Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,30 +1,17 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
license_link: https://huggingface.co/microsoft/Phi-3-medium-4k-instruct/resolve/main/LICENSE
|
| 4 |
-
|
| 5 |
-
language:
|
| 6 |
-
- multilingual
|
| 7 |
-
pipeline_tag: text-generation
|
| 8 |
-
tags:
|
| 9 |
-
- nlp
|
| 10 |
-
- code
|
| 11 |
-
inference:
|
| 12 |
-
parameters:
|
| 13 |
-
temperature: 0.7
|
| 14 |
-
widget:
|
| 15 |
-
- messages:
|
| 16 |
-
- role: user
|
| 17 |
-
content: Can you provide ways to eat combinations of bananas and dragonfruits?
|
| 18 |
quantized_by: bartowski
|
|
|
|
| 19 |
---
|
| 20 |
|
| 21 |
## Llamacpp imatrix Quantizations of Phi-3-medium-4k-instruct
|
| 22 |
|
| 23 |
-
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a>
|
| 24 |
|
| 25 |
Original model: https://huggingface.co/microsoft/Phi-3-medium-4k-instruct
|
| 26 |
|
| 27 |
-
All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/
|
|
|
|
|
|
|
| 28 |
|
| 29 |
## Prompt format
|
| 30 |
|
|
@@ -34,30 +21,42 @@ All quants made using imatrix option with dataset from [here](https://gist.githu
|
|
| 34 |
|
| 35 |
## Download a file (not the whole branch) from below:
|
| 36 |
|
| 37 |
-
| Filename | Quant type | File Size | Description |
|
| 38 |
-
| -------- | ---------- | --------- | ----------- |
|
| 39 |
-
| [Phi-3-medium-4k-instruct-
|
| 40 |
-
| [Phi-3-medium-4k-instruct-
|
| 41 |
-
| [Phi-3-medium-4k-instruct-
|
| 42 |
-
| [Phi-3-medium-4k-instruct-
|
| 43 |
-
| [Phi-3-medium-4k-instruct-
|
| 44 |
-
| [Phi-3-medium-4k-instruct-
|
| 45 |
-
| [Phi-3-medium-4k-instruct-
|
| 46 |
-
| [Phi-3-medium-4k-instruct-
|
| 47 |
-
| [Phi-3-medium-4k-instruct-
|
| 48 |
-
| [Phi-3-medium-4k-instruct-
|
| 49 |
-
| [Phi-3-medium-4k-instruct-
|
| 50 |
-
| [Phi-3-medium-4k-instruct-
|
| 51 |
-
| [Phi-3-medium-4k-instruct-
|
| 52 |
-
| [Phi-3-medium-4k-instruct-
|
| 53 |
-
| [Phi-3-medium-4k-instruct-
|
| 54 |
-
| [Phi-3-medium-4k-instruct-
|
| 55 |
-
| [Phi-3-medium-4k-instruct-
|
| 56 |
-
| [Phi-3-medium-4k-instruct-
|
| 57 |
-
| [Phi-3-medium-4k-instruct-
|
| 58 |
-
| [Phi-3-medium-4k-instruct-
|
| 59 |
-
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
## Downloading using huggingface-cli
|
| 63 |
|
|
@@ -76,7 +75,7 @@ huggingface-cli download bartowski/Phi-3-medium-4k-instruct-GGUF --include "Phi-
|
|
| 76 |
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:
|
| 77 |
|
| 78 |
```
|
| 79 |
-
huggingface-cli download bartowski/Phi-3-medium-4k-instruct-GGUF --include "Phi-3-medium-4k-instruct-Q8_0
|
| 80 |
```
|
| 81 |
|
| 82 |
You can either specify a new local-dir (Phi-3-medium-4k-instruct-Q8_0) or download them all in place (./)
|
|
@@ -106,3 +105,4 @@ These I-quants can also be used on CPU and Apple Metal, but will be slower than
|
|
| 106 |
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.
|
| 107 |
|
| 108 |
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
|
|
|
|
|
|
| 1 |
+
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
quantized_by: bartowski
|
| 3 |
+
pipeline_tag: text-generation
|
| 4 |
---
|
| 5 |
|
| 6 |
## Llamacpp imatrix Quantizations of Phi-3-medium-4k-instruct
|
| 7 |
|
| 8 |
+
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3509">b3509</a> for quantization.
|
| 9 |
|
| 10 |
Original model: https://huggingface.co/microsoft/Phi-3-medium-4k-instruct
|
| 11 |
|
| 12 |
+
All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
|
| 13 |
+
|
| 14 |
+
Run them in [LM Studio](https://lmstudio.ai/)
|
| 15 |
|
| 16 |
## Prompt format
|
| 17 |
|
|
|
|
| 21 |
|
| 22 |
## Download a file (not the whole branch) from below:
|
| 23 |
|
| 24 |
+
| Filename | Quant type | File Size | Split | Description |
|
| 25 |
+
| -------- | ---------- | --------- | ----- | ----------- |
|
| 26 |
+
| [Phi-3-medium-4k-instruct-f32.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/tree/main/Phi-3-medium-4k-instruct-f32) | f32 | 55.84GB | true | Full F32 weights. |
|
| 27 |
+
| [Phi-3-medium-4k-instruct-Q8_0.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q8_0.gguf) | Q8_0 | 14.83GB | false | Extremely high quality, generally unneeded but max available quant. |
|
| 28 |
+
| [Phi-3-medium-4k-instruct-Q6_K_L.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q6_K_L.gguf) | Q6_K_L | 11.53GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, *recommended*. |
|
| 29 |
+
| [Phi-3-medium-4k-instruct-Q6_K.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q6_K.gguf) | Q6_K | 11.45GB | false | Very high quality, near perfect, *recommended*. |
|
| 30 |
+
| [Phi-3-medium-4k-instruct-Q5_K_L.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q5_K_L.gguf) | Q5_K_L | 10.18GB | false | Uses Q8_0 for embed and output weights. High quality, *recommended*. |
|
| 31 |
+
| [Phi-3-medium-4k-instruct-Q5_K_M.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q5_K_M.gguf) | Q5_K_M | 10.07GB | false | High quality, *recommended*. |
|
| 32 |
+
| [Phi-3-medium-4k-instruct-Q5_K_S.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q5_K_S.gguf) | Q5_K_S | 9.62GB | false | High quality, *recommended*. |
|
| 33 |
+
| [Phi-3-medium-4k-instruct-Q4_K_L.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q4_K_L.gguf) | Q4_K_L | 8.69GB | false | Uses Q8_0 for embed and output weights. Good quality, *recommended*. |
|
| 34 |
+
| [Phi-3-medium-4k-instruct-Q4_K_M.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q4_K_M.gguf) | Q4_K_M | 8.57GB | false | Good quality, default size for must use cases, *recommended*. |
|
| 35 |
+
| [Phi-3-medium-4k-instruct-Q4_K_S.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q4_K_S.gguf) | Q4_K_S | 7.95GB | false | Slightly lower quality with more space savings, *recommended*. |
|
| 36 |
+
| [Phi-3-medium-4k-instruct-Q3_K_XL.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q3_K_XL.gguf) | Q3_K_XL | 7.63GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |
|
| 37 |
+
| [Phi-3-medium-4k-instruct-Q3_K_L.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q3_K_L.gguf) | Q3_K_L | 7.49GB | false | Lower quality but usable, good for low RAM availability. |
|
| 38 |
+
| [Phi-3-medium-4k-instruct-IQ4_XS.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-IQ4_XS.gguf) | IQ4_XS | 7.47GB | false | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
|
| 39 |
+
| [Phi-3-medium-4k-instruct-Q3_K_M.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q3_K_M.gguf) | Q3_K_M | 6.92GB | false | Low quality. |
|
| 40 |
+
| [Phi-3-medium-4k-instruct-IQ3_M.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-IQ3_M.gguf) | IQ3_M | 6.47GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
|
| 41 |
+
| [Phi-3-medium-4k-instruct-Q3_K_S.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q3_K_S.gguf) | Q3_K_S | 6.06GB | false | Low quality, not recommended. |
|
| 42 |
+
| [Phi-3-medium-4k-instruct-IQ3_XS.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-IQ3_XS.gguf) | IQ3_XS | 5.81GB | false | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
|
| 43 |
+
| [Phi-3-medium-4k-instruct-Q2_K_L.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q2_K_L.gguf) | Q2_K_L | 5.30GB | false | Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable. |
|
| 44 |
+
| [Phi-3-medium-4k-instruct-Q2_K.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-Q2_K.gguf) | Q2_K | 5.14GB | false | Very low quality but surprisingly usable. |
|
| 45 |
+
| [Phi-3-medium-4k-instruct-IQ2_M.gguf](https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF/blob/main/Phi-3-medium-4k-instruct-IQ2_M.gguf) | IQ2_M | 4.72GB | false | Relatively low quality, uses SOTA techniques to be surprisingly usable. |
|
| 46 |
+
|
| 47 |
+
## Embed/output weights
|
| 48 |
+
|
| 49 |
+
Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.
|
| 50 |
+
|
| 51 |
+
Some say that this improves the quality, others don't notice any difference. If you use these models PLEASE COMMENT with your findings. I would like feedback that these are actually used and useful so I don't keep uploading quants no one is using.
|
| 52 |
+
|
| 53 |
+
Thanks!
|
| 54 |
+
|
| 55 |
+
## Credits
|
| 56 |
+
|
| 57 |
+
Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset
|
| 58 |
+
|
| 59 |
+
Thank you ZeroWw for the inspiration to experiment with embed/output
|
| 60 |
|
| 61 |
## Downloading using huggingface-cli
|
| 62 |
|
|
|
|
| 75 |
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:
|
| 76 |
|
| 77 |
```
|
| 78 |
+
huggingface-cli download bartowski/Phi-3-medium-4k-instruct-GGUF --include "Phi-3-medium-4k-instruct-Q8_0/*" --local-dir ./
|
| 79 |
```
|
| 80 |
|
| 81 |
You can either specify a new local-dir (Phi-3-medium-4k-instruct-Q8_0) or download them all in place (./)
|
|
|
|
| 105 |
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.
|
| 106 |
|
| 107 |
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
|
| 108 |
+
|