bartowski commited on
Commit
f2f95f9
โ€ข
1 Parent(s): f298d51

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +37 -69
README.md CHANGED
@@ -27,93 +27,61 @@ language:
27
  license: cc-by-nc-4.0
28
  quantized_by: bartowski
29
  pipeline_tag: text-generation
 
 
 
 
 
 
 
 
 
 
30
  ---
 
31
 
32
- ## Llamacpp imatrix Quantizations of aya-23-35B
33
 
34
- Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b2965">b2965</a> for quantization.
 
 
35
 
36
- Original model: https://huggingface.co/CohereForAI/aya-23-35B
37
 
38
- All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/b6ac44691e994344625687afe3263b3a)
 
39
 
40
- ## Prompt format
41
 
42
- ```
43
- <BOS_TOKEN><|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{system_prompt}<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|USER_TOKEN|>{prompt}<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|><|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
44
- ```
45
 
46
- ## Download a file (not the whole branch) from below:
47
-
48
- | Filename | Quant type | File Size | Description |
49
- | -------- | ---------- | --------- | ----------- |
50
- | [aya-23-35B-Q8_0.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q8_0.gguf) | Q8_0 | 37.17GB | Extremely high quality, generally unneeded but max available quant. |
51
- | [aya-23-35B-Q6_K.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q6_K.gguf) | Q6_K | 28.70GB | Very high quality, near perfect, *recommended*. |
52
- | [aya-23-35B-Q5_K_M.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q5_K_M.gguf) | Q5_K_M | 25.00GB | High quality, *recommended*. |
53
- | [aya-23-35B-Q5_K_S.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q5_K_S.gguf) | Q5_K_S | 24.33GB | High quality, *recommended*. |
54
- | [aya-23-35B-Q4_K_M.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q4_K_M.gguf) | Q4_K_M | 21.52GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
55
- | [aya-23-35B-Q4_K_S.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q4_K_S.gguf) | Q4_K_S | 20.37GB | Slightly lower quality with more space savings, *recommended*. |
56
- | [aya-23-35B-IQ4_NL.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ4_NL.gguf) | IQ4_NL | 20.22GB | Decent quality, slightly smaller than Q4_K_S with similar performance *recommended*. |
57
- | [aya-23-35B-IQ4_XS.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ4_XS.gguf) | IQ4_XS | 19.20GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
58
- | [aya-23-35B-Q3_K_L.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q3_K_L.gguf) | Q3_K_L | 19.14GB | Lower quality but usable, good for low RAM availability. |
59
- | [aya-23-35B-Q3_K_M.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q3_K_M.gguf) | Q3_K_M | 17.61GB | Even lower quality. |
60
- | [aya-23-35B-IQ3_M.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ3_M.gguf) | IQ3_M | 16.69GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
61
- | [aya-23-35B-IQ3_S.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ3_S.gguf) | IQ3_S | 15.86GB | Lower quality, new method with decent performance, recommended over Q3_K_S quant, same size with better performance. |
62
- | [aya-23-35B-Q3_K_S.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q3_K_S.gguf) | Q3_K_S | 15.86GB | Low quality, not recommended. |
63
- | [aya-23-35B-IQ3_XS.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ3_XS.gguf) | IQ3_XS | 15.09GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
64
- | [aya-23-35B-IQ3_XXS.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ3_XXS.gguf) | IQ3_XXS | 13.83GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
65
- | [aya-23-35B-Q2_K.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-Q2_K.gguf) | Q2_K | 13.81GB | Very low quality but surprisingly usable. |
66
- | [aya-23-35B-IQ2_M.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ2_M.gguf) | IQ2_M | 12.67GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
67
- | [aya-23-35B-IQ2_S.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ2_S.gguf) | IQ2_S | 11.84GB | Very low quality, uses SOTA techniques to be usable. |
68
- | [aya-23-35B-IQ2_XS.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ2_XS.gguf) | IQ2_XS | 11.10GB | Very low quality, uses SOTA techniques to be usable. |
69
- | [aya-23-35B-IQ2_XXS.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ2_XXS.gguf) | IQ2_XXS | 10.18GB | Lower quality, uses SOTA techniques to be usable. |
70
- | [aya-23-35B-IQ1_M.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ1_M.gguf) | IQ1_M | 9.14GB | Extremely low quality, *not* recommended. |
71
- | [aya-23-35B-IQ1_S.gguf](https://huggingface.co/bartowski/aya-23-35B-GGUF/blob/main/aya-23-35B-IQ1_S.gguf) | IQ1_S | 8.52GB | Extremely low quality, *not* recommended. |
72
-
73
- ## Downloading using huggingface-cli
74
-
75
- First, make sure you have hugginface-cli installed:
76
 
77
  ```
78
- pip install -U "huggingface_hub[cli]"
79
- ```
80
-
81
- Then, you can target the specific file you want:
82
-
83
- ```
84
- huggingface-cli download bartowski/aya-23-35B-GGUF --include "aya-23-35B-Q4_K_M.gguf" --local-dir ./
85
- ```
86
 
87
- If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
88
-
89
- ```
90
- huggingface-cli download bartowski/aya-23-35B-GGUF --include "aya-23-35B-Q8_0.gguf/*" --local-dir aya-23-35B-Q8_0
91
  ```
92
 
93
- You can either specify a new local-dir (aya-23-35B-Q8_0) or download them all in place (./)
94
-
95
- ## Which file should I choose?
96
-
97
- A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
98
-
99
- The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
100
-
101
- If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
102
-
103
- If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
104
 
105
- Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
 
106
 
107
- If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
 
108
 
109
- If you want to get more into the weeds, you can check out this extremely useful feature chart:
110
 
111
- [llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
112
 
113
- But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
114
 
115
- These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
116
 
117
- The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
118
 
119
- Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
 
27
  license: cc-by-nc-4.0
28
  quantized_by: bartowski
29
  pipeline_tag: text-generation
30
+ lm_studio:
31
+ param_count: 35b
32
+ use_case: general
33
+ release_date: 23-05-2024
34
+ model_creator: CohereForAI
35
+ prompt_template: Cohere Command R
36
+ system_prompt: You are a helpful AI assistant
37
+ base_model: cohere
38
+ original_repo: CohereForAI/aya-23-35B
39
+ base_model: CohereForAI/aya-23-35B
40
  ---
41
+ ## ๐Ÿ’ซ Community Model> Aya 23 35B by Cohere For AI
42
 
43
+ *๐Ÿ‘พ [LM Studio](https://lmstudio.ai) Community models highlights program. Highlighting new & noteworthy models by the community. Join the conversation on [Discord](https://discord.gg/aPQfnNkxGC)*.
44
 
45
+ **Model creator:** [Cohere For AI](https://huggingface.co/CohereForAI)<br>
46
+ **Original model**: [aya-23-35B](https://huggingface.co/CohereForAI/aya-23-35B)<br>
47
+ **GGUF quantization:** provided by [bartowski](https://huggingface.co/bartowski) based on `llama.cpp` release [b2965](https://github.com/ggerganov/llama.cpp/releases/tag/b2965)<br>
48
 
49
+ ## Model Summary:
50
 
51
+ Aya 23 are brand new instruction tuned multilingual models from Cohere. This model should perform well at logic across a wide variety of languages.<br>
52
+ This is the 35B version of the model. The performance is quite high especially when used for multilingual tasks where most other models in this size range lack training data.
53
 
54
+ ## Prompt template:
55
 
56
+ Choose the `Cohere Command R` preset in your LM Studio.
 
 
57
 
58
+ Under the hood, the model will see a prompt that's formatted like so:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
59
 
60
  ```
61
+ <BOS_TOKEN><|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>
62
+ {system_prompt}
63
+ <|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|USER_TOKEN|>
64
+ {prompt}
65
+ <|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
 
 
 
66
 
 
 
 
 
67
  ```
68
 
69
+ ## Technical Details
 
 
 
 
 
 
 
 
 
 
70
 
71
+ Aya 23 covers the following languages:
72
+ - Arabic, Chinese (simplified & traditional), Czech, Dutch, English, French, German, Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Korean, Persian, Polish, Portuguese, Romanian, Russian, Spanish, Turkish, Ukrainian, and Vietnamese
73
 
74
+ The Aya training dataset can be found here:
75
+ - https://huggingface.co/datasets/CohereForAI/aya_collection
76
 
77
+ More technical details can be found from Cohere [here](https://cohere.com/research/papers/aya-command-23-35b-and-35b-technical-report-2024-05-23)
78
 
79
+ ## Special thanks
80
 
81
+ ๐Ÿ™ Special thanks to [Georgi Gerganov](https://github.com/ggerganov) and the whole team working on [llama.cpp](https://github.com/ggerganov/llama.cpp/) for making all of this possible.
82
 
83
+ ๐Ÿ™ Special thanks to [Kalomaze](https://github.com/kalomaze), [Dampf](https://github.com/Dampfinchen) and [turboderp](https://github.com/turboderp/) for their work on the dataset (linked [here](https://gist.github.com/bartowski1182/b6ac44691e994344625687afe3263b3a)) that was used for calculating the imatrix for all sizes.
84
 
85
+ ## Disclaimers
86
 
87
+ LM Studio is not the creator, originator, or owner of any Model featured in the Community Model Program. Each Community Model is created and provided by third parties. LM Studio does not endorse, support, represent or guarantee the completeness, truthfulness, accuracy, or reliability of any Community Model. You understand that Community Models can produce content that might be offensive, harmful, inaccurate or otherwise inappropriate, or deceptive. Each Community Model is the sole responsibility of the person or entity who originated such Model. LM Studio may not monitor or control the Community Models and cannot, and does not, take responsibility for any such Model. LM Studio disclaims all warranties or guarantees about the accuracy, reliability or benefits of the Community Models. LM Studio further disclaims any warranty that the Community Model will meet your requirements, be secure, uninterrupted or available at any time or location, or error-free, viruses-free, or that any errors will be corrected, or otherwise. You will be solely responsible for any damage resulting from your use of or access to the Community Models, your downloading of any Community Model, or use of any other Community Model provided by or through LM Studio.