diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-10-21 12:16:54 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-10-21 12:16:54 +0200 |
| commit | dbf951df1594a3dec36eca9ab81a0f7ba81b11cd (patch) | |
| tree | a1692718a4602927fe34dc313841aec582e7fe58 /ggml/src/ggml-cuda/template-instances/generate_cu_files.py | |
| parent | f2d315b46f7aacc7df4b86bd8acba387b30e11ca (diff) | |
Enable IQ4_NL for KV-cache in token generation using Flash Attention (#99)
* Enable IQ4_NL for V-cache in token generation
* We don't need these
* Update printour of allowed quantized KV-cache combinations
* Add IQ4_NL + IQ4_NL to FA
This is a better alternative than Q4_0 + Q4_0 for the VRAM poor.
* Remove file added by mistake
* Fix typo, which is not really a bug
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src/ggml-cuda/template-instances/generate_cu_files.py')
| -rwxr-xr-x | ggml/src/ggml-cuda/template-instances/generate_cu_files.py | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/ggml/src/ggml-cuda/template-instances/generate_cu_files.py b/ggml/src/ggml-cuda/template-instances/generate_cu_files.py index d7874e6e..1186112e 100755 --- a/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +++ b/ggml/src/ggml-cuda/template-instances/generate_cu_files.py @@ -3,7 +3,7 @@ from glob import glob import os -TYPES_KV = ["GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0", "GGML_TYPE_F16"] +TYPES_KV = ["GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0", "GGML_TYPE_IQ4_NL", "GGML_TYPE_F16"] SOURCE_FATTN_VEC = """// This file has been autogenerated by generate_cu_files.py, do not edit manually. |