Update GGML models for latest ggml commit with ggml version included.
Browse files
README.md
CHANGED
|
@@ -31,12 +31,12 @@ Please note that these MPT GGMLs are **not compatbile with llama.cpp**. Right no
|
|
| 31 |
## Provided files
|
| 32 |
| Name | Quant method | Bits | Size | RAM required | Use case |
|
| 33 |
| ---- | ---- | ---- | ---- | ---- | ----- |
|
| 34 |
-
`mpt-7b.
|
| 35 |
-
`mpt-7b.
|
| 36 |
-
`mpt-7b.
|
| 37 |
-
`mpt-7b.
|
| 38 |
-
`mpt-7b.
|
| 39 |
-
`mpt-7b.
|
| 40 |
|
| 41 |
## Compatibilty
|
| 42 |
|
|
@@ -57,7 +57,7 @@ mkdir build
|
|
| 57 |
cd build
|
| 58 |
cmake ..
|
| 59 |
cmake --build . --config Release
|
| 60 |
-
bin/mpt -m /path/to/mpt-7b.
|
| 61 |
```
|
| 62 |
|
| 63 |
Please see the ggml repo for other build options.
|
|
|
|
| 31 |
## Provided files
|
| 32 |
| Name | Quant method | Bits | Size | RAM required | Use case |
|
| 33 |
| ---- | ---- | ---- | ---- | ---- | ----- |
|
| 34 |
+
`mpt-7b.ggmlv3.q4_0.bin` | q4_0 | 4bit | 4.16GB | 6.2GB | 4-bit. |
|
| 35 |
+
`mpt-7b.ggmlv3.q4_1.bin` | q4_0 | 4bit | 4.99GB | 7.2GB | 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models. |
|
| 36 |
+
`mpt-7b.ggmlv3.q5_0.bin` | q5_0 | 5bit | 4.57GB | 6.8GB | 5-bit. Higher accuracy, higher resource usage and slower inference. |
|
| 37 |
+
`mpt-7b.ggmlv3.q5_1.bin` | q5_1 | 5bit | 4,99GB | 7.2GB | 5-bit. Even higher accuracy, and higher resource usage and slower inference. |
|
| 38 |
+
`mpt-7b.ggmlv3.q8_0.bin` | q8_0 | 8bit | 7.48GB | 9.6GB | 8-bit. Almost indistinguishable from float16. Huge resource use and slow. Not recommended for normal use. |
|
| 39 |
+
`mpt-7b.ggmlv3.fp16.bin` | fp16 | 16bit | 13.3GB | 15.5GB | Full 16-bit. |
|
| 40 |
|
| 41 |
## Compatibilty
|
| 42 |
|
|
|
|
| 57 |
cd build
|
| 58 |
cmake ..
|
| 59 |
cmake --build . --config Release
|
| 60 |
+
bin/mpt -m /path/to/mpt-7b.ggmlv3.q4_0.bin -t 8 -n 512 -p "Write a story about llamas"
|
| 61 |
```
|
| 62 |
|
| 63 |
Please see the ggml repo for other build options.
|