ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

CLI - Specify GGML_TYPE to quantize for the main tensors. (#91)

Browse Source 
To complement the token_embd.weight and output.weight :

attn_v.weight
attn_k.weight.
attn_q_weight
attn_output.weight
attn_qkv.weight
ffn_gate
ffn_down
ffn_up
This commit is contained in:
Nexes the Elder
2024-10-18 09:48:15 +02:00
committed by GitHub
parent f369c6f921
commit 2b1af6bade
3 changed files with 125 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
examples/quantize/quantize.cpp
View File

@@ -109,19 +109,35 @@ 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] [--include-weights] [--exclude-weights] [--output-tensor-type] [--token-embedding-type] [--override-kv] model-f32.gguf [model-quant.gguf] type [nthreads]\n\n", executable);
printf("usage: %s [--help] [--allow-requantize] [--leave-output-tensor] [--pure] [--imatrix] [--include-weights] [--exclude-weights] [--output-tensor-type] [--token-embedding-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] [--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(" --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");
printf(" --token-embedding-type ggml_type: use this ggml_type for the token embeddings tensor\n");
printf(" --keep-split: will generate quatized model in the same shards as input");
printf(" --output-tensor-type ggml_type: use this ggml_type for the output.weight tensor.\n");
printf(" --token-embedding-type ggml_type: use this ggml_type for the token_embd.weight tensor.\n\n");
printf("Additional specific tensor quantization types used in the custom quant scheme 'CQS (default is Q2_K):\n");
printf(" --attn-q-type ggml_type: use this ggml_type for the attn_q.weight tensor.\n");
printf(" --attn-k-type ggml_type: use this ggml_type for the attn_k.weight tensor.\n");
printf(" --attn-v-type ggml_type: use this ggml_type for the attn_v.weight tensor.\n");
printf(" --attn-qkv-type ggml_type: use this ggml_type for the attn_qkv.weight tensor.\n");
printf(" --attn-output-type ggml_type: use this ggml_type for the attn_output.weight tensor.\n");
printf(" --ffn-gate-type ggml_type: use this ggml_type for the ffn_gate tensor.\n");
printf(" --ffn-down-type ggml_type: use this ggml_type for the ffn_down tensor.\n");
printf(" --ffn-up-type ggml_type: use this ggml_type for the ffn_up tensor.\n\n");
printf(" --keep-split: will generate quantized model in the same shards as input\n");
printf(" --override-kv KEY=TYPE:VALUE\n");
printf(" Advanced option to override model metadata by key in the quantized model. May be specified multiple times.\n");
printf(" Advanced option to override model metadata by key in the quantized model. May be specified multiple times.\n\n");
printf("Note: --include-weights and --exclude-weights cannot be used together\n");
printf("Note: The token embeddings tensor is loaded in system RAM, even in case of full GPU/VRAM offload.\n");
printf("Note: The recommanded type for the output tensor is q6_K for the ffn types > iq3_xxs and < q8_0.\n\n");
printf("Note for the Custom Quant Scheme FTYPE:\n");
printf(" Write the specific tensor legacy quants as qN_N, the K-Quants as qN_K, the IQ-Quants as iqN_xx.\n");
printf(" Usually, attn-q-type can be one type below the chosen ffn type, and attn-v-type should be one type above.\n");
printf(" attn-qkv-type replaces the types attn-q, attn-k and attn-v on some models.\n");
//TODO: - eventually - harmonize the CAPS writing of the FTYPEs, and non CAPS writing of the GGML_TYPEs.
printf("\nAllowed quantization types:\n");
for (auto & it : QUANT_OPTIONS) {
if (it.name != "COPY") {
@@ -277,6 +293,54 @@ int main(int argc, char ** argv) {
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--attn-q-type") == 0) {
if (arg_idx < argc-1) {
params.attn_q_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--attn-k-type") == 0) {
if (arg_idx < argc-1) {
params.attn_k_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--attn-v-type") == 0) {
if (arg_idx < argc-1) {
params.attn_v_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--attn-qkv-type") == 0) {
if (arg_idx < argc-1) {
params.attn_qkv_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--attn-output-type") == 0) {
if (arg_idx < argc-1) {
params.attn_output_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--ffn-gate-type") == 0) {
if (arg_idx < argc-1) {
params.ffn_gate_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--ffn-down-type") == 0) {
if (arg_idx < argc-1) {
params.ffn_down_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--ffn-up-type") == 0) {
if (arg_idx < argc-1) {
params.ffn_up_type = parse_ggml_type(argv[++arg_idx]);
} else {
usage(argv[0]);
}
} else if (strcmp(argv[arg_idx], "--override-kv") == 0) {
if (arg_idx == argc-1 || !string_parse_kv_override(argv[++arg_idx], kv_overrides)) {
usage(argv[0]);