TheBloke commited on
Commit
5786412
1 Parent(s): b57a149

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +471 -0
README.md ADDED
@@ -0,0 +1,471 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: openaccess-ai-collective/jackalope-7b
3
+ datasets:
4
+ - Open-Orca/OpenOrca
5
+ - LDJnr/LessWrong-Amplify-Instruct
6
+ - LDJnr/Pure-Dove
7
+ - LDJnr/Verified-Camel
8
+ - PygmalionAI/PIPPA
9
+ - meta-math/MetaMathQA
10
+ - riddle_sense
11
+ inference: false
12
+ language:
13
+ - en
14
+ library_name: transformers
15
+ license: apache-2.0
16
+ model_creator: Open Access AI Collective
17
+ model_name: Jackalope 7B
18
+ model_type: mistral
19
+ pipeline_tag: text-generation
20
+ prompt_template: '<|im_start|>system
21
+
22
+ {system_message}<|im_end|>
23
+
24
+ <|im_start|>user
25
+
26
+ {prompt}<|im_end|>
27
+
28
+ <|im_start|>assistant
29
+
30
+ '
31
+ quantized_by: TheBloke
32
+ ---
33
+
34
+ <!-- header start -->
35
+ <!-- 200823 -->
36
+ <div style="width: auto; margin-left: auto; margin-right: auto">
37
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
38
+ </div>
39
+ <div style="display: flex; justify-content: space-between; width: 100%;">
40
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
41
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
42
+ </div>
43
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
44
+ <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>
45
+ </div>
46
+ </div>
47
+ <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>
48
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
49
+ <!-- header end -->
50
+
51
+ # Jackalope 7B - AWQ
52
+ - Model creator: [Open Access AI Collective](https://huggingface.co/openaccess-ai-collective)
53
+ - Original model: [Jackalope 7B](https://huggingface.co/openaccess-ai-collective/jackalope-7b)
54
+
55
+ <!-- description start -->
56
+ ## Description
57
+
58
+ This repo contains AWQ model files for [Open Access AI Collective's Jackalope 7B](https://huggingface.co/openaccess-ai-collective/jackalope-7b).
59
+
60
+
61
+ ### About AWQ
62
+
63
+ AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference.
64
+
65
+ It is also now supported by continuous batching server [vLLM](https://github.com/vllm-project/vllm), allowing use of Llama AWQ models for high-throughput concurrent inference in multi-user server scenarios.
66
+
67
+ As of September 25th 2023, preliminary Llama-only AWQ support has also been added to [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference).
68
+
69
+ Note that, at the time of writing, overall throughput is still lower than running vLLM or TGI with unquantised models, however using AWQ enables using much smaller GPUs which can lead to easier deployment and overall cost savings. For example, a 70B model can be run on 1 x 48GB GPU instead of 2 x 80GB.
70
+ <!-- description end -->
71
+ <!-- repositories-available start -->
72
+ ## Repositories available
73
+
74
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/jackalope-7B-AWQ)
75
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/jackalope-7B-GPTQ)
76
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/jackalope-7B-GGUF)
77
+ * [Open Access AI Collective's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/openaccess-ai-collective/jackalope-7b)
78
+ <!-- repositories-available end -->
79
+
80
+ <!-- prompt-template start -->
81
+ ## Prompt template: ChatML
82
+
83
+ ```
84
+ <|im_start|>system
85
+ {system_message}<|im_end|>
86
+ <|im_start|>user
87
+ {prompt}<|im_end|>
88
+ <|im_start|>assistant
89
+
90
+ ```
91
+
92
+ <!-- prompt-template end -->
93
+
94
+
95
+ <!-- README_AWQ.md-provided-files start -->
96
+ ## Provided files, and AWQ parameters
97
+
98
+ For my first release of AWQ models, I am releasing 128g models only. I will consider adding 32g as well if there is interest, and once I have done perplexity and evaluation comparisons, but at this time 32g models are still not fully tested with AutoAWQ and vLLM.
99
+
100
+ Models are released as sharded safetensors files.
101
+
102
+ | Branch | Bits | GS | AWQ Dataset | Seq Len | Size |
103
+ | ------ | ---- | -- | ----------- | ------- | ---- |
104
+ | [main](https://huggingface.co/TheBloke/jackalope-7B-AWQ/tree/main) | 4 | 128 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-v1/test) | 4096 | 4.15 GB
105
+
106
+ <!-- README_AWQ.md-provided-files end -->
107
+
108
+ <!-- README_AWQ.md-use-from-vllm start -->
109
+ ## Serving this model from vLLM
110
+
111
+ Documentation on installing and using vLLM [can be found here](https://vllm.readthedocs.io/en/latest/).
112
+
113
+ Note: at the time of writing, vLLM has not yet done a new release with AWQ support.
114
+
115
+ If you try the vLLM examples below and get an error about `quantization` being unrecognised, or other AWQ-related issues, please install vLLM from Github source.
116
+
117
+ - When using vLLM as a server, pass the `--quantization awq` parameter, for example:
118
+
119
+ ```shell
120
+ python3 python -m vllm.entrypoints.api_server --model TheBloke/jackalope-7B-AWQ --quantization awq --dtype half
121
+ ```
122
+
123
+ When using vLLM from Python code, pass the `quantization=awq` parameter, for example:
124
+
125
+ ```python
126
+ from vllm import LLM, SamplingParams
127
+
128
+ prompts = [
129
+ "Hello, my name is",
130
+ "The president of the United States is",
131
+ "The capital of France is",
132
+ "The future of AI is",
133
+ ]
134
+ sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
135
+
136
+ llm = LLM(model="TheBloke/jackalope-7B-AWQ", quantization="awq", dtype="half")
137
+
138
+ outputs = llm.generate(prompts, sampling_params)
139
+
140
+ # Print the outputs.
141
+ for output in outputs:
142
+ prompt = output.prompt
143
+ generated_text = output.outputs[0].text
144
+ print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
145
+ ```
146
+ <!-- README_AWQ.md-use-from-vllm start -->
147
+
148
+ <!-- README_AWQ.md-use-from-tgi start -->
149
+ ## Serving this model from Text Generation Inference (TGI)
150
+
151
+ Use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0`
152
+
153
+ Example Docker parameters:
154
+
155
+ ```shell
156
+ --model-id TheBloke/jackalope-7B-AWQ --port 3000 --quantize awq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096
157
+ ```
158
+
159
+ Example Python code for interfacing with TGI (requires huggingface-hub 0.17.0 or later):
160
+
161
+ ```shell
162
+ pip3 install huggingface-hub
163
+ ```
164
+
165
+ ```python
166
+ from huggingface_hub import InferenceClient
167
+
168
+ endpoint_url = "https://your-endpoint-url-here"
169
+
170
+ prompt = "Tell me about AI"
171
+ prompt_template=f'''<|im_start|>system
172
+ {system_message}<|im_end|>
173
+ <|im_start|>user
174
+ {prompt}<|im_end|>
175
+ <|im_start|>assistant
176
+
177
+ '''
178
+
179
+ client = InferenceClient(endpoint_url)
180
+ response = client.text_generation(prompt,
181
+ max_new_tokens=128,
182
+ do_sample=True,
183
+ temperature=0.7,
184
+ top_p=0.95,
185
+ top_k=40,
186
+ repetition_penalty=1.1)
187
+
188
+ print(f"Model output: {response}")
189
+ ```
190
+ <!-- README_AWQ.md-use-from-tgi end -->
191
+
192
+ <!-- README_AWQ.md-use-from-python start -->
193
+ ## How to use this AWQ model from Python code
194
+
195
+ ### Install the necessary packages
196
+
197
+ Requires: [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) 0.1.1 or later
198
+
199
+ ```shell
200
+ pip3 install autoawq
201
+ ```
202
+
203
+ If you have problems installing [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) using the pre-built wheels, install it from source instead:
204
+
205
+ ```shell
206
+ pip3 uninstall -y autoawq
207
+ git clone https://github.com/casper-hansen/AutoAWQ
208
+ cd AutoAWQ
209
+ pip3 install .
210
+ ```
211
+
212
+ ### You can then try the following example code
213
+
214
+ ```python
215
+ from awq import AutoAWQForCausalLM
216
+ from transformers import AutoTokenizer
217
+
218
+ model_name_or_path = "TheBloke/jackalope-7B-AWQ"
219
+
220
+ # Load model
221
+ model = AutoAWQForCausalLM.from_quantized(model_name_or_path, fuse_layers=True,
222
+ trust_remote_code=False, safetensors=True)
223
+ tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=False)
224
+
225
+ prompt = "Tell me about AI"
226
+ prompt_template=f'''<|im_start|>system
227
+ {system_message}<|im_end|>
228
+ <|im_start|>user
229
+ {prompt}<|im_end|>
230
+ <|im_start|>assistant
231
+
232
+ '''
233
+
234
+ print("\n\n*** Generate:")
235
+
236
+ tokens = tokenizer(
237
+ prompt_template,
238
+ return_tensors='pt'
239
+ ).input_ids.cuda()
240
+
241
+ # Generate output
242
+ generation_output = model.generate(
243
+ tokens,
244
+ do_sample=True,
245
+ temperature=0.7,
246
+ top_p=0.95,
247
+ top_k=40,
248
+ max_new_tokens=512
249
+ )
250
+
251
+ print("Output: ", tokenizer.decode(generation_output[0]))
252
+
253
+ """
254
+ # Inference should be possible with transformers pipeline as well in future
255
+ # But currently this is not yet supported by AutoAWQ (correct as of September 25th 2023)
256
+ from transformers import pipeline
257
+
258
+ print("*** Pipeline:")
259
+ pipe = pipeline(
260
+ "text-generation",
261
+ model=model,
262
+ tokenizer=tokenizer,
263
+ max_new_tokens=512,
264
+ do_sample=True,
265
+ temperature=0.7,
266
+ top_p=0.95,
267
+ top_k=40,
268
+ repetition_penalty=1.1
269
+ )
270
+
271
+ print(pipe(prompt_template)[0]['generated_text'])
272
+ """
273
+ ```
274
+ <!-- README_AWQ.md-use-from-python end -->
275
+
276
+ <!-- README_AWQ.md-compatibility start -->
277
+ ## Compatibility
278
+
279
+ The files provided are tested to work with:
280
+
281
+ - [AutoAWQ](https://github.com/casper-hansen/AutoAWQ)
282
+ - [vLLM](https://github.com/vllm-project/vllm)
283
+ - [Huggingface Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)
284
+
285
+ TGI merged AWQ support on September 25th, 2023: [TGI PR #1054](https://github.com/huggingface/text-generation-inference/pull/1054). Use the `:latest` Docker container until the next TGI release is made.
286
+
287
+ <!-- README_AWQ.md-compatibility end -->
288
+
289
+ <!-- footer start -->
290
+ <!-- 200823 -->
291
+ ## Discord
292
+
293
+ For further support, and discussions on these models and AI in general, join us at:
294
+
295
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
296
+
297
+ ## Thanks, and how to contribute
298
+
299
+ Thanks to the [chirper.ai](https://chirper.ai) team!
300
+
301
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
302
+
303
+ 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.
304
+
305
+ 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.
306
+
307
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
308
+
309
+ * Patreon: https://patreon.com/TheBlokeAI
310
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
311
+
312
+ **Special thanks to**: Aemon Algiz.
313
+
314
+ **Patreon special mentions**: Pierre Kircher, Stanislav Ovsiannikov, Michael Levine, Eugene Pentland, Andrey, 준교 김, Randy H, Fred von Graf, Artur Olbinski, Caitlyn Gatomon, terasurfer, Jeff Scroggin, James Bentley, Vadim, Gabriel Puliatti, Harry Royden McLaughlin, Sean Connelly, Dan Guido, Edmond Seymore, Alicia Loh, subjectnull, AzureBlack, Manuel Alberto Morcote, Thomas Belote, Lone Striker, Chris Smitley, Vitor Caleffi, Johann-Peter Hartmann, Clay Pascal, biorpg, Brandon Frisco, sidney chen, transmissions 11, Pedro Madruga, jinyuan sun, Ajan Kanaga, Emad Mostaque, Trenton Dambrowitz, Jonathan Leane, Iucharbius, usrbinkat, vamX, George Stoitzev, Luke Pendergrass, theTransient, Olakabola, Swaroop Kallakuri, Cap'n Zoog, Brandon Phillips, Michael Dempsey, Nikolai Manek, danny, Matthew Berman, Gabriel Tamborski, alfie_i, Raymond Fosdick, Tom X Nguyen, Raven Klaugh, LangChain4j, Magnesian, Illia Dulskyi, David Ziegler, Mano Prime, Luis Javier Navarrete Lozano, Erik Bjäreholt, 阿明, Nathan Dryer, Alex, Rainer Wilmers, zynix, TL, Joseph William Delisle, John Villwock, Nathan LeClaire, Willem Michiel, Joguhyik, GodLy, OG, Alps Aficionado, Jeffrey Morgan, ReadyPlayerEmma, Tiffany J. Kim, Sebastain Graf, Spencer Kim, Michael Davis, webtim, Talal Aujan, knownsqashed, John Detwiler, Imad Khwaja, Deo Leter, Jerry Meng, Elijah Stavena, Rooh Singh, Pieter, SuperWojo, Alexandros Triantafyllidis, Stephen Murray, Ai Maven, ya boyyy, Enrico Ros, Ken Nordquist, Deep Realms, Nicholas, Spiking Neurons AB, Elle, Will Dee, Jack West, RoA, Luke @flexchar, Viktor Bowallius, Derek Yates, Subspace Studios, jjj, Toran Billups, Asp the Wyvern, Fen Risland, Ilya, NimbleBox.ai, Chadd, Nitin Borwankar, Emre, Mandus, Leonard Tan, Kalila, K, Trailburnt, S_X, Cory Kujawski
315
+
316
+
317
+ Thank you to all my generous patrons and donaters!
318
+
319
+ And thank you again to a16z for their generous grant.
320
+
321
+ <!-- footer end -->
322
+
323
+ # Original model card: Open Access AI Collective's Jackalope 7B
324
+
325
+
326
+ <p><h1>🐰🦌 Jackalope 7B 🐰🦌</h1></p>
327
+
328
+
329
+ ![Jackalope Logo](https://huggingface.co/openaccess-ai-collective/jackalope-7b/resolve/main/images/jackalope.jpg "Jackalope Logo")
330
+ [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
331
+
332
+
333
+ # Jackalope 7B
334
+
335
+ We have used the [SlimOrca dataset](https://huggingface.co/datasets/Open-Orca/SlimOrca), PIPPA, and various other open datasets
336
+ to fine-tune on top of [Mistral 7B](https://huggingface.co/mistralai/Mistral-7B-v0.1).
337
+
338
+ This dataset is our attempt to reproduce the dataset generated for Microsoft Research's [Orca Paper](https://arxiv.org/abs/2306.02707).
339
+ We use [OpenChat](https://huggingface.co/openchat) packing, trained with [Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl).
340
+
341
+ This release highlights the efficiency of SlimOrca, while improving the ability of the model's multi-turn chat.
342
+
343
+ HF Leaderboard evals puts this model only slightly below the MistralOrca release, but can be considered a
344
+ reasonable tradeoff for a more general model that can handle multi-turn chat.
345
+
346
+ If you'd like to try the model now, we have it running on fast GPUs unquantized: https://huggingface.co/spaces/openaccess-ai-collective/jackalope-7b
347
+
348
+
349
+ Join the OpenAccess AI Collective Discord for more information about Axolotl trainer and other OAAIC models here:
350
+
351
+ https://discord.gg/5y8STgB3P3
352
+
353
+ Also join the AlignmentLab Discord for sneak-peak announcements:
354
+
355
+ https://AlignmentLab.ai
356
+
357
+
358
+
359
+ # Quantized Models
360
+
361
+ Quantized versions of this model are generously made available by [TheBloke](https://huggingface.co/TheBloke).
362
+
363
+ - AWQ: https://huggingface.co/TheBloke/Jackalope-7B-AWQ
364
+ - GPTQ: https://huggingface.co/TheBloke/Jackalope-7B-GPTQ
365
+ - GGUF: https://huggingface.co/TheBloke/Jackalope-7B-GGUF
366
+
367
+
368
+ # Prompt Template
369
+
370
+ We used [OpenAI's Chat Markup Language (ChatML)](https://github.com/openai/openai-python/blob/main/chatml.md) format, with `<|im_start|>` and `<|im_end|>` tokens added to support this.
371
+
372
+ This means that, e.g., in [oobabooga](https://github.com/oobabooga/text-generation-webui/) the "`MPT-Chat`" instruction template should work, as it also uses ChatML.
373
+
374
+ This formatting is also available via a pre-defined [Transformers chat template](https://huggingface.co/docs/transformers/main/chat_templating),
375
+ which means that lists of messages can be formatted for you with the `apply_chat_template()` method:
376
+
377
+ ```python
378
+ chat = [
379
+ {"role": "system", "content": "You are JackalopeAI, a large language model trained by OpenAccess AI Collective. Write out your reasoning step-by-step to be sure you get the right answers!"}
380
+ {"role": "user", "content": "How are you?"},
381
+ {"role": "assistant", "content": "I am doing well!"},
382
+ {"role": "user", "content": "Please tell me about the mythical creatures called jackalopes."},
383
+ ]
384
+ tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
385
+ ```
386
+
387
+ which will yield:
388
+
389
+ ```
390
+ <|im_start|>system
391
+ You are JackalopeAI. Write out your reasoning step-by-step to be sure you get the right answers!
392
+ <|im_end|>
393
+ <|im_start|>user
394
+ How are you?<|im_end|>
395
+ <|im_start|>assistant
396
+ I am doing well!<|im_end|>
397
+ <|im_start|>user
398
+ Please tell me about the mythical creatures called jackalopes.<|im_end|>
399
+ <|im_start|>assistant
400
+ ```
401
+
402
+ If you use `tokenize=True` and `return_tensors="pt"` instead, then you will get a tokenized
403
+ and formatted conversation ready to pass to `model.generate()`.
404
+
405
+
406
+ # Evaluation
407
+
408
+ ## HuggingFace Leaderboard Performance
409
+
410
+ ![All benchmarks](https://huggingface.co/openaccess-ai-collective/jackalope-7b/resolve/main/images/bench.png)
411
+
412
+
413
+ | Metric | Value |
414
+ |-----------------------|--|
415
+ | MMLU (5-shot) | 63.63 |
416
+ | ARC (25-shot) | 63.31 |
417
+ | HellaSwag (10-shot) | 83.29 |
418
+ | TruthfulQA (0-shot) | 49.99 |
419
+ | Avg. | 65.06 |
420
+
421
+ We use [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) to run the benchmark tests above, using the same version as the HuggingFace LLM Leaderboard.
422
+
423
+ # Dataset
424
+
425
+ We used a verified, curated, filtered selection of most of the GPT-4 augmented data from the OpenOrca dataset.
426
+ Additionally we include multi-turn chat from PIPPA, various datasets
427
+ by LDJ from Nous Research, MetaMathQA, and Chain-of-Thought augmented data from the train split of RiddleSense.
428
+
429
+ - [Open-Orca/OpenOrca](https://huggingface.co/datasets/Open-Orca/OpenOrca)
430
+ - [LDJnr/LessWrong-Amplify-Instruct](https://huggingface.co/datasets/LDJnr/LessWrong-Amplify-Instruct)
431
+ - [LDJnr/Pure-Dove](https://huggingface.co/datasets/LDJnr/Pure-Dove)
432
+ - [LDJnr/Verified-Camel](https://huggingface.co/datasets/LDJnr/Verified-Camel)
433
+ - [PygmalionAI/PIPPA](https://huggingface.co/datasets/PygmalionAI/PIPPA)
434
+ - [meta-math/MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA)
435
+ - [riddle_sense](https://huggingface.co/datasets/riddle_sense)
436
+
437
+
438
+ # Training
439
+
440
+ We trained with 8x A6000 GPUs for 96 hours, completing 4 epochs of full fine tuning on our dataset in one training run.
441
+ Commodity cost was ~$650.
442
+
443
+
444
+ # Citation
445
+
446
+ ```bibtex
447
+ @software{lian2023jackalope,
448
+ title = {Jackalope 7B: Mistral-7B Model Multi-Turn Chat tuned on Filtered OpenOrcaV1 GPT-4 Dataset},
449
+ author = {Wing Lian and Bleys Goodson and Guan Wang and Eugene Pentland and Austin Cook and Chanvichet Vong and "Teknium"},
450
+ year = {2023},
451
+ publisher = {HuggingFace},
452
+ journal = {HuggingFace repository},
453
+ howpublished = {\url{openaccess-ai-collective/jackalope-7b},
454
+ }
455
+ @misc{mukherjee2023orca,
456
+ title={Orca: Progressive Learning from Complex Explanation Traces of GPT-4},
457
+ author={Subhabrata Mukherjee and Arindam Mitra and Ganesh Jawahar and Sahaj Agarwal and Hamid Palangi and Ahmed Awadallah},
458
+ year={2023},
459
+ eprint={2306.02707},
460
+ archivePrefix={arXiv},
461
+ primaryClass={cs.CL}
462
+ }
463
+ @misc{longpre2023flan,
464
+ title={The Flan Collection: Designing Data and Methods for Effective Instruction Tuning},
465
+ author={Shayne Longpre and Le Hou and Tu Vu and Albert Webson and Hyung Won Chung and Yi Tay and Denny Zhou and Quoc V. Le and Barret Zoph and Jason Wei and Adam Roberts},
466
+ year={2023},
467
+ eprint={2301.13688},
468
+ archivePrefix={arXiv},
469
+ primaryClass={cs.AI}
470
+ }
471
+ ```