furyhawk/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-07-21 10:15:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Little maintenance (#1579)

Browse Source 
* Little maintenance

* llama-quantize : Add the missing items in the help

* Add GGML_MAX_CONTEXTS define in the general cmakelist.txt

* Make the KV cache (CPU) based warnings clearer

* Correct placement of GGML_MAX_CONTEXTS definition

* Revert wrong indents

This reverts commit d0728cbb6c.

* Moving the GGML_MAX_CONTEXTS definition to src/CMakeLists.txt

* Update warning message for unsupported KV cache types

* forgotten antislash
This commit is contained in:
Nexes the Elder
2026-04-08 07:58:49 +02:00
committed by GitHub
parent 67fc9c5eb9
commit 0a6e4335f7
3 changed files with 13 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/quantize/quantize.cpp
+3 -1
View File
@@ -151,12 +151,14 @@ static bool try_parse_ftype(const std::string & ftype_str_in, llama_ftype & ftyp
//
[[noreturn]]
static void usage(const char * executable) {
printf("usage: %s [--help] [--allow-requantize] [--leave-output-tensor] [--pure] [--imatrix] [--hide-imatrix] [--include-weights] [--exclude-weights] [--output-tensor-type] [--token-embedding-type] [--ffn-gate-inp-type] [--attn-q-type] [--attn-k-type] [--attn-v-type] [--attn-qkv-type] [--attn-output-type] [--ffn-gate-type] [--ffn-down-type] [--ffn-up-type] [--keep-split] [--partial-requant] [--override-kv] model-f32.gguf [model-quant.gguf] type [nthreads]\n\n", executable);
printf("usage: %s [--help] [--allow-requantize] [--leave-output-tensor] [--pure] [--imatrix] [--hide-imatrix] [--ignore-imatrix-rules] [--dry-run] [--include-weights] [--exclude-weights] [--output-tensor-type] [--token-embedding-type] [--ffn-gate-inp-type] [--attn-q-type] [--attn-k-type] [--attn-v-type] [--attn-qkv-type] [--attn-output-type] [--ffn-gate-type] [--ffn-down-type] [--ffn-up-type] [--repack] [--repack-pattern] [--keep-split] [--partial-requant] [--override-kv] model-f32.gguf [model-quant.gguf] type [nthreads]\n\n", executable);
printf(" --allow-requantize: Allows requantizing tensors that have already been quantized. Warning: This can severely reduce quality compared to quantizing from 16bit or 32bit\n");
printf(" --leave-output-tensor: Will leave output.weight un(re)quantized. Increases model size but may also increase quality, especially when requantizing\n");
printf(" --pure: Disable k-quant mixtures and quantize all tensors to the same type\n");
printf(" --imatrix file_name: use data in file_name as importance matrix for quant optimizations\n");
printf(" --hide-imatrix: do not store imatrix details in the quantized model\n");
printf(" --ignore-imatrix-rules: ignore importance matrix rules when quantizing\n");
printf(" --dry-run: show what would be quantized without actually writing the output file\n");
printf(" --include-weights tensor_name: use importance matrix for this/these tensor(s)\n");
printf(" --exclude-weights tensor_name: use importance matrix for this/these tensor(s)\n");
printf(" --output-tensor-type ggml_type: use this ggml_type for the output.weight tensor.\n");
ggml/src/iqk/iqk_flash_attn.cpp
+6 -2
View File
@@ -169,15 +169,19 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float
if (auto type_k = ggml_type(int_type_k_in), type_v = ggml_type(int_type_v); !are_kv_types_supported(type_k, type_v)) {
if (ith == 0) {
fprintf(stderr, "\n==================== KV cache types %s, %s are not supported on the CPU\n",
fprintf(stderr, "\n==================== K cache %s coupled with V cache %s is not a supported combination on the CPU backend.\n",
ggml_type_name(type_k), ggml_type_name(type_v));
auto & supported = supported_kv_types();
fprintf(stderr, "Sopprted types are:\n");
fprintf(stderr, "Supported types are:\n");
for (auto type : supported) {
fprintf(stderr, " %s\n", ggml_type_name(type));
}
fprintf(stderr, " Warning: ik_llama.cpp does not support Q5_0 or Q5_1 KV cache on the CPU.\n");
#ifdef __AVX512BF16__
fprintf(stderr, " %s, but only if K and V are both %s\n", ggml_type_name(GGML_TYPE_BF16), ggml_type_name(GGML_TYPE_BF16));
#endif
#ifndef GGML_IQK_FA_ALL_QUANTS
fprintf(stderr, " To enable q4_0, q4_1, and iq4_nl KV cache types, recompile with -DGGML_IQK_FA_ALL_QUANTS=ON\n");
#endif
}
barrier(barrier_data);
src/CMakeLists.txt
+4
View File
@@ -72,3 +72,7 @@ if (BUILD_SHARED_LIBS)
set_target_properties(llama PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_compile_definitions(llama PRIVATE LLAMA_SHARED LLAMA_BUILD)
endif()
if (GGML_MAX_CONTEXTS)
add_compile_definitions(GGML_MAX_CONTEXTS=${GGML_MAX_CONTEXTS})
endif()