TheBloke commited on
Commit
f116503
1 Parent(s): e9f12bc

Initial GGML model commit

Browse files
Files changed (1) hide show
  1. README.md +51 -34
README.md CHANGED
@@ -1,6 +1,7 @@
1
  ---
2
  inference: false
3
  license: other
 
4
  ---
5
 
6
  <!-- header start -->
@@ -9,7 +10,7 @@ license: other
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/Jq4vkcDakD">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>
@@ -17,38 +18,48 @@ license: other
17
  </div>
18
  <!-- header end -->
19
 
20
- # Meta's LLaMA 7B GGML
21
 
22
- These files are GGML format model files for [Meta's LLaMA 7B](https://ai.facebook.com/blog/large-language-model-llama-meta-ai/).
23
 
24
  GGML files are for CPU + GPU inference using [llama.cpp](https://github.com/ggerganov/llama.cpp) and libraries and UIs which support this format, such as:
25
- * [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
26
- * [KoboldCpp](https://github.com/LostRuins/koboldcpp)
27
- * [ParisNeo/GPT4All-UI](https://github.com/ParisNeo/gpt4all-ui)
28
- * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python)
29
- * [ctransformers](https://github.com/marella/ctransformers)
 
 
 
30
 
31
  ## Repositories available
32
 
33
- * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/llama-7B-GGML)
34
- * [huggyllama's unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/huggyllama/llama-7b)
 
 
 
 
 
 
 
35
 
36
  <!-- compatibility_ggml start -->
37
  ## Compatibility
38
 
39
  ### Original llama.cpp quant methods: `q4_0, q4_1, q5_0, q5_1, q8_0`
40
 
41
- I have quantized these 'original' quantisation methods using an older version of llama.cpp so that they remain compatible with llama.cpp as of May 19th, commit `2d5db48`.
42
-
43
- They should be compatible with all current UIs and libraries that use llama.cpp, such as those listed at the top of this README.
44
 
45
  ### New k-quant methods: `q2_K, q3_K_S, q3_K_M, q3_K_L, q4_K_S, q4_K_M, q5_K_S, q6_K`
46
 
47
- These new quantisation methods are only compatible with llama.cpp as of June 6th, commit `2d43387`.
48
 
49
- They will NOT be compatible with koboldcpp, text-generation-ui, and other UIs and libraries yet. Support is expected to come over the next few days.
50
 
51
  ## Explanation of the new k-quant methods
 
 
52
 
53
  The new methods available are:
54
  * 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)
@@ -59,26 +70,26 @@ The new methods available are:
59
  * 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.
60
 
61
  Refer to the Provided Files table below to see what files use which methods, and how.
 
62
  <!-- compatibility_ggml end -->
63
 
64
  ## Provided files
65
  | Name | Quant method | Bits | Size | Max RAM required | Use case |
66
  | ---- | ---- | ---- | ---- | ---- | ----- |
67
- | llama-7b.ggmlv3.q2_K.bin | q2_K | 2 | 2.80 GB | 5.30 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. |
68
- | llama-7b.ggmlv3.q3_K_L.bin | q3_K_L | 3 | 3.55 GB | 6.05 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 |
69
- | llama-7b.ggmlv3.q3_K_M.bin | q3_K_M | 3 | 3.23 GB | 5.73 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 |
70
- | llama-7b.ggmlv3.q3_K_S.bin | q3_K_S | 3 | 2.90 GB | 5.40 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors |
71
- | llama-7b.ggmlv3.q4_0.bin | q4_0 | 4 | 3.79 GB | 6.29 GB | Original llama.cpp quant method, 4-bit. |
72
- | llama-7b.ggmlv3.q4_1.bin | q4_1 | 4 | 4.21 GB | 6.71 GB | Original llama.cpp quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models. |
73
- | llama-7b.ggmlv3.q4_K_M.bin | q4_K_M | 4 | 4.05 GB | 6.55 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 |
74
- | llama-7b.ggmlv3.q4_K_S.bin | q4_K_S | 4 | 3.79 GB | 6.29 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors |
75
- | llama-7b.ggmlv3.q5_0.bin | q5_0 | 5 | 4.63 GB | 7.13 GB | Original llama.cpp quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
76
- | llama-7b.ggmlv3.q5_1.bin | q5_1 | 5 | 5.06 GB | 7.56 GB | Original llama.cpp quant method, 5-bit. Even higher accuracy, resource usage and slower inference. |
77
- | llama-7b.ggmlv3.q5_K_M.bin | q5_K_M | 5 | 4.77 GB | 7.27 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 |
78
- | llama-7b.ggmlv3.q5_K_S.bin | q5_K_S | 5 | 4.63 GB | 7.13 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
79
- | llama-7b.ggmlv3.q6_K.bin | q6_K | 6 | 5.53 GB | 8.03 GB | New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors |
80
- | llama-7b.ggmlv3.q8_0.bin | q8_0 | 8 | 7.16 GB | 9.66 GB | Original llama.cpp quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users. |
81
-
82
 
83
  **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.
84
 
@@ -87,7 +98,7 @@ Refer to the Provided Files table below to see what files use which methods, and
87
  I use the following command line; adjust for your tastes and needs:
88
 
89
  ```
90
- ./main -t 10 -ngl 32 -m llama-7b.ggmlv3.q5_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction: Write a story about llamas\n### Response:"
91
  ```
92
  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`.
93
 
@@ -104,7 +115,7 @@ Further instructions here: [text-generation-webui/docs/llama.cpp-models.md](http
104
 
105
  For further support, and discussions on these models and AI in general, join us at:
106
 
107
- [TheBloke AI's Discord server](https://discord.gg/Jq4vkcDakD)
108
 
109
  ## Thanks, and how to contribute.
110
 
@@ -119,11 +130,17 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
119
  * Patreon: https://patreon.com/TheBlokeAI
120
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
121
 
122
- **Special thanks to**: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.
123
 
124
- **Patreon special mentions**: Oscar Rangel, Eugene Pentland, Talal Aujan, Cory Kujawski, Luke, Asp the Wyvern, Ai Maven, Pyrater, Alps Aficionado, senxiiz, Willem Michiel, Junyu Yang, trip7s trip, Sebastain Graf, Joseph William Delisle, Lone Striker, Jonathan Leane, Johann-Peter Hartmann, David Flickinger, Spiking Neurons AB, Kevin Schuppel, Mano Prime, Dmitriy Samsonov, Sean Connelly, Nathan LeClaire, Alain Rossmann, Fen Risland, Derek Yates, Luke Pendergrass, Nikolai Manek, Khalefa Al-Ahmad, Artur Olbinski, John Detwiler, Ajan Kanaga, Imad Khwaja, Trenton Dambrowitz, Kalila, vamX, webtim, Illia Dulskyi.
125
 
126
  Thank you to all my generous patrons and donaters!
127
 
128
  <!-- footer end -->
129
 
 
 
 
 
 
 
 
1
  ---
2
  inference: false
3
  license: other
4
+ model_type: llama
5
  ---
6
 
7
  <!-- header start -->
 
10
  </div>
11
  <div style="display: flex; justify-content: space-between; width: 100%;">
12
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
13
+ <p><a href="https://discord.gg/theblokeai">Chat & support: my new Discord server</a></p>
14
  </div>
15
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
16
  <p><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
 
18
  </div>
19
  <!-- header end -->
20
 
21
+ # Meta's LLaMA 7b GGML
22
 
23
+ These files are GGML format model files for [Meta's LLaMA 7b](https://ai.meta.com/blog/large-language-model-llama-meta-ai).
24
 
25
  GGML files are for CPU + GPU inference using [llama.cpp](https://github.com/ggerganov/llama.cpp) and libraries and UIs which support this format, such as:
26
+ * [KoboldCpp](https://github.com/LostRuins/koboldcpp), a powerful GGML web UI with full GPU acceleration out of the box. Especially good for story telling.
27
+ * [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui), a great web UI with GPU acceleration via the c_transformers backend.
28
+ * [LM Studio](https://lmstudio.ai/), a fully featured local GUI. Supports full GPU accel on macOS. Also supports Windows, without GPU accel.
29
+ * [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most popular web UI. Requires extra steps to enable GPU accel via llama.cpp backend.
30
+ * [ctransformers](https://github.com/marella/ctransformers), a Python library with LangChain support and OpenAI-compatible AI server.
31
+ * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), a Python library with OpenAI-compatible API server.
32
+
33
+ These files were quantised using hardware kindly provided by [Latitude.sh](https://www.latitude.sh/accelerate).
34
 
35
  ## Repositories available
36
 
37
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/LLaMA-7b-GPTQ)
38
+ * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/LLaMA-7b-GGML)
39
+ * [Unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/huggyllama/llama-7b)
40
+
41
+ ## Prompt template: None
42
+
43
+ ```
44
+ {prompt}
45
+ ```
46
 
47
  <!-- compatibility_ggml start -->
48
  ## Compatibility
49
 
50
  ### Original llama.cpp quant methods: `q4_0, q4_1, q5_0, q5_1, q8_0`
51
 
52
+ These are guaranteed to be compatible with any UIs, tools and libraries released since late May. They may be phased out soon, as they are largely superseded by the new k-quant methods.
 
 
53
 
54
  ### New k-quant methods: `q2_K, q3_K_S, q3_K_M, q3_K_L, q4_K_S, q4_K_M, q5_K_S, q6_K`
55
 
56
+ These new quantisation methods are compatible with llama.cpp as of June 6th, commit `2d43387`.
57
 
58
+ They are now also compatible with recent releases of text-generation-webui, KoboldCpp, llama-cpp-python, ctransformers, rustformers and most others. For compatibility with other tools and libraries, please check their documentation.
59
 
60
  ## Explanation of the new k-quant methods
61
+ <details>
62
+ <summary>Click to see details</summary>
63
 
64
  The new methods available are:
65
  * 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)
 
70
  * 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.
71
 
72
  Refer to the Provided Files table below to see what files use which methods, and how.
73
+ </details>
74
  <!-- compatibility_ggml end -->
75
 
76
  ## Provided files
77
  | Name | Quant method | Bits | Size | Max RAM required | Use case |
78
  | ---- | ---- | ---- | ---- | ---- | ----- |
79
+ | llama-7b.ggmlv3.q2_K.bin | q2_K | 2 | 2.80 GB| 5.30 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. |
80
+ | llama-7b.ggmlv3.q3_K_L.bin | q3_K_L | 3 | 3.55 GB| 6.05 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 |
81
+ | llama-7b.ggmlv3.q3_K_M.bin | q3_K_M | 3 | 3.23 GB| 5.73 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 |
82
+ | llama-7b.ggmlv3.q3_K_S.bin | q3_K_S | 3 | 2.90 GB| 5.40 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors |
83
+ | llama-7b.ggmlv3.q4_0.bin | q4_0 | 4 | 3.79 GB| 6.29 GB | Original quant method, 4-bit. |
84
+ | llama-7b.ggmlv3.q4_1.bin | q4_1 | 4 | 4.21 GB| 6.71 GB | Original quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models. |
85
+ | llama-7b.ggmlv3.q4_K_M.bin | q4_K_M | 4 | 4.05 GB| 6.55 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 |
86
+ | llama-7b.ggmlv3.q4_K_S.bin | q4_K_S | 4 | 3.79 GB| 6.29 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors |
87
+ | llama-7b.ggmlv3.q5_0.bin | q5_0 | 5 | 4.63 GB| 7.13 GB | Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
88
+ | llama-7b.ggmlv3.q5_1.bin | q5_1 | 5 | 5.06 GB| 7.56 GB | Original quant method, 5-bit. Even higher accuracy, resource usage and slower inference. |
89
+ | llama-7b.ggmlv3.q5_K_M.bin | q5_K_M | 5 | 4.77 GB| 7.27 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 |
90
+ | llama-7b.ggmlv3.q5_K_S.bin | q5_K_S | 5 | 4.63 GB| 7.13 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
91
+ | llama-7b.ggmlv3.q6_K.bin | q6_K | 6 | 5.53 GB| 8.03 GB | New k-quant method. Uses GGML_TYPE_Q8_K for all tensors - 6-bit quantization |
92
+ | llama-7b.ggmlv3.q8_0.bin | q8_0 | 8 | 7.16 GB| 9.66 GB | Original quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users. |
 
93
 
94
  **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.
95
 
 
98
  I use the following command line; adjust for your tastes and needs:
99
 
100
  ```
101
+ ./main -t 10 -ngl 32 -m llama-7b.ggmlv3.q4_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction: Write a story about llamas\n### Response:"
102
  ```
103
  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`.
104
 
 
115
 
116
  For further support, and discussions on these models and AI in general, join us at:
117
 
118
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
119
 
120
  ## Thanks, and how to contribute.
121
 
 
130
  * Patreon: https://patreon.com/TheBlokeAI
131
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
132
 
133
+ **Special thanks to**: Luke from CarbonQuill, Aemon Algiz.
134
 
135
+ **Patreon special mentions**: Space Cruiser, Nikolai Manek, Sam, Chris McCloskey, Rishabh Srivastava, Kalila, Spiking Neurons AB, Khalefa Al-Ahmad, WelcomeToTheClub, Chadd, Lone Striker, Viktor Bowallius, Edmond Seymore, Ai Maven, Chris Smitley, Dave, Alexandros Triantafyllidis, Luke @flexchar, Elle, ya boyyy, Talal Aujan, Alex , Jonathan Leane, Deep Realms, Randy H, subjectnull, Preetika Verma, Joseph William Delisle, Michael Levine, chris gileta, K, Oscar Rangel, LangChain4j, Trenton Dambrowitz, Eugene Pentland, Johann-Peter Hartmann, Femi Adebogun, Illia Dulskyi, senxiiz, Daniel P. Andersen, Sean Connelly, Artur Olbinski, RoA, Mano Prime, Derek Yates, Raven Klaugh, David Flickinger, Willem Michiel, Pieter, Willian Hasse, vamX, Luke Pendergrass, webtim, Ghost , Rainer Wilmers, Nathan LeClaire, Will Dee, Cory Kujawski, John Detwiler, Fred von Graf, biorpg, Iucharbius , Imad Khwaja, Pierre Kircher, terasurfer , Asp the Wyvern, John Villwock, theTransient, zynix , Gabriel Tamborski, Fen Risland, Gabriel Puliatti, Matthew Berman, Pyrater, SuperWojo, Stephen Murray, Karl Bernard, Ajan Kanaga, Greatston Gnanesh, Junyu Yang.
136
 
137
  Thank you to all my generous patrons and donaters!
138
 
139
  <!-- footer end -->
140
 
141
+ # Original model card: Meta's LLaMA 7b
142
+
143
+
144
+ This contains the weights for the LLaMA-7b model. This model is under a non-commercial license (see the LICENSE file).
145
+ You should only use this repository if you have been granted access to the model by filling out [this form](https://docs.google.com/forms/d/e/1FAIpQLSfqNECQnMkycAp2jP4Z9TFX0cGR4uf7b_fBxjY_OjhJILlKGA/viewform?usp=send_form) but either lost your copy of the weights or got some trouble converting them to the Transformers format.
146
+