TheBloke commited on
Commit
c99263d
1 Parent(s): ef08eda

Upload new k-quant GGML quantised models.

Browse files
Files changed (1) hide show
  1. README.md +73 -64
README.md CHANGED
@@ -1,17 +1,8 @@
1
  ---
2
- license: other
3
- library_name: transformers
4
- pipeline_tag: text-generation
5
- datasets:
6
- - RyokoAI/ShareGPT52K
7
- - Hello-SimpleAI/HC3
8
- tags:
9
- - koala
10
- - ShareGPT
11
- - llama
12
- - gptq
13
  inference: false
 
14
  ---
 
15
  <!-- header start -->
16
  <div style="width: 100%;">
17
  <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
@@ -25,78 +16,89 @@ inference: false
25
  </div>
26
  </div>
27
  <!-- header end -->
28
- # Koala: A Dialogue Model for Academic Research
29
- This repo contains the weights of the Koala 13B model produced at Berkeley. It is the result of combining the diffs from https://huggingface.co/young-geng/koala with the original Llama 13B model.
30
 
31
- This version has then been quantized to 4-bit and 5-bit GGML for use with [llama.cpp](https://github.com/ggerganov/llama.cpp).
32
 
33
- ## My Koala repos
34
- I have the following Koala model repositories available:
35
 
36
- **13B models:**
37
- * [Unquantized 13B model in HF format](https://huggingface.co/TheBloke/koala-13B-HF)
38
- * [GPTQ quantized 4bit 13B model in `pt` and `safetensors` formats](https://huggingface.co/TheBloke/koala-13B-GPTQ-4bit-128g)
39
- * [4bit and 5bit models in GGML format for `llama.cpp`](https://huggingface.co/TheBloke/koala-13B-GGML)
 
 
40
 
41
- **7B models:**
42
- * [Unquantized 7B model in HF format](https://huggingface.co/TheBloke/koala-7B-HF)
43
- * [Unquantized 7B model in GGML format for llama.cpp](https://huggingface.co/TheBloke/koala-7b-ggml-unquantized)
44
- * [GPTQ quantized 4bit 7B model in `pt` and `safetensors` formats](https://huggingface.co/TheBloke/koala-7B-GPTQ-4bit-128g)
45
- * [4bit and 5bit models in GGML format for `llama.cpp`](https://huggingface.co/TheBloke/koala-7B-GGML)
46
 
47
- ## THE FILES IN MAIN BRANCH REQUIRES LATEST LLAMA.CPP (May 19th 2023 - commit 2d5db48)!
 
 
48
 
49
- llama.cpp recently made another breaking change to its quantisation methods - https://github.com/ggerganov/llama.cpp/pull/1508
 
50
 
51
- I have quantised the GGML files in this repo with the latest version. Therefore you will require llama.cpp compiled on May 19th or later (commit `2d5db48` or later) to use them.
52
 
53
- For files compatible with the previous version of llama.cpp, please see branch `previous_llama_ggmlv2`.
54
 
55
- ## How to run in `llama.cpp`
56
 
57
- I use the following command line; adjust for your tastes and needs:
58
 
59
- ```
60
- ./main -t 18 -m koala-13B-4bit-128g.GGML.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "BEGINNING OF CONVERSATION:
61
- USER: <PROMPT GOES HERE>
62
- GPT:"
63
- ```
64
 
65
- Change `-t 18` to the number of physical CPU cores you have. For example if your system has 8 cores, 16 threads, use `-t 8`.
66
 
67
- You will require 16GB or more RAM to run this model without swapping.
68
 
69
- ## How the Koala delta weights were merged
 
 
 
 
 
 
70
 
71
- The Koala delta weights were originally merged using the following commands, producing [koala-13B-HF](https://huggingface.co/TheBloke/koala-13B-HF):
72
- ```
73
- git clone https://github.com/young-geng/EasyLM
74
 
75
- git clone https://huggingface.co/TheBloke/llama-13b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
76
 
77
- mkdir koala_diffs && cd koala_diffs && wget https://huggingface.co/young-geng/koala/resolve/main/koala_13b_diff_v2
78
 
79
- cd EasyLM
80
 
81
- PYTHON_PATH="${PWD}:$PYTHONPATH" python \
82
- -m EasyLM.models.llama.convert_torch_to_easylm \
83
- --checkpoint_dir=/content/llama-13b \
84
- --output_file=/content/llama-13b-LM \
85
- --streaming=True
86
 
87
- PYTHON_PATH="${PWD}:$PYTHONPATH" python \
88
- -m EasyLM.scripts.diff_checkpoint --recover_diff=True \
89
- --load_base_checkpoint='params::/content/llama-13b-LM' \
90
- --load_target_checkpoint='params::/content/koala_diffs/koala_13b_diff_v2' \
91
- --output_file=/content/koala_13b.diff.weights \
92
- --streaming=True
93
 
94
- PYTHON_PATH="${PWD}:$PYTHONPATH" python \
95
- -m EasyLM.models.llama.convert_easylm_to_hf --model_size=13b \
96
- --output_dir=/content/koala-13B-HF \
97
- --load_checkpoint='params::/content/koala_13b.diff.weights' \
98
- --tokenizer_path=/content/llama-13b/tokenizer.model
99
  ```
 
 
 
 
 
 
 
 
 
 
 
100
 
101
  <!-- footer start -->
102
  ## Discord
@@ -118,13 +120,19 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
118
  * Patreon: https://patreon.com/TheBlokeAI
119
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
120
 
121
- **Patreon special mentions**: Aemon Algiz, Dmitriy Samsonov, Nathan LeClaire, Trenton Dambrowitz, Mano Prime, David Flickinger, vamX, Nikolai Manek, senxiiz, Khalefa Al-Ahmad, Illia Dulskyi, Jonathan Leane, Talal Aujan, V. Lukas, Joseph William Delisle, Pyrater, Oscar Rangel, Lone Striker, Luke Pendergrass, Eugene Pentland, Sebastain Graf, Johann-Peter Hartman.
 
 
122
 
123
  Thank you to all my generous patrons and donaters!
 
124
  <!-- footer end -->
125
- ## Further info
126
 
127
- Check out the following links to learn more about the Berkeley Koala model.
 
 
 
 
128
  * [Blog post](https://bair.berkeley.edu/blog/2023/04/03/koala/)
129
  * [Online demo](https://koala.lmsys.org/)
130
  * [EasyLM: training and serving framework on GitHub](https://github.com/young-geng/EasyLM)
@@ -136,3 +144,4 @@ The model weights are intended for academic research only, subject to the
136
  [Terms of Use of the data generated by OpenAI](https://openai.com/policies/terms-of-use),
137
  and [Privacy Practices of ShareGPT](https://chrome.google.com/webstore/detail/sharegpt-share-your-chatg/daiacboceoaocpibfodeljbdfacokfjb).
138
  Any other usage of the model weights, including but not limited to commercial usage, is strictly prohibited.
 
 
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;">
 
16
  </div>
17
  </div>
18
  <!-- header end -->
 
 
19
 
20
+ # Koala 13B GGML
21
 
22
+ These files are GGML format model files for [Koala 13B](https://huggingface.co/young-geng/koala).
 
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
+ * [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/koala-13B-GPTQ-4bit-128g)
34
+ * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/koala-13B-GGML)
35
+ * [Unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/TheBloke/koala-13B-HF)
36
 
37
+ <!-- compatibility_ggml start -->
38
+ ## Compatibility
39
 
40
+ ### Original llama.cpp quant methods: `q4_0, q4_1, q5_0, q5_1, q8_0`
41
 
42
+ 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`.
43
 
44
+ They should be compatible with all current UIs and libraries that use llama.cpp, such as those listed at the top of this README.
45
 
46
+ ### 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`
47
 
48
+ These new quantisation methods are only compatible with llama.cpp as of June 6th, commit `2d43387`.
 
 
 
 
49
 
50
+ 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.
51
 
52
+ ## Explanation of the new k-quant methods
53
 
54
+ The new methods available are:
55
+ * 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)
56
+ * 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.
57
+ * 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.
58
+ * GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
59
+ * 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
60
+ * 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.
61
 
62
+ Refer to the Provided Files table below to see what files use which methods, and how.
63
+ <!-- compatibility_ggml end -->
 
64
 
65
+ ## Provided files
66
+ | Name | Quant method | Bits | Size | Max RAM required | Use case |
67
+ | ---- | ---- | ---- | ---- | ---- | ----- |
68
+ | koala-13B.ggmlv3.q2_K.bin | q2_K | 2 | 5.43 GB | 7.93 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. |
69
+ | koala-13B.ggmlv3.q3_K_L.bin | q3_K_L | 3 | 6.87 GB | 9.37 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 |
70
+ | koala-13B.ggmlv3.q3_K_M.bin | q3_K_M | 3 | 6.25 GB | 8.75 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 |
71
+ | koala-13B.ggmlv3.q3_K_S.bin | q3_K_S | 3 | 5.59 GB | 8.09 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors |
72
+ | koala-13B.ggmlv3.q4_0.bin | q4_0 | 4 | 7.32 GB | 9.82 GB | Original llama.cpp quant method, 4-bit. |
73
+ | koala-13B.ggmlv3.q4_1.bin | q4_1 | 4 | 8.14 GB | 10.64 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. |
74
+ | koala-13B.ggmlv3.q4_K_M.bin | q4_K_M | 4 | 7.82 GB | 10.32 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 |
75
+ | koala-13B.ggmlv3.q4_K_S.bin | q4_K_S | 4 | 7.32 GB | 9.82 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors |
76
+ | koala-13B.ggmlv3.q5_0.bin | q5_0 | 5 | 8.95 GB | 11.45 GB | Original llama.cpp quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
77
+ | koala-13B.ggmlv3.q5_1.bin | q5_1 | 5 | 9.76 GB | 12.26 GB | Original llama.cpp quant method, 5-bit. Even higher accuracy, resource usage and slower inference. |
78
+ | koala-13B.ggmlv3.q5_K_M.bin | q5_K_M | 5 | 9.21 GB | 11.71 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 |
79
+ | koala-13B.ggmlv3.q5_K_S.bin | q5_K_S | 5 | 8.95 GB | 11.45 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
80
+ | koala-13B.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 |
81
+ | koala-13B.ggmlv3.q8_0.bin | q8_0 | 8 | 13.83 GB | 16.33 GB | Original llama.cpp quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users. |
82
 
 
83
 
84
+ **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.
85
 
86
+ ## How to run in `llama.cpp`
 
 
 
 
87
 
88
+ I use the following command line; adjust for your tastes and needs:
 
 
 
 
 
89
 
 
 
 
 
 
90
  ```
91
+ ./main -t 10 -ngl 32 -m koala-13B.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:"
92
+ ```
93
+ 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`.
94
+
95
+ Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
96
+
97
+ If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
98
+
99
+ ## How to run in `text-generation-webui`
100
+
101
+ Further instructions here: [text-generation-webui/docs/llama.cpp-models.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp-models.md).
102
 
103
  <!-- footer start -->
104
  ## Discord
 
120
  * Patreon: https://patreon.com/TheBlokeAI
121
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
122
 
123
+ **Special thanks to**: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.
124
+
125
+ **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.
126
 
127
  Thank you to all my generous patrons and donaters!
128
+
129
  <!-- footer end -->
 
130
 
131
+ # Original model card: Koala 13B
132
+
133
+
134
+ # Koala: A Dialogue Model for Academic Research
135
+ This repo contains the weights diff against the base LLaMA for the Koala model. Check out the following links to get started:
136
  * [Blog post](https://bair.berkeley.edu/blog/2023/04/03/koala/)
137
  * [Online demo](https://koala.lmsys.org/)
138
  * [EasyLM: training and serving framework on GitHub](https://github.com/young-geng/EasyLM)
 
144
  [Terms of Use of the data generated by OpenAI](https://openai.com/policies/terms-of-use),
145
  and [Privacy Practices of ShareGPT](https://chrome.google.com/webstore/detail/sharegpt-share-your-chatg/daiacboceoaocpibfodeljbdfacokfjb).
146
  Any other usage of the model weights, including but not limited to commercial usage, is strictly prohibited.
147
+ Please contact us If you find any potential violations. Our training and inference code is released under the Apache License 2.0.