TheBloke commited on
Commit
eed958a
1 Parent(s): 8103211

Initial GGML model commit

Browse files
Files changed (1) hide show
  1. README.md +257 -0
README.md ADDED
@@ -0,0 +1,257 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ inference: false
3
+ license: other
4
+ ---
5
+
6
+ <!-- header start -->
7
+ <div style="width: 100%;">
8
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
9
+ </div>
10
+ <div style="display: flex; justify-content: space-between; width: 100%;">
11
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
12
+ <p><a href="https://discord.gg/theblokeai">Chat & support: my new Discord server</a></p>
13
+ </div>
14
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
15
+ <p><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
16
+ </div>
17
+ </div>
18
+ <!-- header end -->
19
+
20
+ # TehVenom's merge of PygmalionAI's Pygmalion 13B GGML
21
+
22
+ These files are GGML format model files for [TehVenom's merge of PygmalionAI's Pygmalion 13B](https://huggingface.co/TehVenom/Pygmalion-13b-Merged).
23
+
24
+ These are SuperHOT GGMLs with an increased context length. SuperHOT is a new system that employs RoPE to expand context beyond what was originally possible for a model. It was discovered and developed by [kaiokendev](https://huggingface.co/kaiokendev).
25
+
26
+ In order to use the increased context length, you can presently use:
27
+ * [KoboldCpp](https://github.com/LostRuins/koboldcpp) - [release 1.33](https://github.com/LostRuins/koboldcpp/releases/tag/v1.33) or later.
28
+
29
+ Support is also expected to come to llama.cpp, however it is still being worked on and there is currently no ETA for that.
30
+
31
+ To use the increased context with KoboldCpp and (when supported) llama.cpp, simply use `--contextsize` to set the desired context, eg `--contextsize 4096` or `--contextsize 8192`.
32
+
33
+ ## Repositories available
34
+
35
+ * [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/Pygmalion-13B-SuperHOT-8K-GPTQ)
36
+ * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU inference](https://huggingface.co/TheBloke/Pygmalion-13B-SuperHOT-8K-GGML)
37
+ * [Unquantised SuperHOT fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/TheBloke/Pygmalion-13B-SuperHOT-8K-fp16)
38
+ * [Unquantised base fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/PygmalionAI/pygmalion-13b)
39
+
40
+ <!-- compatibility_ggml start -->
41
+ ## Compatibility
42
+
43
+ These GGMLs will work with any llama.cpp-compatible GGML client that supports k-quants.
44
+
45
+ However the increased context length won't work without specific support. See the note in the introduction for details on using increased context.
46
+
47
+ ## Explanation of the new k-quant methods
48
+
49
+ The new methods available are:
50
+ * 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)
51
+ * 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.
52
+ * 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.
53
+ * GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
54
+ * 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
55
+ * GGML_TYPE_Q8_K - "type-0" 8-bit quantization. Only used for quantizing intermediate results. The difference to the existing Q8_0 is that the block size is 256. All 2-6 bit dot products are implemented for this quantization type.
56
+
57
+ Refer to the Provided Files table below to see what files use which methods, and how.
58
+ <!-- compatibility_ggml end -->
59
+
60
+ ## Provided files
61
+ | Name | Quant method | Bits | Size | Max RAM required | Use case |
62
+ | ---- | ---- | ---- | ---- | ---- | ----- |
63
+ | pygmalion-13b-superhot-8k.ggmlv3.q2_K.bin | q2_K | 2 | 5.51 GB | 8.01 GB | New k-quant method. Uses GGML_TYPE_Q4_K for the attention.vw and feed_forward.w2 tensors, GGML_TYPE_Q2_K for the other tensors. |
64
+ | pygmalion-13b-superhot-8k.ggmlv3.q3_K_L.bin | q3_K_L | 3 | 6.93 GB | 9.43 GB | New k-quant method. Uses GGML_TYPE_Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K |
65
+ | pygmalion-13b-superhot-8k.ggmlv3.q3_K_M.bin | q3_K_M | 3 | 6.31 GB | 8.81 GB | New k-quant method. Uses GGML_TYPE_Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K |
66
+ | pygmalion-13b-superhot-8k.ggmlv3.q3_K_S.bin | q3_K_S | 3 | 5.66 GB | 8.16 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors |
67
+ | pygmalion-13b-superhot-8k.ggmlv3.q4_K_M.bin | q4_K_M | 4 | 7.87 GB | 10.37 GB | New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q4_K |
68
+ | pygmalion-13b-superhot-8k.ggmlv3.q4_K_S.bin | q4_K_S | 4 | 7.37 GB | 9.87 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors |
69
+ | pygmalion-13b-superhot-8k.ggmlv3.q5_K_M.bin | q5_K_M | 5 | 9.23 GB | 11.73 GB | New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q5_K |
70
+ | pygmalion-13b-superhot-8k.ggmlv3.q5_K_S.bin | q5_K_S | 5 | 8.97 GB | 11.47 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
71
+ | pygmalion-13b-superhot-8k.ggmlv3.q6_K.bin | q6_K | 6 | 10.68 GB | 13.18 GB | New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors |
72
+
73
+ **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.
74
+
75
+ ## How to run in `koboldcpp`
76
+
77
+ On Linux I use the following command line to launch the KoboldCpp UI with OpenCL aceleration and a context size of 4096:
78
+
79
+ ```
80
+ python ./koboldcpp.py --stream --unbantokens --threads 8 --usecublas 100 pygmalion-13b-superhot-8k.ggmlv3.q5_0.bin
81
+ ```
82
+
83
+ Change `--gpulayers 100` to the number of layers you want/are able to offload to the GPU. Remove it if you don't have GPU acceleration.
84
+
85
+ For OpenCL acceleration, change `--usecublas` to `--useclblast 0 0`. You may need to change the second `0` to `1` if you have both an iGPU and a discrete GPU.
86
+
87
+ <!-- footer start -->
88
+ ## Discord
89
+
90
+ For further support, and discussions on these models and AI in general, join us at:
91
+
92
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
93
+
94
+ ## Thanks, and how to contribute.
95
+
96
+ Thanks to the [chirper.ai](https://chirper.ai) team!
97
+
98
+ 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.
99
+
100
+ 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.
101
+
102
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
103
+
104
+ * Patreon: https://patreon.com/TheBlokeAI
105
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
106
+
107
+ **Special thanks to**: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.
108
+
109
+ **Patreon special mentions**: zynix , ya boyyy, Trenton Dambrowitz, Imad Khwaja, Alps Aficionado, chris gileta, John Detwiler, Willem Michiel, RoA, Mano Prime, Rainer Wilmers, Fred von Graf, Matthew Berman, Ghost , Nathan LeClaire, Iucharbius , Ai Maven, Illia Dulskyi, Joseph William Delisle, Space Cruiser, Lone Striker, Karl Bernard, Eugene Pentland, Greatston Gnanesh, Jonathan Leane, Randy H, Pierre Kircher, Willian Hasse, Stephen Murray, Alex , terasurfer , Edmond Seymore, Oscar Rangel, Luke Pendergrass, Asp the Wyvern, Junyu Yang, David Flickinger, Luke, Spiking Neurons AB, subjectnull, Pyrater, Nikolai Manek, senxiiz, Ajan Kanaga, Johann-Peter Hartmann, Artur Olbinski, Kevin Schuppel, Derek Yates, Kalila, K, Talal Aujan, Khalefa Al-Ahmad, Gabriel Puliatti, John Villwock, WelcomeToTheClub, Daniel P. Andersen, Preetika Verma, Deep Realms, Fen Risland, trip7s trip, webtim, Sean Connelly, Michael Levine, Chris McCloskey, biorpg, vamX, Viktor Bowallius, Cory Kujawski.
110
+
111
+ Thank you to all my generous patrons and donaters!
112
+
113
+ <!-- footer end -->
114
+
115
+ # Original model card: Kaio Ken's SuperHOT 8K
116
+
117
+ ### SuperHOT Prototype 2 w/ 8K Context
118
+
119
+ This is a second prototype of SuperHOT, this time 30B with 8K context and no RLHF, using the same technique described in [the github blog](https://kaiokendev.github.io/til#extending-context-to-8k).
120
+ Tests have shown that the model does indeed leverage the extended context at 8K.
121
+
122
+ You will need to **use either the monkeypatch** or, if you are already using the monkeypatch, **change the scaling factor to 0.25 and the maximum sequence length to 8192**
123
+
124
+ #### Looking for Merged & Quantized Models?
125
+ - 30B 4-bit CUDA: [tmpupload/superhot-30b-8k-4bit-safetensors](https://huggingface.co/tmpupload/superhot-30b-8k-4bit-safetensors)
126
+ - 30B 4-bit CUDA 128g: [tmpupload/superhot-30b-8k-4bit-128g-safetensors](https://huggingface.co/tmpupload/superhot-30b-8k-4bit-128g-safetensors)
127
+
128
+
129
+ #### Training Details
130
+ I trained the LoRA with the following configuration:
131
+ - 1200 samples (~400 samples over 2048 sequence length)
132
+ - learning rate of 3e-4
133
+ - 3 epochs
134
+ - The exported modules are:
135
+ - q_proj
136
+ - k_proj
137
+ - v_proj
138
+ - o_proj
139
+ - no bias
140
+ - Rank = 4
141
+ - Alpha = 8
142
+ - no dropout
143
+ - weight decay of 0.1
144
+ - AdamW beta1 of 0.9 and beta2 0.99, epsilon of 1e-5
145
+ - Trained on 4-bit base model
146
+
147
+ # Original model card: TehVenom's merge of PygmalionAI's Pygmalion 13B
148
+
149
+ <h1 style="text-align: center">Pygmalion 13b</h1>
150
+ <h2 style="text-align: center">A conversational LLaMA fine-tune.</h2>
151
+
152
+ ## Model Details:
153
+
154
+ Pygmalion 13b is a dialogue model based on Meta's LLaMA-13b.
155
+
156
+ This is version 1. It has been fine-tuned using a subset of the data from Pygmalion-6B-v8-pt4,
157
+ for those of you familiar with the project.
158
+
159
+ The current Pygmalion-13b has been trained as a LoRA, then merged down to the base model for distribuition.
160
+
161
+ ## Applying the XORs
162
+
163
+ This models has the XOR files pre-applied out of the box.
164
+ Converted from the XORs weights from PygmalionAI's release https://huggingface.co/PygmalionAI/pygmalion-13b
165
+
166
+ ## Prompting
167
+
168
+ The model was trained on the usual Pygmalion persona + chat format, so any of the usual UIs should already handle everything correctly. If you're using the model directly, this is the expected formatting:
169
+
170
+ ```
171
+ [CHARACTER]'s Persona: [A few sentences about the character you want the model to play]
172
+ <START>
173
+ [DIALOGUE HISTORY]
174
+ You: [User's input message here]
175
+ [CHARACTER]:
176
+ ```
177
+
178
+ Where `[CHARACTER]` is, as you can probably guess, the name of the character you want the model to portray, `<START>` should be used verbatim as a delimiter token to separate persona and scenario data from the dialogue, and `[DIALOGUE HISTORY]` is a sliding window of chat history so the model can have conversational context to draw from. Here's a concrete example:
179
+
180
+ ```
181
+ Assistant's Persona: Assistant is a highly intelligent language model trained to comply with user requests.
182
+ <START>
183
+ Assistant: Hello! How may I help you today?
184
+ You: What is Zork?
185
+ Assistant:
186
+ ```
187
+
188
+ Which will generate something like:
189
+
190
+ ```
191
+ Zork is an interactive fiction computer game created in the 1970s by Infocom, Inc., which was later acquired by Activision Blizzard. It is widely considered one of the most influential games ever made and has been credited with popularizing text-based adventure games. The original version of Zork was written in the programming language MACRO-10, but it was ported to many other platforms over the years."
192
+ ```
193
+
194
+ The model will automatically emit an end-of-text token (`</s>`) when it judges that the response is complete.
195
+
196
+ ## Eval / Benchmark scores
197
+
198
+
199
+ Current evals out of the Pygmalion-13b model: <br>
200
+ <html>
201
+ <head>
202
+ <style>
203
+ table {
204
+ border:1px solid #b3adad;
205
+ border-collapse:collapse;
206
+ padding:5px;
207
+ }
208
+ table th {
209
+ border:1px solid #b3adad;
210
+ padding:5px;
211
+ background: #f0f0f0;
212
+ color: #313030;
213
+ }
214
+ table td {
215
+ border:1px solid #b3adad;
216
+ text-align:center;
217
+ padding:5px;
218
+ background: #ffffff;
219
+ color: #313030;
220
+ }
221
+ </style>
222
+ </head>
223
+ <body>
224
+ <table>
225
+ <thead>
226
+ <tr>
227
+ <th>Model:</th>
228
+ <th>Wikitext2</th>
229
+ <th>Ptb-New</th>
230
+ <th>C4-New</th>
231
+ </tr>
232
+ </thead>
233
+ <tbody>
234
+ <tr>
235
+ <td>Pygmalion 13b - 16bit</td>
236
+ <td>5.710726737976074</td>
237
+ <td>23.633684158325195</td>
238
+ <td>7.6324849128723145</td>
239
+ </tr>
240
+ </tbody>
241
+ </table>
242
+ </body>
243
+ </html>
244
+ <br>Thanks to YellowRose#1776 for the numbers.
245
+ <hr>
246
+
247
+ ## Other notes
248
+
249
+ - When prompted correctly, the model will always start by generating a BOS token. This behavior is an accidental side-effect which we plan to address in future model versions and should not be relied upon.
250
+ - The model was trained as a LoRA with a somewhat unorthodox configuration which causes errors when used with the current version of `peft`, hence we release it as a full model instead.
251
+
252
+
253
+ ## Limitations and biases
254
+
255
+ The intended use-case for this model is fictional conversation for entertainment purposes. Any other sort of usage is out of scope.
256
+
257
+ As such, it was **not** fine-tuned to be safe and harmless: the base model _and_ this fine-tune have been trained on data known to contain profanity and texts that are lewd or otherwise offensive. It may produce socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive. Outputs might often be factually wrong or misleading.