vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix preci failures

Browse source
This commit is contained in:
z5269887 2024-04-23 20:50:00 +08:00
parent 141eb5107f
commit e0a3679aeb
2 changed files with 5 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

6
examples/quantize/test.sh
View file

@ -42,7 +42,7 @@ echo PASS
echo
# 3. Requant model with '--keep_split'
$QUANTIZE --allow-requantize --keep_split $WORK_PATH/ggml-model-split-00001-of-00006.gguf $WORK_PATH/ggml-model-requant.gguf Q4_K
$QUANTIZE --allow-requantize --keep_split $WORK_PATH/ggml-model-split-00001-of-00006.gguf $WORK_PATH/ggml-model-requant.gguf Q4_K
echo PASS
echo
@ -52,7 +52,7 @@ echo PASS
echo
# 4. Requant mode without '--keep_split'
$QUANTIZE --allow-requantize $WORK_PATH/ggml-model-split-00001-of-00006.gguf $WORK_PATH/ggml-model-requant-merge.gguf Q4_K
$QUANTIZE --allow-requantize $WORK_PATH/ggml-model-split-00001-of-00006.gguf $WORK_PATH/ggml-model-requant-merge.gguf Q4_K
echo PASS
echo
@ -62,4 +62,4 @@ echo PASS
echo
# Clean up
rm -f $WORK_PATH/ggml-model-split*.gguf $WORK_PATH/ggml-model-requant*.gguf
rm -f $WORK_PATH/ggml-model-split*.gguf $WORK_PATH/ggml-model-requant*.gguf

4
llama.cpp
View file

@ -13618,13 +13618,13 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
// Set split info if needed
if (n_split > 1) {
for (int i = 0; i < ctx_outs.size(); ++i) {
for (size_t i = 0; i < ctx_outs.size(); ++i) {
gguf_set_val_u16(ctx_outs[i], ml.llm_kv(LLM_KV_SPLIT_NO).c_str(), i);
gguf_set_val_u16(ctx_outs[i], ml.llm_kv(LLM_KV_SPLIT_COUNT).c_str(), n_split);
gguf_set_val_i32(ctx_outs[i], ml.llm_kv(LLM_KV_SPLIT_TENSORS_COUNT).c_str(), ml.n_tensors);
}
}
int cur_split = -1;
std::ofstream fout;
auto close_ofstream = [&]() {