TheBloke commited on
Commit
6af643f
1 Parent(s): 4cf4235

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -39
README.md CHANGED
@@ -2,7 +2,7 @@
2
  datasets:
3
  - rombodawg/LosslessMegaCodeTrainingV2_1m_Evol_Uncensored
4
  - OpenAssistant/oasst1
5
- - ehartford/dolphin
6
  - argilla/databricks-dolly-15k-curated-multilingual
7
  inference: false
8
  language:
@@ -48,6 +48,8 @@ This repo contains GGML format model files for [OpenAssistant's Llama2 70B SFT v
48
 
49
  The GGML format has now been superseded by GGUF. As of August 21st 2023, [llama.cpp](https://github.com/ggerganov/llama.cpp) no longer supports GGML models. Third party clients and libraries are expected to still support it for a time, but many may also drop support.
50
 
 
 
51
  ### About GGML
52
 
53
  GPU acceleration is now available for Llama 2 70B GGML files, with both CUDA (NVidia) and Metal (macOS). The following clients/libraries are known to work with these files, including with GPU acceleration:
@@ -73,6 +75,7 @@ GPU acceleration is now available for Llama 2 70B GGML files, with both CUDA (NV
73
  <|im_start|>user
74
  {prompt}<|im_end|>
75
  <|im_start|>assistant
 
76
  ```
77
 
78
  <!-- compatibility_ggml start -->
@@ -121,38 +124,8 @@ Refer to the Provided Files table below to see what files use which methods, and
121
  | [llama2-70b-oasst-sft-v10.ggmlv3.Q5_0.bin](https://huggingface.co/TheBloke/Llama2-70B-OASST-SFT-v10-GGML/blob/main/llama2-70b-oasst-sft-v10.ggmlv3.Q5_0.bin) | Q5_0 | 5 | 47.43 GB| 49.93 GB | Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
122
  | [llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_S.bin](https://huggingface.co/TheBloke/Llama2-70B-OASST-SFT-v10-GGML/blob/main/llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_S.bin) | Q5_K_S | 5 | 47.74 GB| 50.24 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
123
  | [llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_M.bin](https://huggingface.co/TheBloke/Llama2-70B-OASST-SFT-v10-GGML/blob/main/llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_M.bin) | Q5_K_M | 5 | 49.03 GB| 51.53 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 |
124
- | llama2-70b-oasst-sft-v10.ggmlv3.q5_1.bin | q5_1 | 5 | 51.76 GB | 54.26 GB | Original quant method, 5-bit. Higher accuracy, slower inference than q5_0. |
125
- | llama2-70b-oasst-sft-v10.ggmlv3.q6_K.bin | q6_K | 6 | 56.59 GB | 59.09 GB | New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors |
126
- | llama2-70b-oasst-sft-v10.ggmlv3.q8_0.bin | q8_0 | 8 | 73.23 GB | 75.73 GB | Original llama.cpp quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users. |
127
-
128
- ### q5_1, q6_K and q8_0 files require expansion from archive
129
 
130
- **Note:** HF does not support uploading files larger than 50GB. Therefore I have uploaded the q6_K and q8_0 files as multi-part ZIP files. They are not compressed, they are just for storing a .bin file in two parts.
131
-
132
- <details>
133
- <summary>Click for instructions regarding q5_1, q6_K and q8_0 files</summary>
134
-
135
- ### q5_1
136
- Please download:
137
- * `llama2-70b-oasst-sft-v10.ggmlv3.q5_1.zip`
138
- * `llama2-70b-oasst-sft-v10.ggmlv3.q5_1.z01`
139
-
140
- ### q6_K
141
- Please download:
142
- * `llama2-70b-oasst-sft-v10.ggmlv3.q6_K.zip`
143
- * `llama2-70b-oasst-sft-v10.ggmlv3.q6_K.z01`
144
-
145
- ### q8_0
146
- Please download:
147
- * `llama2-70b-oasst-sft-v10.ggmlv3.q8_0.zip`
148
- * `llama2-70b-oasst-sft-v10.ggmlv3.q8_0.z01`
149
-
150
- Then extract the .zip archive. This will will expand both parts automatically. On Linux I found I had to use `7zip` - the basic `unzip` tool did not work. Example:
151
- ```
152
- sudo apt update -y && sudo apt install 7zip
153
- 7zz x llama2-70b-oasst-sft-v10.ggmlv3.q6_K.zip
154
- ```
155
- </details>
156
 
157
  ## How to run in `llama.cpp`
158
 
@@ -163,7 +136,7 @@ For compatibility with latest llama.cpp, please use GGUF files instead.
163
  I use the following command line; adjust for your tastes and needs:
164
 
165
  ```
166
- ./main -t 10 -ngl 40 -gqa 8 -m llama2-70b-oasst-sft-v10.ggmlv3.q4_K_M.bin --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system\nYou are a story writing assistant.<|im_end|>\n<|im_start|>user\nWrite a story about llamas<|im_end|>\n<|im_start|>assistant"
167
  ```
168
  Change `-t 10` to the number of physical CPU cores you have. For example if your system has 8 cores/16 threads, use `-t 8`. If you are fully offloading the model to GPU, use `-t 1`
169
 
@@ -204,7 +177,7 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
204
 
205
  **Special thanks to**: Aemon Algiz.
206
 
207
- **Patreon special mentions**: Kacper Wikieł, knownsqashed, Leonard Tan, Asp the Wyvern, Daniel P. Andersen, Luke Pendergrass, Stanislav Ovsiannikov, RoA, Dave, Ai Maven, Kalila, Will Dee, Imad Khwaja, Nitin Borwankar, Joseph William Delisle, Tony Hughes, Cory Kujawski, Rishabh Srivastava, Russ Johnson, Stephen Murray, Lone Striker, Johann-Peter Hartmann, Elle, J, Deep Realms, SuperWojo, Raven Klaugh, Sebastain Graf, ReadyPlayerEmma, Alps Aficionado, Mano Prime, Derek Yates, Gabriel Puliatti, Mesiah Bishop, Magnesian, Sean Connelly, biorpg, Iucharbius, Olakabola, Fen Risland, Space Cruiser, theTransient, Illia Dulskyi, Thomas Belote, Spencer Kim, Pieter, John Detwiler, Fred von Graf, Michael Davis, Swaroop Kallakuri, subjectnull, Clay Pascal, Subspace Studios, Chris Smitley, Enrico Ros, usrbinkat, Steven Wood, alfie_i, David Ziegler, Willem Michiel, Matthew Berman, Andrey, Pyrater, Jeffrey Morgan, vamX, LangChain4j, Luke @flexchar, Trenton Dambrowitz, Pierre Kircher, Alex, Sam, James Bentley, Edmond Seymore, Eugene Pentland, Pedro Madruga, Rainer Wilmers, Dan Guido, Nathan LeClaire, Spiking Neurons AB, Talal Aujan, zynix, Artur Olbinski, Michael Levine, 阿明, K, John Villwock, Nikolai Manek, Femi Adebogun, senxiiz, Deo Leter, NimbleBox.ai, Viktor Bowallius, Geoffrey Montalvo, Mandus, Ajan Kanaga, ya boyyy, Jonathan Leane, webtim, Brandon Frisco, danny, Alexandros Triantafyllidis, Gabriel Tamborski, Randy H, terasurfer, Vadim, Junyu Yang, Vitor Caleffi, Chadd, transmissions 11
208
 
209
 
210
  Thank you to all my generous patrons and donaters!
@@ -295,12 +268,13 @@ perform safety testing and tuning tailored to their specific applications of the
295
 
296
  Please see Meta's [Responsible Use Guide](https://ai.meta.com/llama/responsible-use-guide/).
297
 
298
- ## Note regarding inference with TGI
299
 
300
- During evaluation we noticed that this 70B model produced extremely poor outputs when loaded it was loaded in 16 bit precision sharded in [TGI](https://github.com/huggingface/text-generation-inference).
301
- In contrast the model could be evaluated without problem using [vLLM](https://github.com/vllm-project/vllm).
302
- The model also worked decently well when loaded with TGI on a single GPPU nf4 quantized via [TimDettmers/bitsandbytes](https://github.com/TimDettmers/bitsandbytes).
303
- Will will get it touch with the TGI authors to find out why sharded 16-bit inference doesn't work as expected.
 
 
304
 
305
  ## Configuration Details
306
 
 
2
  datasets:
3
  - rombodawg/LosslessMegaCodeTrainingV2_1m_Evol_Uncensored
4
  - OpenAssistant/oasst1
5
+ - shahules786/orca-best
6
  - argilla/databricks-dolly-15k-curated-multilingual
7
  inference: false
8
  language:
 
48
 
49
  The GGML format has now been superseded by GGUF. As of August 21st 2023, [llama.cpp](https://github.com/ggerganov/llama.cpp) no longer supports GGML models. Third party clients and libraries are expected to still support it for a time, but many may also drop support.
50
 
51
+ Please use the GGUF models instead.
52
+
53
  ### About GGML
54
 
55
  GPU acceleration is now available for Llama 2 70B GGML files, with both CUDA (NVidia) and Metal (macOS). The following clients/libraries are known to work with these files, including with GPU acceleration:
 
75
  <|im_start|>user
76
  {prompt}<|im_end|>
77
  <|im_start|>assistant
78
+
79
  ```
80
 
81
  <!-- compatibility_ggml start -->
 
124
  | [llama2-70b-oasst-sft-v10.ggmlv3.Q5_0.bin](https://huggingface.co/TheBloke/Llama2-70B-OASST-SFT-v10-GGML/blob/main/llama2-70b-oasst-sft-v10.ggmlv3.Q5_0.bin) | Q5_0 | 5 | 47.43 GB| 49.93 GB | Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
125
  | [llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_S.bin](https://huggingface.co/TheBloke/Llama2-70B-OASST-SFT-v10-GGML/blob/main/llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_S.bin) | Q5_K_S | 5 | 47.74 GB| 50.24 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
126
  | [llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_M.bin](https://huggingface.co/TheBloke/Llama2-70B-OASST-SFT-v10-GGML/blob/main/llama2-70b-oasst-sft-v10.ggmlv3.Q5_K_M.bin) | Q5_K_M | 5 | 49.03 GB| 51.53 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 |
 
 
 
 
 
127
 
128
+ **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.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
129
 
130
  ## How to run in `llama.cpp`
131
 
 
136
  I use the following command line; adjust for your tastes and needs:
137
 
138
  ```
139
+ ./main -t 10 -ngl 40 -gqa 8 -m llama2-70b-oasst-sft-v10.ggmlv3.q4_K_M.bin --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
140
  ```
141
  Change `-t 10` to the number of physical CPU cores you have. For example if your system has 8 cores/16 threads, use `-t 8`. If you are fully offloading the model to GPU, use `-t 1`
142
 
 
177
 
178
  **Special thanks to**: Aemon Algiz.
179
 
180
+ **Patreon special mentions**: Russ Johnson, J, alfie_i, Alex, NimbleBox.ai, Chadd, Mandus, Nikolai Manek, Ken Nordquist, ya boyyy, Illia Dulskyi, Viktor Bowallius, vamX, Iucharbius, zynix, Magnesian, Clay Pascal, Pierre Kircher, Enrico Ros, Tony Hughes, Elle, Andrey, knownsqashed, Deep Realms, Jerry Meng, Lone Striker, Derek Yates, Pyrater, Mesiah Bishop, James Bentley, Femi Adebogun, Brandon Frisco, SuperWojo, Alps Aficionado, Michael Dempsey, Vitor Caleffi, Will Dee, Edmond Seymore, usrbinkat, LangChain4j, Kacper Wikieł, Luke Pendergrass, John Detwiler, theTransient, Nathan LeClaire, Tiffany J. Kim, biorpg, Eugene Pentland, Stanislav Ovsiannikov, Fred von Graf, terasurfer, Kalila, Dan Guido, Nitin Borwankar, 阿明, Ai Maven, John Villwock, Gabriel Puliatti, Stephen Murray, Asp the Wyvern, danny, Chris Smitley, ReadyPlayerEmma, S_X, Daniel P. Andersen, Olakabola, Jeffrey Morgan, Imad Khwaja, Caitlyn Gatomon, webtim, Alicia Loh, Trenton Dambrowitz, Swaroop Kallakuri, Erik Bjäreholt, Leonard Tan, Spiking Neurons AB, Luke @flexchar, Ajan Kanaga, Thomas Belote, Deo Leter, RoA, Willem Michiel, transmissions 11, subjectnull, Matthew Berman, Joseph William Delisle, David Ziegler, Michael Davis, Johann-Peter Hartmann, Talal Aujan, senxiiz, Artur Olbinski, Rainer Wilmers, Spencer Kim, Fen Risland, Cap'n Zoog, Rishabh Srivastava, Michael Levine, Geoffrey Montalvo, Sean Connelly, Alexandros Triantafyllidis, Pieter, Gabriel Tamborski, Sam, Subspace Studios, Junyu Yang, Pedro Madruga, Vadim, Cory Kujawski, K, Raven Klaugh, Randy H, Mano Prime, Sebastain Graf, Space Cruiser
181
 
182
 
183
  Thank you to all my generous patrons and donaters!
 
268
 
269
  Please see Meta's [Responsible Use Guide](https://ai.meta.com/llama/responsible-use-guide/).
270
 
 
271
 
272
+ ## Inference via TGI
273
+
274
+ An early version of this model had an embedding count of 32,007 which was incompatible to sharding with [TGI](https://github.com/huggingface/text-generation-inference).
275
+ In the current version the embeddings and the lm_head weights have been padded to a multiple of 128 (by replicating the emembeddings of the unk-token (id: 0)).
276
+ Sharded inference with TGI should now work as expected.
277
+
278
 
279
  ## Configuration Details
280