root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-14 23:09:53 +00:00
Code Issues Actions 14 Packages Projects Releases Wiki Activity

ggml : add asserts for type conversion in fattn kernels (#9971)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-10-21 16:20:46 +03:00 committed by GitHub
parent d5ebd79c76
commit f594bc80ba
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
common/common.cpp
View File

@ -1035,7 +1035,7 @@ static ggml_type kv_cache_type_from_str(const std::string & s) {
return GGML_TYPE_Q5_1; return GGML_TYPE_Q5_1;
} }
throw std::runtime_error("Invalid cache type: " + s); throw std::runtime_error("Unsupported cache type: " + s);
} }
struct llama_context_params common_context_params_to_llama(const common_params & params) { struct llama_context_params common_context_params_to_llama(const common_params & params) {

6
ggml/src/ggml.c
View File

@ -324,8 +324,9 @@ struct ggml_logger_state {
static struct ggml_logger_state g_logger_state = {ggml_log_callback_default, NULL}; static struct ggml_logger_state g_logger_state = {ggml_log_callback_default, NULL};
static void ggml_log_internal_v(enum ggml_log_level level, const char * format, va_list args) { static void ggml_log_internal_v(enum ggml_log_level level, const char * format, va_list args) {
if (format == NULL) if (format == NULL) {
return; return;
}
va_list args_copy; va_list args_copy;
va_copy(args_copy, args); va_copy(args_copy, args);
char buffer[128]; char buffer[128];
@ -15723,6 +15724,9 @@ static void ggml_compute_forward_flash_attn_ext_f16(
ggml_vec_dot_t const kq_vec_dot = type_traits[k->type].vec_dot; ggml_vec_dot_t const kq_vec_dot = type_traits[k->type].vec_dot;
ggml_to_float_t const v_to_float = type_traits[v->type].to_float; ggml_to_float_t const v_to_float = type_traits[v->type].to_float;
GGML_ASSERT(q_to_vec_dot && "fattn: unsupported K-type");
GGML_ASSERT(v_to_float && "fattn: unsupported V-type");
// loop over n_batch and n_head // loop over n_batch and n_head
for (int ir = ir0; ir < ir1; ++ir) { for (int ir = ir0; ir < ir1; ++ir) {
// q indices // q indices

2
src/llama.cpp
View File

@ -19243,7 +19243,7 @@ struct llama_context * llama_new_context_with_model(
params.flash_attn = false; params.flash_attn = false;
} }
if (params.type_v != GGML_TYPE_F16 && !params.flash_attn) { if (ggml_is_quantized(params.type_v) && !params.flash_attn) {
LLAMA_LOG_ERROR("%s: V cache quantization requires flash_attn\n", __func__); LLAMA_LOG_ERROR("%s: V cache quantization requires flash_attn\n", __func__);
return nullptr; return nullptr;
} }