jartine commited on
Commit
b950db1
1 Parent(s): 9f71e7e

Add README.md to repo

Browse files
Files changed (1) hide show
  1. README.md +47 -57
README.md CHANGED
@@ -31,38 +31,36 @@ quantized_by: TheBloke
31
  <!-- header start -->
32
  <!-- 200823 -->
33
  <div style="width: auto; margin-left: auto; margin-right: auto">
34
- <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
35
  </div>
36
  <div style="display: flex; justify-content: space-between; width: 100%;">
37
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
38
- <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
39
  </div>
40
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
41
- <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>
42
  </div>
43
  </div>
44
- <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>
45
  <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
46
  <!-- header end -->
47
 
48
- # Nous Hermes Llama 2 13B - GGUF
49
  - Model creator: [NousResearch](https://huggingface.co/NousResearch)
50
  - Original model: [Nous Hermes Llama 2 13B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b)
51
 
52
  <!-- description start -->
53
  ## Description
54
 
55
- This repo contains GGUF format model files for [Nous Research's Nous Hermes Llama 2 13B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b).
56
 
57
- <!-- description end -->
58
- <!-- README_GGUF.md-about-gguf start -->
59
- ### About GGUF
60
 
61
- GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp. GGUF offers numerous advantages over GGML, such as better tokenisation, and support for special tokens. It is also supports metadata, and is designed to be extensible.
62
 
63
- Here is an incomplate list of clients and libraries that are known to support GGUF:
64
 
65
- * [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for GGUF. Offers a CLI and a server option.
66
  * [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
67
  * [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
68
  * [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
@@ -72,13 +70,13 @@ Here is an incomplate list of clients and libraries that are known to support GG
72
  * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
73
  * [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
74
 
75
- <!-- README_GGUF.md-about-gguf end -->
76
  <!-- repositories-available start -->
77
  ## Repositories available
78
 
79
- * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-AWQ)
80
- * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GPTQ)
81
- * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF)
82
  * [NousResearch's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b)
83
  <!-- repositories-available end -->
84
 
@@ -105,10 +103,10 @@ As this model is based on Llama 2, it is also subject to the Meta Llama 2 licens
105
 
106
  In the meantime, any questions regarding licensing, and in particular how these two licenses might interact, should be directed to the original model repository: [Nous Research's Nous Hermes Llama 2 13B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b).
107
  <!-- licensing end -->
108
- <!-- compatibility_gguf start -->
109
  ## Compatibility
110
 
111
- These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d36d5be95a0d9088b674dbb27354107221](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
112
 
113
  They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
114
 
@@ -125,34 +123,34 @@ The new methods available are:
125
 
126
  Refer to the Provided Files table below to see what files use which methods, and how.
127
  </details>
128
- <!-- compatibility_gguf end -->
129
 
130
- <!-- README_GGUF.md-provided-files start -->
131
  ## Provided files
132
 
133
  | Name | Quant method | Bits | Size | Max RAM required | Use case |
134
  | ---- | ---- | ---- | ---- | ---- | ----- |
135
- | [nous-hermes-llama2-13b.Q2_K.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q2_K.gguf) | Q2_K | 2 | 5.43 GB| 7.93 GB | smallest, significant quality loss - not recommended for most purposes |
136
- | [nous-hermes-llama2-13b.Q3_K_S.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q3_K_S.gguf) | Q3_K_S | 3 | 5.66 GB| 8.16 GB | very small, high quality loss |
137
- | [nous-hermes-llama2-13b.Q3_K_M.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q3_K_M.gguf) | Q3_K_M | 3 | 6.34 GB| 8.84 GB | very small, high quality loss |
138
- | [nous-hermes-llama2-13b.Q3_K_L.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q3_K_L.gguf) | Q3_K_L | 3 | 6.93 GB| 9.43 GB | small, substantial quality loss |
139
- | [nous-hermes-llama2-13b.Q4_0.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q4_0.gguf) | Q4_0 | 4 | 7.37 GB| 9.87 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
140
- | [nous-hermes-llama2-13b.Q4_K_S.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q4_K_S.gguf) | Q4_K_S | 4 | 7.41 GB| 9.91 GB | small, greater quality loss |
141
- | [nous-hermes-llama2-13b.Q4_K_M.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q4_K_M.gguf) | Q4_K_M | 4 | 7.87 GB| 10.37 GB | medium, balanced quality - recommended |
142
- | [nous-hermes-llama2-13b.Q5_0.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q5_0.gguf) | Q5_0 | 5 | 8.97 GB| 11.47 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
143
- | [nous-hermes-llama2-13b.Q5_K_S.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q5_K_S.gguf) | Q5_K_S | 5 | 8.97 GB| 11.47 GB | large, low quality loss - recommended |
144
- | [nous-hermes-llama2-13b.Q5_K_M.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q5_K_M.gguf) | Q5_K_M | 5 | 9.23 GB| 11.73 GB | large, very low quality loss - recommended |
145
- | [nous-hermes-llama2-13b.Q6_K.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q6_K.gguf) | Q6_K | 6 | 10.68 GB| 13.18 GB | very large, extremely low quality loss |
146
- | [nous-hermes-llama2-13b.Q8_0.gguf](https://huggingface.co/TheBloke/Nous-Hermes-Llama2-GGUF/blob/main/nous-hermes-llama2-13b.Q8_0.gguf) | Q8_0 | 8 | 13.83 GB| 16.33 GB | very large, extremely low quality loss - not recommended |
147
 
148
  **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.
149
 
150
 
151
 
152
- <!-- README_GGUF.md-provided-files end -->
153
 
154
- <!-- README_GGUF.md-how-to-download start -->
155
- ## How to download GGUF files
156
 
157
  **Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
158
 
@@ -163,7 +161,7 @@ The following clients/libraries will automatically download models for you, prov
163
 
164
  ### In `text-generation-webui`
165
 
166
- Under Download Model, you can enter the model repo: TheBloke/Nous-Hermes-Llama2-GGUF and below it, a specific filename to download, such as: nous-hermes-llama2-13b.q4_K_M.gguf.
167
 
168
  Then click Download.
169
 
@@ -178,7 +176,7 @@ pip3 install huggingface-hub>=0.17.1
178
  Then you can download any individual model file to the current directory, at high speed, with a command like this:
179
 
180
  ```shell
181
- huggingface-cli download TheBloke/Nous-Hermes-Llama2-GGUF nous-hermes-llama2-13b.q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
182
  ```
183
 
184
  <details>
@@ -187,7 +185,7 @@ huggingface-cli download TheBloke/Nous-Hermes-Llama2-GGUF nous-hermes-llama2-13b
187
  You can also download multiple files at once with a pattern:
188
 
189
  ```shell
190
- huggingface-cli download TheBloke/Nous-Hermes-Llama2-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
191
  ```
192
 
193
  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).
@@ -201,25 +199,25 @@ pip3 install hf_transfer
201
  And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
202
 
203
  ```shell
204
- HUGGINGFACE_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/Nous-Hermes-Llama2-GGUF nous-hermes-llama2-13b.q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
205
  ```
206
 
207
  Windows CLI users: Use `set HUGGINGFACE_HUB_ENABLE_HF_TRANSFER=1` before running the download command.
208
  </details>
209
- <!-- README_GGUF.md-how-to-download end -->
210
 
211
- <!-- README_GGUF.md-how-to-run start -->
212
  ## Example `llama.cpp` command
213
 
214
  Make sure you are using `llama.cpp` from commit [d0cee0d36d5be95a0d9088b674dbb27354107221](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
215
 
216
  ```shell
217
- ./main -ngl 32 -m nous-hermes-llama2-13b.q4_K_M.gguf --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{prompt}\n\n### Response:"
218
  ```
219
 
220
  Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
221
 
222
- Change `-c 4096` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.
223
 
224
  If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
225
 
@@ -231,7 +229,7 @@ Further instructions here: [text-generation-webui/docs/llama.cpp.md](https://git
231
 
232
  ## How to run from Python code
233
 
234
- You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries.
235
 
236
  ### How to load this model from Python using ctransformers
237
 
@@ -248,13 +246,13 @@ CT_HIPBLAS=1 pip install ctransformers>=0.2.24 --no-binary ctransformers
248
  CT_METAL=1 pip install ctransformers>=0.2.24 --no-binary ctransformers
249
  ```
250
 
251
- #### Simple example code to load one of these GGUF models
252
 
253
  ```python
254
  from ctransformers import AutoModelForCausalLM
255
 
256
  # Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
257
- llm = AutoModelForCausalLM.from_pretrained("TheBloke/Nous-Hermes-Llama2-GGUF", model_file="nous-hermes-llama2-13b.q4_K_M.gguf", model_type="llama", gpu_layers=50)
258
 
259
  print(llm("AI is going to"))
260
  ```
@@ -266,7 +264,7 @@ Here's guides on using llama-cpp-python or ctransformers with LangChain:
266
  * [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
267
  * [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
268
 
269
- <!-- README_GGUF.md-how-to-run end -->
270
 
271
  <!-- footer start -->
272
  <!-- 200823 -->
@@ -274,31 +272,23 @@ Here's guides on using llama-cpp-python or ctransformers with LangChain:
274
 
275
  For further support, and discussions on these models and AI in general, join us at:
276
 
277
- [TheBloke AI's Discord server](https://discord.gg/theblokeai)
278
 
279
  ## Thanks, and how to contribute
280
 
281
- Thanks to the [chirper.ai](https://chirper.ai) team!
282
 
283
- Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
284
 
285
  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.
286
 
287
  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.
288
 
289
- Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
290
 
291
- * Patreon: https://patreon.com/TheBlokeAI
292
- * Ko-Fi: https://ko-fi.com/TheBlokeAI
293
 
294
- **Special thanks to**: Aemon Algiz.
295
 
296
- **Patreon special mentions**: Alicia Loh, Stephen Murray, K, Ajan Kanaga, RoA, Magnesian, Deo Leter, Olakabola, Eugene Pentland, zynix, Deep Realms, Raymond Fosdick, Elijah Stavena, Iucharbius, Erik Bjäreholt, Luis Javier Navarrete Lozano, Nicholas, theTransient, John Detwiler, alfie_i, knownsqashed, Mano Prime, Willem Michiel, Enrico Ros, LangChain4j, OG, Michael Dempsey, Pierre Kircher, Pedro Madruga, James Bentley, Thomas Belote, Luke @flexchar, Leonard Tan, Johann-Peter Hartmann, Illia Dulskyi, Fen Risland, Chadd, S_X, Jeff Scroggin, Ken Nordquist, Sean Connelly, Artur Olbinski, Swaroop Kallakuri, Jack West, Ai Maven, David Ziegler, Russ Johnson, transmissions 11, John Villwock, Alps Aficionado, Clay Pascal, Viktor Bowallius, Subspace Studios, Rainer Wilmers, Trenton Dambrowitz, vamX, Michael Levine, 준교 김, Brandon Frisco, Kalila, Trailburnt, Randy H, Talal Aujan, Nathan Dryer, Vadim, 阿明, ReadyPlayerEmma, Tiffany J. Kim, George Stoitzev, Spencer Kim, Jerry Meng, Gabriel Tamborski, Cory Kujawski, Jeffrey Morgan, Spiking Neurons AB, Edmond Seymore, Alexandros Triantafyllidis, Lone Striker, Cap'n Zoog, Nikolai Manek, danny, ya boyyy, Derek Yates, usrbinkat, Mandus, TL, Nathan LeClaire, subjectnull, Imad Khwaja, webtim, Raven Klaugh, Asp the Wyvern, Gabriel Puliatti, Caitlyn Gatomon, Joseph William Delisle, Jonathan Leane, Luke Pendergrass, SuperWojo, Sebastain Graf, Will Dee, Fred von Graf, Andrey, Dan Guido, Daniel P. Andersen, Nitin Borwankar, Elle, Vitor Caleffi, biorpg, jjj, NimbleBox.ai, Pieter, Matthew Berman, terasurfer, Michael Davis, Alex, Stanislav Ovsiannikov
297
 
298
 
299
- Thank you to all my generous patrons and donaters!
300
 
301
- And thank you again to a16z for their generous grant.
302
 
303
  <!-- footer end -->
304
 
 
31
  <!-- header start -->
32
  <!-- 200823 -->
33
  <div style="width: auto; margin-left: auto; margin-right: auto">
 
34
  </div>
35
  <div style="display: flex; justify-content: space-between; width: 100%;">
36
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
37
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/FwAVVu7eJ4">Chat & support: jartine's Discord server</a></p>
38
  </div>
39
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
 
40
  </div>
41
  </div>
42
+ <div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">jartine's LLM work is generously supported by a grant from <a href="https://mozilla.org">mozilla</a></p></div>
43
  <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
44
  <!-- header end -->
45
 
46
+ # Nous Hermes Llama 2 13B - llamafile
47
  - Model creator: [NousResearch](https://huggingface.co/NousResearch)
48
  - Original model: [Nous Hermes Llama 2 13B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b)
49
 
50
  <!-- description start -->
51
  ## Description
52
 
53
+ This repo contains llamafile format model files for [Nous Research's Nous Hermes Llama 2 13B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b).
54
 
55
+ WARNING: This README may contain inaccuracies. It was generated automatically by forking <a href=/TheBloke/Nous-Hermes-Llama2-GGUF>TheBloke/Nous-Hermes-Llama2-GGUF</a> and piping the README through sed. Errors should be reported to jartine, and do not reflect TheBloke. You can support his work on [Patreon](https://www.patreon.com/TheBlokeAI).
56
+ <!-- README_llamafile.md-about-llamafile start -->
57
+ ### About llamafile
58
 
59
+ llamafile is a new format introduced by Mozilla Ocho on Nov 20th 2023. It uses Cosmopolitan Libc to turn LLM weights into runnable llama.cpp binaries that run on the stock installs of six OSes for both ARM64 and AMD64. llamafile offers numerous advantages over GGML, such as better tokenisation, and support for special tokens. It is also supports metadata, and is designed to be extensible.
60
 
61
+ Here is an incomplate list of clients and libraries that are known to support llamafile:
62
 
63
+ * [llama.cpp](https://github.com/ggerganov/llama.cpp). The source project for llamafile. Offers a CLI and a server option.
64
  * [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
65
  * [KoboldCpp](https://github.com/LostRuins/koboldcpp), a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
66
  * [LM Studio](https://lmstudio.ai/), an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
 
70
  * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
71
  * [candle](https://github.com/huggingface/candle), a Rust ML framework with a focus on performance, including GPU support, and ease of use.
72
 
73
+ <!-- README_llamafile.md-about-llamafile end -->
74
  <!-- repositories-available start -->
75
  ## Repositories available
76
 
77
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/jartine/Nous-Hermes-Llama2-AWQ)
78
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/jartine/Nous-Hermes-Llama2-GPTQ)
79
+ * [2, 3, 4, 5, 6 and 8-bit llamafile models for CPU+GPU inference](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile)
80
  * [NousResearch's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b)
81
  <!-- repositories-available end -->
82
 
 
103
 
104
  In the meantime, any questions regarding licensing, and in particular how these two licenses might interact, should be directed to the original model repository: [Nous Research's Nous Hermes Llama 2 13B](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b).
105
  <!-- licensing end -->
106
+ <!-- compatibility_llamafile start -->
107
  ## Compatibility
108
 
109
+ These quantised llamafilev2 files are compatible with llama.cpp from August 27th onwards, as of commit [d0cee0d36d5be95a0d9088b674dbb27354107221](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221)
110
 
111
  They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
112
 
 
123
 
124
  Refer to the Provided Files table below to see what files use which methods, and how.
125
  </details>
126
+ <!-- compatibility_llamafile end -->
127
 
128
+ <!-- README_llamafile.md-provided-files start -->
129
  ## Provided files
130
 
131
  | Name | Quant method | Bits | Size | Max RAM required | Use case |
132
  | ---- | ---- | ---- | ---- | ---- | ----- |
133
+ | [nous-hermes-llama2-13b.Q2_K.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q2_K.llamafile) | Q2_K | 2 | 5.43 GB| 7.93 GB | smallest, significant quality loss - not recommended for most purposes |
134
+ | [nous-hermes-llama2-13b.Q3_K_S.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q3_K_S.llamafile) | Q3_K_S | 3 | 5.66 GB| 8.16 GB | very small, high quality loss |
135
+ | [nous-hermes-llama2-13b.Q3_K_M.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q3_K_M.llamafile) | Q3_K_M | 3 | 6.34 GB| 8.84 GB | very small, high quality loss |
136
+ | [nous-hermes-llama2-13b.Q3_K_L.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q3_K_L.llamafile) | Q3_K_L | 3 | 6.93 GB| 9.43 GB | small, substantial quality loss |
137
+ | [nous-hermes-llama2-13b.Q4_0.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q4_0.llamafile) | Q4_0 | 4 | 7.37 GB| 9.87 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
138
+ | [nous-hermes-llama2-13b.Q4_K_S.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q4_K_S.llamafile) | Q4_K_S | 4 | 7.41 GB| 9.91 GB | small, greater quality loss |
139
+ | [nous-hermes-llama2-13b.Q4_K_M.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q4_K_M.llamafile) | Q4_K_M | 4 | 7.87 GB| 10.37 GB | medium, balanced quality - recommended |
140
+ | [nous-hermes-llama2-13b.Q5_0.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q5_0.llamafile) | Q5_0 | 5 | 8.97 GB| 11.47 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
141
+ | [nous-hermes-llama2-13b.Q5_K_S.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q5_K_S.llamafile) | Q5_K_S | 5 | 8.97 GB| 11.47 GB | large, low quality loss - recommended |
142
+ | [nous-hermes-llama2-13b.Q5_K_M.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q5_K_M.llamafile) | Q5_K_M | 5 | 9.23 GB| 11.73 GB | large, very low quality loss - recommended |
143
+ | [nous-hermes-llama2-13b.Q6_K.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q6_K.llamafile) | Q6_K | 6 | 10.68 GB| 13.18 GB | very large, extremely low quality loss |
144
+ | [nous-hermes-llama2-13b.Q8_0.llamafile](https://huggingface.co/jartine/Nous-Hermes-Llama2-llamafile/blob/main/nous-hermes-llama2-13b.Q8_0.llamafile) | Q8_0 | 8 | 13.83 GB| 16.33 GB | very large, extremely low quality loss - not recommended |
145
 
146
  **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.
147
 
148
 
149
 
150
+ <!-- README_llamafile.md-provided-files end -->
151
 
152
+ <!-- README_llamafile.md-how-to-download start -->
153
+ ## How to download llamafile files
154
 
155
  **Note for manual downloaders:** You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
156
 
 
161
 
162
  ### In `text-generation-webui`
163
 
164
+ Under Download Model, you can enter the model repo: jartine/Nous-Hermes-Llama2-llamafile and below it, a specific filename to download, such as: nous-hermes-llama2-13b.q4_K_M.llamafile.
165
 
166
  Then click Download.
167
 
 
176
  Then you can download any individual model file to the current directory, at high speed, with a command like this:
177
 
178
  ```shell
179
+ huggingface-cli download jartine/Nous-Hermes-Llama2-llamafile nous-hermes-llama2-13b.q4_K_M.llamafile --local-dir . --local-dir-use-symlinks False
180
  ```
181
 
182
  <details>
 
185
  You can also download multiple files at once with a pattern:
186
 
187
  ```shell
188
+ huggingface-cli download jartine/Nous-Hermes-Llama2-llamafile --local-dir . --local-dir-use-symlinks False --include='*Q4_K*llamafile'
189
  ```
190
 
191
  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).
 
199
  And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
200
 
201
  ```shell
202
+ HUGGINGFACE_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download jartine/Nous-Hermes-Llama2-llamafile nous-hermes-llama2-13b.q4_K_M.llamafile --local-dir . --local-dir-use-symlinks False
203
  ```
204
 
205
  Windows CLI users: Use `set HUGGINGFACE_HUB_ENABLE_HF_TRANSFER=1` before running the download command.
206
  </details>
207
+ <!-- README_llamafile.md-how-to-download end -->
208
 
209
+ <!-- README_llamafile.md-how-to-run start -->
210
  ## Example `llama.cpp` command
211
 
212
  Make sure you are using `llama.cpp` from commit [d0cee0d36d5be95a0d9088b674dbb27354107221](https://github.com/ggerganov/llama.cpp/commit/d0cee0d36d5be95a0d9088b674dbb27354107221) or later.
213
 
214
  ```shell
215
+ ./main -ngl 32 -m nous-hermes-llama2-13b.q4_K_M.llamafile --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{prompt}\n\n### Response:"
216
  ```
217
 
218
  Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
219
 
220
+ Change `-c 4096` to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the llamafile file and set by llama.cpp automatically.
221
 
222
  If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
223
 
 
229
 
230
  ## How to run from Python code
231
 
232
+ You can use llamafile models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) or [ctransformers](https://github.com/marella/ctransformers) libraries.
233
 
234
  ### How to load this model from Python using ctransformers
235
 
 
246
  CT_METAL=1 pip install ctransformers>=0.2.24 --no-binary ctransformers
247
  ```
248
 
249
+ #### Simple example code to load one of these llamafile models
250
 
251
  ```python
252
  from ctransformers import AutoModelForCausalLM
253
 
254
  # Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
255
+ llm = AutoModelForCausalLM.from_pretrained("jartine/Nous-Hermes-Llama2-llamafile", model_file="nous-hermes-llama2-13b.q4_K_M.llamafile", model_type="llama", gpu_layers=50)
256
 
257
  print(llm("AI is going to"))
258
  ```
 
264
  * [LangChain + llama-cpp-python](https://python.langchain.com/docs/integrations/llms/llamacpp)
265
  * [LangChain + ctransformers](https://python.langchain.com/docs/integrations/providers/ctransformers)
266
 
267
+ <!-- README_llamafile.md-how-to-run end -->
268
 
269
  <!-- footer start -->
270
  <!-- 200823 -->
 
272
 
273
  For further support, and discussions on these models and AI in general, join us at:
274
 
275
+ [jartine AI's Discord server](https://discord.gg/FwAVVu7eJ4)
276
 
277
  ## Thanks, and how to contribute
278
 
 
279
 
 
280
 
281
  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.
282
 
283
  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.
284
 
 
285
 
 
 
286
 
 
287
 
 
288
 
289
 
 
290
 
291
+ And thank you again to mozilla for their generous grant.
292
 
293
  <!-- footer end -->
294