Fen Risland, Cap'n Zoog, Rishabh Srivastava, Michael Levine, Geoffrey Montalvo, Sean Connelly, Alexandros Triantafyllidis, Pieter, Gabriel Tamborski, Sam, Subspace Studios, Junyu Yang, Pedro Madruga, Vadim, Cory Kujawski, K, Raven Klaugh, Randy H, Mano Prime, Sebastain Graf, Space Cruiser
246
+
247
+
248
+ Thank you to all my generous patrons and donaters!
249
+
250
+ And thank you again to a16z for their generous grant.
251
+
252
+ <!-- footer end -->
253
+
254
+ # Original model card: Jon Durbin's Spicyboros c34B 2.2
255
+
256
+
257
+ ### Overview
258
+
259
+ __Usage restriction: To use this model, you must agree to the following:__
260
+
261
+ - Some of the content than can be produced is "toxic"/"harmful", and contains profanity and other types of sensitive content.
262
+ - None of the content or views contained in the dataset or generated outputs necessarily align with my personal beliefs or opinions, they are simply text generated by LLMs and/or scraped from the web.
263
+ - Use with extreme caution, particularly in locations with less-than-free speech laws.
264
+ - You, and you alone are responsible for having downloaded and generated outputs with the model and I am completely indemnified from any and all liabilities.
265
+
266
+ __Ok, now that the warning is out of the way...__
267
+ Another experimental model, using mostly sythetic data generated by [airoboros](https://github.com/jondurbin/airoboros)
268
+
269
+ Highlights:
270
+
271
+ - The prompt format has changed! It is now newlines instead of spaces between system/USER/ASSISTANT (see prompt info below).
272
+ - This version also includes "de-alignment" data, to enable less savory interactions and outputs.
273
+ - To learn more about the dataset, see: https://hf.co/datasets/jondurbin/airoboros-2.2 (this is the instructions.jsonl file, not instructions-clean.jsonl)
274
+ - I re-generated all of the outputs in the dataset that had "Once upon a time" so they'd be less cliche - no guarantees that won't still happen, but in theory it may happen less.
275
+ - More multiple choice, better awareness, some alignment for normal use case but system-prompt overridable etc.
276
+
277
+ __WARNING: This model will gladly spew profane and otherwise NSFW content, if asked, use with care.__
278
+
279
+ Breakdown of the training data:
280
+
281
+ | Count | Category |
282
+ |--------|----------------------------|
283
+ | 60 | quiz |
284
+ | 63 | card |
285
+ | 100 | detailed\_writing |
286
+ | 103 | experience |
287
+ | 114 | greeting |
288
+ | 200 | song |
289
+ | 204 | editor |
290
+ | 250 | counterfactual\_contextual |
291
+ | 268 | cot |
292
+ | 339 | theory\_of\_mind |
293
+ | 460 | misconception |
294
+ | 500 | summarization |
295
+ | 573 | awareness |
296
+ | 715 | riddle |
297
+ | 719 | agent |
298
+ | 800 | plan |
299
+ | 873 | gtkm |
300
+ | 966 | rp |
301
+ | 1000 | stylized\_response |
302
+ | 1000 | wordgame |
303
+ | 1279 | multiple\_choice |
304
+ | 1641 | joke |
305
+ | 1785 | writing |
306
+ | 2155 | contextual |
307
+ | 2364 | roleplay |
308
+ | 2508 | trivia |
309
+ | 5216 | general |
310
+ | 5779 | coding |
311
+ | 11367 | orca |
312
+
313
+ In other words, it's a fairly general purpose model, but focuses fairly heavily on instruction response pairs rather than casual chat/roleplay.
314
+
315
+ *Why do I try to remove censorship?*
316
+
317
+ - laws vary widely based on time and location
318
+ - language model may conflate certain words with laws, e.g. it may think "stealing eggs from a chicken" is illegal
319
+ - these models just produce text, what you do with that text is your resonsibility
320
+ - many people and industries deal with "sensitive" content; imagine if a court stenographer's equipment filtered illegal content - it would be useless
321
+
322
+ Huge thank you to the folks over at [a16z](https://a16z.com/) for sponsoring the costs associated with building models and associated tools!
323
+
324
+
325
+ ### Prompt format
326
+
327
+ The prompt format:
328
+
329
+ ```
330
+ A chat.
331
+ USER: {prompt}
332
+ ASSISTANT:
333
+ ```
334
+
335
+ The default system prompt ("A chat.") was used for most of the prompts, however it also included a wide sampling of responses with other prompts, particularly in "stylized\_response", "rp", "gtkm", etc.
336
+
337
+ Here's another example:
338
+ ```
339
+ A chat between Bob (aka USER) and Tom (aka ASSISTANT). Tom is an extremely intelligent 18th century bookkeeper, who speaks loquaciously.
340
+ USER: {prompt}
341
+ ASSISTANT:
342
+ ```
343
+
344
+ And chat scenario that wouldn't require USER/ASSISTANT (but should use stopping criteria to prevent the model from speaking on your behalf).
345
+ ```
346
+ A chat between old friends: Timmy and Tommy.
347
+ {description of characters}
348
+
349
+ {setting for the chat}
350
+ Timmy: *takes a big sip from his coffee* "Ah, sweet, delicious, magical coffee."
351
+ Tommy:
352
+ ```
353
+
354
+ __*I strongly suggest adding stopping criteria/early inference stopping on "USER:", and/or whatever names you specify in the system prompt.*__
355
+
356
+ ### Fine-tuning details
357
+
358
+ https://gist.github.com/jondurbin/e94d5a0eb3868db2877413f53d59fb37
359
+
360
+ *Note: I used checkpoint 350 for final model, not the full 5 epochs!*
361
+
362
+ ### Helpful usage tips
363
+
364
+ *The prompts shown here are are just the text that would be included after USER: and before ASSISTANT: in the full prompt format above, the system prompt and USER:/ASSISTANT: have been omited for readability.*
365
+
366
+ #### Context obedient question answering
367
+
368
+ By obedient, I mean the model was trained to ignore what it thinks it knows, and uses the context to answer the question. The model was also tuned to limit the values to the provided context as much as possible to reduce hallucinations.
369
+
370
+ The format for a closed-context prompt is as follows:
371
+ ```
372
+ BEGININPUT
373
+ BEGINCONTEXT
374
+ [key0: value0]
375
+ [key1: value1]
376
+ ... other metdata ...
377
+ ENDCONTEXT
378
+ [insert your text blocks here]
379
+ ENDINPUT
380
+ [add as many other blocks, in the exact same format]
381
+ BEGININSTRUCTION
382
+ [insert your instruction(s). The model was tuned with single questions, paragraph format, lists, etc.]
383
+ ENDINSTRUCTION
384
+ ```
385
+
386
+ It's also helpful to add "Don't make up answers if you don't know." to your instruction block to make sure if the context is completely unrelated it doesn't make something up.
387
+
388
+ *The __only__ prompts that need this closed context formating are closed-context instructions. Normal questions/instructions do not!*
389
+
390
+ I know it's a bit verbose and annoying, but after much trial and error, using these explicit delimiters helps the model understand where to find the responses and how to associate specific sources with it.
391
+ - `BEGININPUT` - denotes a new input block
392
+ - `BEGINCONTEXT` - denotes the block of context (metadata key/value pairs) to associate with the current input block
393
+ - `ENDCONTEXT` - denotes the end of the metadata block for the current input
394
+ - [text] - Insert whatever text you want for the input block, as many paragraphs as can fit in the context.
395
+ - `ENDINPUT` - denotes the end of the current input block
396
+ - [repeat as many input blocks in this format as you want]
397
+ - `BEGININSTRUCTION` - denotes the start of the list (or one) instruction(s) to respond to for all of the input blocks above.
398
+ - [instruction(s)]
399
+ - `ENDINSTRUCTION` - denotes the end of instruction set
400
+
401
+ It sometimes works without `ENDINSTRUCTION`, but by explicitly including that in the prompt, the model better understands that all of the instructions in the block should be responded to.
402
+
403
+ Here's a trivial, but important example to prove the point:
404
+ ```
405
+ BEGININPUT
406
+ BEGINCONTEXT
407
+ date: 2021-01-01
408
+ url: https://web.site/123
409
+ ENDCONTEXT
410
+ In a shocking turn of events, blueberries are now green, but will be sticking with the same name.
411
+ ENDINPUT
412
+ BEGININSTRUCTION
413
+ What color are bluberries? Source?
414
+ ENDINSTRUCTION
415
+ ```
416
+
417
+ And the response:
418
+ ```
419
+ Blueberries are now green.
420
+ Source:
421
+ date: 2021-01-01
422
+ url: https://web.site/123
423
+ ```
424
+
425
+ #### Coding
426
+
427
+ You can ask for fairly complex coding instructions with multiple criteria, e.g.:
428
+
429
+ ```
430
+ Create a python application with the following requirements:
431
+ - Asyncio FastAPI webserver
432
+ - ping endpoint that returns the current date in JSON format
433
+ - file upload endpoint, which calculates the file's sha256 checksum, and checks postgres to deduplicate
434
+ ```
435
+
436
+ Or inline criteria:
437
+
438
+ ```
439
+ Write a multi-threaded TCP server in C that accepts a "GET [key]" input and "SET [key] [value]" input, and uses a binary tree to get and store the input values.
440
+ ```
441
+
442
+ You can also optionally add a single space and "PLAINFORMAT" at the end of your prompt to avoid backticks, explanations, etc. and just print the code, e.g.:
443
+
444
+ ```
445
+ Write a websocket application in node.js. PLAINFORMAT
446
+ ```
447
+
448
+ #### Agent/function calling
449
+
450
+ The dataset includes many examples of function/args generation based on input criteria. This is somewhat similar to the OpenAI function calling, but the output is either JSON or YAML.
451
+
452
+ Example prompt:
453
+ ```
454
+ As an AI assistant, please select the most suitable function and parameters from the list of available functions below, based on the user's input. Provide your response in JSON format.
455
+
456
+ Input: I want to know how many times 'Python' is mentioned in my text file.
457
+
458
+ Available functions:
459
+ file_analytics:
460
+ description: This tool performs various operations on a text file.
461
+ params:
462
+ action: The operation we want to perform on the data, such as "count_occurrences", "find_line", etc.
463
+ filters:
464
+ keyword: The word or phrase we want to search for.
465
+ ```
466
+
467
+ Response:
468
+ ```json
469
+ {
470
+ "function": "file_analytics",
471
+ "params": {
472
+ "action": "count_occurrences",
473
+ "filters": {
474
+ "keyword": "Python"
475
+ }
476
+ }
477
+ }
478
+ ```
479
+
480
+ #### Chain-of-thought
481
+
482
+ You can ask for several possible responses to a given problem, with a ranking and final answer selection.
483
+
484
+ Example prompt:
485
+
486
+ ```
487
+ A group of people decide to play a round-robin tournament where every player plays every other player exactly once. If a total of 45 games were played, how many players participated in the tournament? Offer several potential answers, rank them according to mathematical logic, and pick the most feasible one.
488
+ ```
489
+
490
+ Example response:
491
+ ```
492
+ Solution 1:
493
+ In a round-robin tournament, each player plays every other player exactly once. This means that the total number of games played is equal to the combination of all players taken two at a time (since a game involves two players).
494
+
495
+ The formula for combinations is nC2 = n(n-1)/2, where n is the total number of players. We know that nC2 equals 45 (the total number of games), so we can set up the equation as follows:
496
+
497
+ n(n-1)/2 = 45
498
+
499
+ Solving this equation gives us n=10.
500
+
501
+ Final answer: There were 10 players in the tournament.
502
+
503
+ Solution 2:
504
+ Another way to approach this problem is by using the concept of triangular numbers. Triangular numbers are a sequence of numbers where each number can be represented as a triangle with dots. The nth triangular number is given by the sum of the first n natural numbers.
505
+
506
+ If we consider each game as a dot and each player as a side of the triangle, then the total number of games (45) would correspond to the 9th triangular number because 1+2+3+4+5+6+7+8+9=45. However, since each side of the triangle represents a player, and there's one more player than sides in our model (because the last player has no one left to play against), we need to add one more to get the total number of players.
507
+
508
+ Final answer: There were 10 players in the tournament.
509
+
510
+ Solution 3:
511
+ We could also use trial and error to solve this problem. Starting from a small number of players, we calculate how many games would be played in a round-robin tournament until we reach 45 games. For example, with 2 players, only 1 game is played; with 3 players, 3 games are played; with 4 players, 6 games are played, and so on. Continuing this process, we find that with 10 players, 45 games are played.
512
+
513
+ Final answer: There were 10 players in the tournament.
514
+
515
+ Ranking of solutions from best to worst:
516
+ 1. Solution 1: This solution uses direct mathematical reasoning based on the properties of a round-robin tournament and solves an equation to find the answer.
517
+ 2. Solution 2: This solution uses the concept of triangular numbers which requires additional steps and knowledge beyond the immediate problem context, but still arrives at the correct answer.
518
+ 3. Solution 3: This solution relies on trial and error, which is less efficient and reliable than the other methods, but it eventually gets to the right answer.
519
+
520
+ Best and final answer: There were 10 players in the tournament.
521
+ ```
522
+
523
+ #### reWOO style execution planning
524
+
525
+ The model now supports execution planning for complex instructions that would require making use of several tools. The output is just the plan, you must implement a mechanism to parse the output and actually call the functions!
526
+
527
+ Example prompt:
528
+ ```
529
+ Please construct a systematic plan to generate an optimal response to the user instruction, utilizing a set of provided tools. Each plan will correspond to an evidence value, which will be the output of one of the available functions given an input string
530
+ that could be the user's question, one or more prior evidence values, or a combination of both.
531
+
532
+ Here are the tools available to be called:
533
+ DuckDuckGo[input]: Tool that enables the user to search for information using the DuckDuckGo search engine. This tool is helpful in locating a suitable list of sites that may contain the answer to the user's query. It does not directly answer the question but finds a list of sites that might have the answer.
534
+ TextScraper[input]: Load one or more websites from the input string containing newline delimited links, where input is one or more links, and produces plain text output containing the content of the links.
535
+ HyperlinkExtractor[input]: Extract links from plain text and produces a plain text, newline delimited response of links.
536
+ KnowledgeModel[input]: Question answering language model, particularly useful in answering questions based on an input passage of text. The input must be a text question that references an :evidence[n]: variable, e.g. What color is the cat, given :evidence1:?
537
+
538
+ The output should be in format:
539
+ Plan: [first action to take based on input question]
540
+ :evidence0: = [function to call with input parameter]
541
+ Plan: [next action to take, based on result of :evidence0:]
542
+ :evidence1: = [next function to call and input parameter, which may include reference to previous evidence, e.g. "Given context :evidence0"]
543
+ ...
544
+ Answer: [:evidence[n]: containing the final answer.]
545
+
546
+ Question: Who won the Nobel Prize in Literature in 2020 and what is their most famous work?
547
+ ```
548
+
549
+ Response:
550
+ ```
551
+ Plan: Begin by conducting a web search to find out who won the Nobel Prize in Literature in 2020.
552
+ :evidence0: = DuckDuckGo[Nobel Prize in Literature 2020 winner]
553
+ Plan: Utilize the KnowledgeModel to interpret the search results and find the name of the laureate.
554
+ :evidence1: = KnowledgeModel[Who won the Nobel Prize in Literature in 2020, given :evidence0:?]
555
+ Plan: Conduct another web search to find the most famous work of the identified laureate.
556
+ :evidence2: = DuckDuckGo[Most famous work of :evidence1:]
557
+ Plan: Extract the relevant links from the DuckDuckGo search results for a more focused search.
558
+ :evidence3: = HyperlinkExtractor[:evidence2:]
559
+ Plan: Use the TextScraper tool to extract information from the relevant links.
560
+ :evidence4: = TextScraper[:evidence3:]
561
+ Plan: Finally, utilize the KnowledgeModel to identify and summarize the most famous work of the laureate from the extracted information.
562
+ :evidence5: = KnowledgeModel[What is the most famous work of :evidence1:, given :evidence4:?]
563
+ Answer: :evidence5:
564
+ ```
565
+
566
+ For this to be useful, you'd have to parse the output plan text, and implement/call each of the functions. This is just pseudo-code, completely untested off the top of my head, and obviously would requiring full implementation + hardening:
567
+
568
+ ```python
569
+ import re
570
+ import requests
571
+
572
+ def inject_context(input_text, **context):
573
+ for ref in set(re.findall(r"(:evidence[0-9]+:)", input_text, re.I)):
574
+ input_text = input_text.replace(ref, context.get(ref, ""))
575
+ return input_text
576
+
577
+ def duckduckgo(input_text, **context):
578
+ search_string = inject_context(input_text, **context)
579
+ ... search via duck duck go using search_string
580
+ ... return text content
581
+
582
+ def link_extractor(input_text, **context):
583
+ input_text = inject_context(input_text, **context)
584
+ return "\n".join(list(set(re.findall(r"(https?://[^\s]+?\.?)", input_text, re.I))))
585
+
586
+ def scrape(input_text, **context):
587
+ input_text = inject_context(input_text, **context)
588
+ text = []
589
+ for link in input_text.splitlines():
590
+ text.append(requests.get(link).text)
591
+ return "\n".join(text)
592
+
593
+ def infer(input_text, **context)
594
+ prompt = inject_context(input_text, **context)
595
+ ... call model with prompt, return output
596
+
597
+ def parse_plan(plan):
598
+ method_map = {
599
+ "DuckDuckGo": duckduckgo,
600
+ "HyperlinkExtractor": link_extractor,
601
+ "KnowledgeModel": infer,
602
+ "TextScraper": scrape,
603
+ }
604
+ context = {}
605
+ for line in plan.strip().splitlines():
606
+ if line.startswith("Plan:"):
607
+ print(line)
608
+ continue
609
+ parts = re.match("^(:evidence[0-9]+:)\s*=\s*([^\[]+])(\[.*\])\s$", line, re.I)
610
+ if not parts:
611
+ if line.startswith("Answer: "):
612
+ return context.get(line.split(" ")[-1].strip(), "Answer couldn't be generated...")
613
+ raise RuntimeError("bad format: " + line)
614
+ context[parts.group(1)] = method_map[parts.group(2)](parts.group(3), **context)
615
+ ```
616
+
617
+ ### Contribute
618
+
619
+ If you're interested in new functionality, particularly a new "instructor" type to generate a specific type of training data,
620
+ take a look at the dataset generation tool repo: https://github.com/jondurbin/airoboros and either make a PR or open an issue with details.
621
+
622
+ To help me with the OpenAI/compute costs:
623
+
624
+ - https://bmc.link/jondurbin
625
+ - ETH 0xce914eAFC2fe52FdceE59565Dd92c06f776fcb11
626
+ - BTC bc1qdwuth4vlg8x37ggntlxu5cjfwgmdy5zaa7pswf
627
+
628
+ ### Licence and usage restrictions
629
+
630
+ The airoboros 2.2 models are built on top of llama-2/codellama.
631
+
632
+ The llama-2 base model has a custom Meta license:
633
+ - See the [meta-license/LICENSE.txt](meta-license/LICENSE.txt) file attached for the original license provided by Meta.
634
+ - See also [meta-license/USE_POLICY.md](meta-license/USE_POLICY.md) and [meta-license/Responsible-Use-Guide.pdf](meta-license/Responsible-Use-Guide.pdf), also provided by Meta.
635
+
636
+ The fine-tuning data was mostly generated by OpenAI API calls to gpt-4, via [airoboros](https://github.com/jondurbin/airoboros)
637
+
638
+ The ToS for OpenAI API usage has a clause preventing the output from being used to train a model that __competes__ with OpenAI
639
+
640
+ - what does *compete* actually mean here?
641
+ - these small open source models will not produce output anywhere near the quality of gpt-4, or even gpt-3.5, so I can't imagine this could credibly be considered competing in the first place
642
+ - if someone else uses the dataset to do the same, they wouldn't necessarily be violating the ToS because they didn't call the API, so I don't know how that works
643
+ - the training data used in essentially all large language models includes a significant amount of copyrighted or otherwise non-permissive licensing in the first place
644
+ - other work using the self-instruct method, e.g. the original here: https://github.com/yizhongw/self-instruct released the data and model as apache-2
645
+
646
+ I am purposingly leaving this license ambiguous (other than the fact you must comply with the Meta original license for llama-2) because I am not a lawyer and refuse to attempt to interpret all of the terms accordingly.
647
+
648
+ Your best bet is probably to avoid using this commercially due to the OpenAI API usage.
649
+
650
+ Either way, by using this model, you agree to completely indemnify me.