Spaces:
Sleeping
Sleeping
Neo Zhang Jianyu
commited on
Support multiple GPUs (split mode) on SYCL backend (llama/5806)
Browse files* suport multiple cards: split-mode - layer|row
* rm warning
* rebase with master, support tow new OPs, close feature for -sm=row, fix for unit test
* update news
* fix merge error
* update according to review comments
- ggml-sycl.cpp +0 -0
- ggml-sycl.h +5 -0
ggml-sycl.cpp
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ggml-sycl.h
CHANGED
|
@@ -24,6 +24,11 @@ GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type(void);
|
|
| 24 |
GGML_API void ggml_backend_sycl_print_sycl_devices(void);
|
| 25 |
GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
|
| 26 |
GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
#ifdef __cplusplus
|
| 28 |
}
|
| 29 |
#endif
|
|
|
|
| 24 |
GGML_API void ggml_backend_sycl_print_sycl_devices(void);
|
| 25 |
GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
|
| 26 |
GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
|
| 27 |
+
GGML_API GGML_CALL int ggml_backend_sycl_get_device_count();
|
| 28 |
+
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split);
|
| 29 |
+
GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
|
| 30 |
+
GGML_API GGML_CALL int ggml_backend_sycl_get_device_index(int device_id);
|
| 31 |
+
|
| 32 |
#ifdef __cplusplus
|
| 33 |
}
|
| 34 |
#endif
|