Upload new k-quant GGML quantised models.

README.md

@@ -1,17 +1,8 @@
 ---
-license: other
-library_name: transformers
-pipeline_tag: text-generation
-datasets:
-- RyokoAI/ShareGPT52K
-- Hello-SimpleAI/HC3
-tags:
-- koala
-- ShareGPT
-- llama
-- gptq
 inference: false
+license: other
 ---
+
 <!-- header start -->
 <div style="width: 100%;">
     <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
     </div>
 </div>
 <!-- header end -->
-# Koala: A Dialogue Model for Academic Research
-This repo contains the weights of the Koala 7B model produced at Berkeley. It is the result of combining the diffs from https://huggingface.co/young-geng/koala with the original Llama 7B model.
 
-This version has then been quantized to 4-bit and 5-bit GGML for use with [llama.cpp](https://github.com/ggerganov/llama.cpp).
+# Koala 7B GGML
 
-## My Koala repos
-I have the following Koala model repositories available:
+These files are GGML format model files for [Koala 7B](https://huggingface.co/young-geng/koala).
 
-**13B models:**
-* [Unquantized 13B model in HF format](https://huggingface.co/TheBloke/koala-13B-HF)
-* [GPTQ quantized 4bit 13B model in `pt` and `safetensors` formats](https://huggingface.co/TheBloke/koala-13B-GPTQ-4bit-128g)
-* [4-bit, 5-bit and 8-bit GGML models for `llama.cpp`](https://huggingface.co/TheBloke/koala-13B-GGML)
+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:
+* [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
+* [KoboldCpp](https://github.com/LostRuins/koboldcpp)
+* [ParisNeo/GPT4All-UI](https://github.com/ParisNeo/gpt4all-ui)
+* [llama-cpp-python](https://github.com/abetlen/llama-cpp-python)
+* [ctransformers](https://github.com/marella/ctransformers)
 
-**7B models:**
-* [Unquantized 7B model in HF format](https://huggingface.co/TheBloke/koala-7B-HF)
-* [Unquantized 7B model in GGML format for llama.cpp](https://huggingface.co/TheBloke/koala-7b-ggml-unquantized)
-* [GPTQ quantized 4bit 7B model in `pt` and `safetensors` formats](https://huggingface.co/TheBloke/koala-7B-GPTQ-4bit-128g)
-* [4-bit, 5-bit and 8-bit GGML models for `llama.cpp`](https://huggingface.co/TheBloke/koala-7B-GGML)
+## Repositories available
 
-## THE FILES IN MAIN BRANCH REQUIRES LATEST LLAMA.CPP (May 19th 2023 - commit 2d5db48)!
+* [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/koala-7B-GPTQ-4bit-128g)
+* [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/koala-7B-GGML)
+* [Unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/TheBloke/koala-7B-HF)
 
-llama.cpp recently made another breaking change to its quantisation methods - https://github.com/ggerganov/llama.cpp/pull/1508
+<!-- compatibility_ggml start -->
+## Compatibility
 
-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.
+### Original llama.cpp quant methods: `q4_0, q4_1, q5_0, q5_1, q8_0`
 
-For files compatible with the previous version of llama.cpp, please see branch `previous_llama_ggmlv2`.
+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`.
 
-## How to run in `llama.cpp`
+They should be compatible with all current UIs and libraries that use llama.cpp, such as those listed at the top of this README.
 
-I use the following command line; adjust for your tastes and needs:
+### 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`
 
-```
-./main -t 18 -m koala-7B-4bit-128g.ggmlv3.q5_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "BEGINNING OF CONVERSATION:
-USER: <PROMPT GOES HERE>
-GPT:"
-```
+These new quantisation methods are only compatible with llama.cpp as of June 6th, commit `2d43387`.
 
-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`.
+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.
 
-This model should be able to run in 8GB RAM without swapping.
+## Explanation of the new k-quant methods
 
-## How the Koala delta weights were merged
+The new methods available are:
+* 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)
+* 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.
+* 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.
+* GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
+* 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
+* 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.
 
-The Koala delta weights were originally merged using the following commands, producing [koala-7B-HF](https://huggingface.co/TheBloke/koala-7B-HF):
-```
-git clone https://github.com/young-geng/EasyLM
+Refer to the Provided Files table below to see what files use which methods, and how.
+<!-- compatibility_ggml end -->
 
-git clone https://huggingface.co/nyanko7/LLaMA-7B
+## Provided files
+| Name | Quant method | Bits | Size | Max RAM required | Use case |
+| ---- | ---- | ---- | ---- | ---- | ----- |
+| koala-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. |
+| koala-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 |
+| koala-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 |
+| koala-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 |
+| koala-7B.ggmlv3.q4_0.bin | q4_0 | 4 | 3.79 GB | 6.29 GB | Original llama.cpp quant method, 4-bit. |
+| koala-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. |
+| koala-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 |
+| koala-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 |
+| koala-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. |
+| koala-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. |
+| koala-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 |
+| koala-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 |
+| koala-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 |
+| koala-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. |
 
-mkdir koala_diffs && cd koala_diffs && wget https://huggingface.co/young-geng/koala/resolve/main/koala_7b_diff_v2
 
-cd EasyLM
+**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.
 
-PYTHON_PATH="${PWD}:$PYTHONPATH" python \
--m EasyLM.models.llama.convert_torch_to_easylm \
---checkpoint_dir=/content/LLaMA-7B \
---output_file=/content/llama-7B-LM \
---streaming=True
+## How to run in `llama.cpp`
 
-PYTHON_PATH="${PWD}:$PYTHONPATH" python \
--m EasyLM.scripts.diff_checkpoint --recover_diff=True \
---load_base_checkpoint='params::/content/llama-7B-LM' \
---load_target_checkpoint='params::/content/koala_diffs/koala_7b_diff_v2' \
---output_file=/content/koala_7b.diff.weights \
---streaming=True
+I use the following command line; adjust for your tastes and needs:
 
-PYTHON_PATH="${PWD}:$PYTHONPATH" python \
--m EasyLM.models.llama.convert_easylm_to_hf --model_size=7b \
---output_dir=/content/koala-7B-HF \
---load_checkpoint='params::/content/koala_7b.diff.weights' \
---tokenizer_path=/content/LLaMA-7B/tokenizer.model
 ```
+./main -t 10 -ngl 32 -m koala-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:"
+```
+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`.
+
+Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
+
+If you want to have a chat-style conversation, replace the `-p <PROMPT>` argument with `-i -ins`
+
+## How to run in `text-generation-webui`
+
+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).
 
 <!-- footer start -->
 ## Discord
@@ -118,13 +120,19 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
 * Patreon: https://patreon.com/TheBlokeAI
 * Ko-Fi: https://ko-fi.com/TheBlokeAI
 
-**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.
+**Special thanks to**: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.
+
+**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.
 
 Thank you to all my generous patrons and donaters!
+
 <!-- footer end -->
-## Further info
 
-Check out the following links to learn more about the Berkeley Koala model.
+# Original model card: Koala 7B
+
+
+# Koala: A Dialogue Model for Academic Research
+This repo contains the weights diff against the base LLaMA for the Koala model. Check out the following links to get started:
 * [Blog post](https://bair.berkeley.edu/blog/2023/04/03/koala/)
 * [Online demo](https://koala.lmsys.org/)
 * [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
 [Terms of Use of the data generated by OpenAI](https://openai.com/policies/terms-of-use),
 and [Privacy Practices of ShareGPT](https://chrome.google.com/webstore/detail/sharegpt-share-your-chatg/daiacboceoaocpibfodeljbdfacokfjb).
 Any other usage of the model weights, including but not limited to commercial usage, is strictly prohibited.
+Please contact us If you find any potential violations. Our training and inference code is released under the Apache License 2.0.