TheBloke commited on
Commit
1c657a3
1 Parent(s): 77dbf48

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +465 -0
README.md ADDED
@@ -0,0 +1,465 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: 01-ai/Yi-6B
3
+ inference: false
4
+ license: other
5
+ license_link: LICENSE
6
+ license_name: yi-license
7
+ model_creator: 01-ai
8
+ model_name: Yi 6B
9
+ model_type: yi
10
+ pipeline_tag: text-generation
11
+ prompt_template: '{prompt}
12
+
13
+ '
14
+ quantized_by: TheBloke
15
+ widget:
16
+ - output:
17
+ text: " an eerie sense that something is just not right\u2026\nBetween the two\
18
+ \ worlds lies The Forgotten Kingdom - home to creatures long since thought extinct\
19
+ \ and ancient magic so strong it defies belief! Only here can you find what\
20
+ \ has been lost for centuries: An Elixir Of Life which will restore youth and\
21
+ \ vitality if only those who seek its power are brave enough to face up against\
22
+ \ all manner of dangers lurking in this mysterious land! But beware; some say\
23
+ \ there may even exist powerful entities beyond our comprehension whose intentions\
24
+ \ towards humanity remain unclear at best ---- they might want nothing more\
25
+ \ than destruction itself rather then anything else from their quest after immortality\
26
+ \ (and maybe someone should tell them about modern medicine)? In any event though\
27
+ \ \u2013 one thing remains true regardless : whether or not success comes easy\
28
+ \ depends entirely upon how much effort we put into conquering whatever challenges\
29
+ \ lie ahead along with having faith deep down inside ourselves too ;) So let\u2019\
30
+ s get started now shall We?"
31
+ text: There's a place where time stands still. A place of breath taking wonder,
32
+ but also
33
+ ---
34
+ <!-- markdownlint-disable MD041 -->
35
+
36
+ <!-- header start -->
37
+ <!-- 200823 -->
38
+ <div style="width: auto; margin-left: auto; margin-right: auto">
39
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
40
+ </div>
41
+ <div style="display: flex; justify-content: space-between; width: 100%;">
42
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
43
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
44
+ </div>
45
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
46
+ <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>
47
+ </div>
48
+ </div>
49
+ <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>
50
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
51
+ <!-- header end -->
52
+
53
+ # Yi 6B - GPTQ
54
+ - Model creator: [01-ai](https://huggingface.co/01-ai)
55
+ - Original model: [Yi 6B](https://huggingface.co/01-ai/Yi-6B)
56
+
57
+ <!-- description start -->
58
+ ## Description
59
+
60
+ This repo contains GPTQ model files for [01-ai's Yi 6B](https://huggingface.co/01-ai/Yi-6B).
61
+
62
+ 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.
63
+
64
+ These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
65
+
66
+ <!-- description end -->
67
+ <!-- repositories-available start -->
68
+ ## Repositories available
69
+
70
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/Yi-6B-AWQ)
71
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Yi-6B-GPTQ)
72
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Yi-6B-GGUF)
73
+ * [01-ai's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/01-ai/Yi-6B)
74
+ <!-- repositories-available end -->
75
+
76
+ <!-- prompt-template start -->
77
+ ## Prompt template: None
78
+
79
+ ```
80
+ {prompt}
81
+
82
+ ```
83
+
84
+ <!-- prompt-template end -->
85
+
86
+
87
+
88
+ <!-- README_GPTQ.md-compatible clients start -->
89
+ ## Known compatible clients / servers
90
+
91
+ These GPTQ models are known to work in the following inference servers/webuis.
92
+
93
+ - [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
94
+ - [KoboldAI United](https://github.com/henk717/koboldai)
95
+ - [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui)
96
+ - [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)
97
+
98
+ This may not be a complete list; if you know of others, please let me know!
99
+ <!-- README_GPTQ.md-compatible clients end -->
100
+
101
+ <!-- README_GPTQ.md-provided-files start -->
102
+ ## Provided files, and GPTQ parameters
103
+
104
+ Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
105
+
106
+ Each separate quant is in a different branch. See below for instructions on fetching from different branches.
107
+
108
+ Most GPTQ files are made with AutoGPTQ. Mistral models are currently made with Transformers.
109
+
110
+ <details>
111
+ <summary>Explanation of GPTQ parameters</summary>
112
+
113
+ - Bits: The bit size of the quantised model.
114
+ - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
115
+ - 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.
116
+ - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
117
+ - GPTQ dataset: The calibration dataset used during quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ calibration 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).
118
+ - 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.
119
+ - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama and Mistral models in 4-bit.
120
+
121
+ </details>
122
+
123
+ | Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
124
+ | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |
125
+ | [main](https://huggingface.co/TheBloke/Yi-6B-GPTQ/tree/main) | 4 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-raw-v1) | 4096 | 3.93 GB | Yes | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. |
126
+ | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/Yi-6B-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-raw-v1) | 4096 | 4.26 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. |
127
+ | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/Yi-6B-GPTQ/tree/gptq-8bit--1g-actorder_True) | 8 | None | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-raw-v1) | 4096 | 4.99 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements. |
128
+ | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/Yi-6B-GPTQ/tree/gptq-8bit-128g-actorder_True) | 8 | 128 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-raw-v1) | 4096 | 5.00 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. |
129
+ | [gptq-8bit-32g-actorder_True](https://huggingface.co/TheBloke/Yi-6B-GPTQ/tree/gptq-8bit-32g-actorder_True) | 8 | 32 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-raw-v1) | 4096 | 4.97 GB | No | 8-bit, with group size 32g and Act Order for maximum inference quality. |
130
+ | [gptq-4bit-64g-actorder_True](https://huggingface.co/TheBloke/Yi-6B-GPTQ/tree/gptq-4bit-64g-actorder_True) | 4 | 64 | Yes | 0.1 | [wikitext](https://huggingface.co/datasets/wikitext/viewer/wikitext-2-raw-v1) | 4096 | 4.04 GB | Yes | 4-bit, with Act Order and group size 64g. Uses less VRAM than 32g, but with slightly lower accuracy. |
131
+
132
+ <!-- README_GPTQ.md-provided-files end -->
133
+
134
+ <!-- README_GPTQ.md-download-from-branches start -->
135
+ ## How to download, including from branches
136
+
137
+ ### In text-generation-webui
138
+
139
+ To download from the `main` branch, enter `TheBloke/Yi-6B-GPTQ` in the "Download model" box.
140
+
141
+ To download from another branch, add `:branchname` to the end of the download name, eg `TheBloke/Yi-6B-GPTQ:gptq-4bit-32g-actorder_True`
142
+
143
+ ### From the command line
144
+
145
+ I recommend using the `huggingface-hub` Python library:
146
+
147
+ ```shell
148
+ pip3 install huggingface-hub
149
+ ```
150
+
151
+ To download the `main` branch to a folder called `Yi-6B-GPTQ`:
152
+
153
+ ```shell
154
+ mkdir Yi-6B-GPTQ
155
+ huggingface-cli download TheBloke/Yi-6B-GPTQ --local-dir Yi-6B-GPTQ --local-dir-use-symlinks False
156
+ ```
157
+
158
+ To download from a different branch, add the `--revision` parameter:
159
+
160
+ ```shell
161
+ mkdir Yi-6B-GPTQ
162
+ huggingface-cli download TheBloke/Yi-6B-GPTQ --revision gptq-4bit-32g-actorder_True --local-dir Yi-6B-GPTQ --local-dir-use-symlinks False
163
+ ```
164
+
165
+ <details>
166
+ <summary>More advanced huggingface-cli download usage</summary>
167
+
168
+ If you remove the `--local-dir-use-symlinks False` parameter, the files will instead be stored in the central Hugging Face cache directory (default location on Linux is: `~/.cache/huggingface`), and symlinks will be added to the specified `--local-dir`, pointing to their real location in the cache. This allows for interrupted downloads to be resumed, and allows you to quickly clone the repo to multiple places on disk without triggering a download again. The downside, and the reason why I don't list that as the default option, is that the files are then hidden away in a cache folder and it's harder to know where your disk space is being used, and to clear it up if/when you want to remove a download model.
169
+
170
+ The cache location can be changed with the `HF_HOME` environment variable, and/or the `--cache-dir` parameter to `huggingface-cli`.
171
+
172
+ 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).
173
+
174
+ To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
175
+
176
+ ```shell
177
+ pip3 install hf_transfer
178
+ ```
179
+
180
+ And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
181
+
182
+ ```shell
183
+ mkdir Yi-6B-GPTQ
184
+ HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/Yi-6B-GPTQ --local-dir Yi-6B-GPTQ --local-dir-use-symlinks False
185
+ ```
186
+
187
+ Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
188
+ </details>
189
+
190
+ ### With `git` (**not** recommended)
191
+
192
+ To clone a specific branch with `git`, use a command like this:
193
+
194
+ ```shell
195
+ git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/Yi-6B-GPTQ
196
+ ```
197
+
198
+ Note that using Git with HF repos is strongly discouraged. It will be much slower than using `huggingface-hub`, and will use twice as much disk space as it has to store the model files twice (it stores every byte both in the intended target folder, and again in the `.git` folder as a blob.)
199
+
200
+ <!-- README_GPTQ.md-download-from-branches end -->
201
+ <!-- README_GPTQ.md-text-generation-webui start -->
202
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
203
+
204
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
205
+
206
+ 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.
207
+
208
+ 1. Click the **Model tab**.
209
+ 2. Under **Download custom model or LoRA**, enter `TheBloke/Yi-6B-GPTQ`.
210
+
211
+ - To download from a specific branch, enter for example `TheBloke/Yi-6B-GPTQ:gptq-4bit-32g-actorder_True`
212
+ - see Provided Files above for the list of branches for each option.
213
+
214
+ 3. Click **Download**.
215
+ 4. The model will start downloading. Once it's finished it will say "Done".
216
+ 5. In the top left, click the refresh icon next to **Model**.
217
+ 6. In the **Model** dropdown, choose the model you just downloaded: `Yi-6B-GPTQ`
218
+ 7. The model will automatically load, and is now ready for use!
219
+ 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.
220
+
221
+ - 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`.
222
+
223
+ 9. Once you're ready, click the **Text Generation** tab and enter a prompt to get started!
224
+
225
+ <!-- README_GPTQ.md-text-generation-webui end -->
226
+
227
+ <!-- README_GPTQ.md-use-from-tgi start -->
228
+ ## Serving this model from Text Generation Inference (TGI)
229
+
230
+ It's recommended to use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0`
231
+
232
+ Example Docker parameters:
233
+
234
+ ```shell
235
+ --model-id TheBloke/Yi-6B-GPTQ --port 3000 --quantize gptq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096
236
+ ```
237
+
238
+ Example Python code for interfacing with TGI (requires huggingface-hub 0.17.0 or later):
239
+
240
+ ```shell
241
+ pip3 install huggingface-hub
242
+ ```
243
+
244
+ ```python
245
+ from huggingface_hub import InferenceClient
246
+
247
+ endpoint_url = "https://your-endpoint-url-here"
248
+
249
+ prompt = "Tell me about AI"
250
+ prompt_template=f'''{prompt}
251
+ '''
252
+
253
+ client = InferenceClient(endpoint_url)
254
+ response = client.text_generation(prompt,
255
+ max_new_tokens=128,
256
+ do_sample=True,
257
+ temperature=0.7,
258
+ top_p=0.95,
259
+ top_k=40,
260
+ repetition_penalty=1.1)
261
+
262
+ print(f"Model output: {response}")
263
+ ```
264
+ <!-- README_GPTQ.md-use-from-tgi end -->
265
+ <!-- README_GPTQ.md-use-from-python start -->
266
+ ## How to use this GPTQ model from Python code
267
+
268
+ ### Install the necessary packages
269
+
270
+ Requires: Transformers 4.33.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
271
+
272
+ ```shell
273
+ pip3 install transformers optimum
274
+ pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
275
+ ```
276
+
277
+ If you have problems installing AutoGPTQ using the pre-built wheels, install it from source instead:
278
+
279
+ ```shell
280
+ pip3 uninstall -y auto-gptq
281
+ git clone https://github.com/PanQiWei/AutoGPTQ
282
+ cd AutoGPTQ
283
+ git checkout v0.4.2
284
+ pip3 install .
285
+ ```
286
+
287
+ ### You can then use the following code
288
+
289
+ ```python
290
+ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
291
+
292
+ model_name_or_path = "TheBloke/Yi-6B-GPTQ"
293
+ # To use a different branch, change revision
294
+ # For example: revision="gptq-4bit-32g-actorder_True"
295
+ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
296
+ device_map="auto",
297
+ trust_remote_code=True,
298
+ revision="main")
299
+
300
+ tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
301
+
302
+ prompt = "Tell me about AI"
303
+ prompt_template=f'''{prompt}
304
+ '''
305
+
306
+ print("\n\n*** Generate:")
307
+
308
+ input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
309
+ output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512)
310
+ print(tokenizer.decode(output[0]))
311
+
312
+ # Inference can also be done using transformers' pipeline
313
+
314
+ print("*** Pipeline:")
315
+ pipe = pipeline(
316
+ "text-generation",
317
+ model=model,
318
+ tokenizer=tokenizer,
319
+ max_new_tokens=512,
320
+ do_sample=True,
321
+ temperature=0.7,
322
+ top_p=0.95,
323
+ top_k=40,
324
+ repetition_penalty=1.1
325
+ )
326
+
327
+ print(pipe(prompt_template)[0]['generated_text'])
328
+ ```
329
+ <!-- README_GPTQ.md-use-from-python end -->
330
+
331
+ <!-- README_GPTQ.md-compatibility start -->
332
+ ## Compatibility
333
+
334
+ The files provided are tested to work with Transformers. For non-Mistral models, AutoGPTQ can also be used directly.
335
+
336
+ [ExLlama](https://github.com/turboderp/exllama) is compatible with Llama and Mistral models in 4-bit. Please see the Provided Files table above for per-file compatibility.
337
+
338
+ For a list of clients/servers, please see "Known compatible clients / servers", above.
339
+ <!-- README_GPTQ.md-compatibility end -->
340
+
341
+ <!-- footer start -->
342
+ <!-- 200823 -->
343
+ ## Discord
344
+
345
+ For further support, and discussions on these models and AI in general, join us at:
346
+
347
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
348
+
349
+ ## Thanks, and how to contribute
350
+
351
+ Thanks to the [chirper.ai](https://chirper.ai) team!
352
+
353
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
354
+
355
+ 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.
356
+
357
+ 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.
358
+
359
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
360
+
361
+ * Patreon: https://patreon.com/TheBlokeAI
362
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
363
+
364
+ **Special thanks to**: Aemon Algiz.
365
+
366
+ **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
367
+
368
+
369
+ Thank you to all my generous patrons and donaters!
370
+
371
+ And thank you again to a16z for their generous grant.
372
+
373
+ <!-- footer end -->
374
+
375
+ # Original model card: 01-ai's Yi 6B
376
+
377
+ <div align="center">
378
+
379
+ <img src="./Yi.svg" width="200px">
380
+
381
+ </div>
382
+
383
+ ## Introduction
384
+
385
+ The **Yi** series models are large language models trained from scratch by
386
+ developers at [01.AI](https://01.ai/). The first public release contains two
387
+ bilingual(English/Chinese) base models with the parameter sizes of 6B([`Yi-6B`](https://huggingface.co/01-ai/Yi-6B))
388
+ and 34B([`Yi-34B`](https://huggingface.co/01-ai/Yi-34B)). Both of them are trained
389
+ with 4K sequence length and can be extended to 32K during inference time.
390
+ The [`Yi-6B-200K`](https://huggingface.co/01-ai/Yi-6B-200K)
391
+ and [`Yi-34B-200K`](https://huggingface.co/01-ai/Yi-34B-200K) are base model with
392
+ 200K context length.
393
+
394
+ ## News
395
+
396
+ - 🎯 **2023/11/06**: The base model of [`Yi-6B-200K`](https://huggingface.co/01-ai/Yi-6B-200K)
397
+ and [`Yi-34B-200K`](https://huggingface.co/01-ai/Yi-34B-200K) with 200K context length.
398
+ - 🎯 **2023/11/02**: The base model of [`Yi-6B`](https://huggingface.co/01-ai/Yi-6B) and
399
+ [`Yi-34B`](https://huggingface.co/01-ai/Yi-34B).
400
+
401
+
402
+ ## Model Performance
403
+
404
+ | Model | MMLU | CMMLU | C-Eval | GAOKAO | BBH | Common-sense Reasoning | Reading Comprehension | Math & Code |
405
+ | :------------ | :------: | :------: | :------: | :------: | :------: | :--------------------: | :-------------------: | :---------: |
406
+ | | 5-shot | 5-shot | 5-shot | 0-shot | 3-shot@1 | - | - | - |
407
+ | LLaMA2-34B | 62.6 | - | - | - | 44.1 | 69.9 | 68.0 | 26.0 |
408
+ | LLaMA2-70B | 68.9 | 53.3 | - | 49.8 | 51.2 | 71.9 | 69.4 | 36.8 |
409
+ | Baichuan2-13B | 59.2 | 62.0 | 58.1 | 54.3 | 48.8 | 64.3 | 62.4 | 23.0 |
410
+ | Qwen-14B | 66.3 | 71.0 | 72.1 | 62.5 | 53.4 | 73.3 | 72.5 | **39.8** |
411
+ | Skywork-13B | 62.1 | 61.8 | 60.6 | 68.1 | 41.7 | 72.4 | 61.4 | 24.9 |
412
+ | InternLM-20B | 62.1 | 59.0 | 58.8 | 45.5 | 52.5 | 78.3 | - | 30.4 |
413
+ | Aquila-34B | 67.8 | 71.4 | 63.1 | - | - | - | - | - |
414
+ | Falcon-180B | 70.4 | 58.0 | 57.8 | 59.0 | 54.0 | 77.3 | 68.8 | 34.0 |
415
+ | Yi-6B | 63.2 | 75.5 | 72.0 | 72.2 | 42.8 | 72.3 | 68.7 | 19.8 |
416
+ | Yi-6B-200K | 64.0 | 75.3 | 73.5 | 73.9 | 42.0 | 72.0 | 69.1 | 19.0 |
417
+ | **Yi-34B** | **76.3** | **83.7** | 81.4 | 82.8 | **54.3** | **80.1** | 76.4 | 37.1 |
418
+ | Yi-34B-200K | 76.1 | 83.6 | **81.9** | **83.4** | 52.7 | 79.7 | **76.6** | 36.3 |
419
+
420
+ While benchmarking open-source models, we have observed a disparity between the
421
+ results generated by our pipeline and those reported in public sources (e.g.
422
+ OpenCompass). Upon conducting a more in-depth investigation of this difference,
423
+ we have discovered that various models may employ different prompts,
424
+ post-processing strategies, and sampling techniques, potentially resulting in
425
+ significant variations in the outcomes. Our prompt and post-processing strategy
426
+ remains consistent with the original benchmark, and greedy decoding is employed
427
+ during evaluation without any post-processing for the generated content. For
428
+ scores that were not reported by the original authors (including scores reported
429
+ with different settings), we try to get results with our pipeline.
430
+
431
+ To evaluate the model's capability extensively, we adopted the methodology
432
+ outlined in Llama2. Specifically, we included PIQA, SIQA, HellaSwag, WinoGrande,
433
+ ARC, OBQA, and CSQA to assess common sense reasoning. SquAD, QuAC, and BoolQ
434
+ were incorporated to evaluate reading comprehension. CSQA was exclusively tested
435
+ using a 7-shot setup, while all other tests were conducted with a 0-shot
436
+ configuration. Additionally, we introduced GSM8K (8-shot@1), MATH (4-shot@1),
437
+ HumanEval (0-shot@1), and MBPP (3-shot@1) under the category "Math & Code". Due
438
+ to technical constraints, we did not test Falcon-180 on QuAC and OBQA; the score
439
+ is derived by averaging the scores on the remaining tasks. Since the scores for
440
+ these two tasks are generally lower than the average, we believe that
441
+ Falcon-180B's performance was not underestimated.
442
+
443
+ ## Usage
444
+
445
+ Please visit our [github repository](https://github.com/01-ai/Yi) for general
446
+ guidance on how to use this model.
447
+
448
+ ## Disclaimer
449
+
450
+ Although we use data compliance checking algorithms during the training process
451
+ to ensure the compliance of the trained model to the best of our ability, due to
452
+ the complexity of the data and the diversity of language model usage scenarios,
453
+ we cannot guarantee that the model will generate correct and reasonable output
454
+ in all scenarios. Please be aware that there is still a risk of the model
455
+ producing problematic outputs. We will not be responsible for any risks and
456
+ issues resulting from misuse, misguidance, illegal usage, and related
457
+ misinformation, as well as any associated data security concerns.
458
+
459
+ ## License
460
+
461
+ The Yi series models are fully open for academic research and free commercial
462
+ usage with permission via applications. All usage must adhere to the [Model
463
+ License Agreement 2.0](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE). To
464
+ apply for the official commercial license, please contact us
465