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

quantize : add '--keep-split' to quantize model into shards (#6688)

Browse source 
* Implement '--keep-split' to quantize model into several shards

* Add test script

* Update examples/quantize/quantize.cpp

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Split model correctly even if tensor id is out-of-order

* Update llama_model_quantize_params

* Fix preci failures

---------

Co-authored-by: z5269887 <z5269887@unsw.edu.au>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
jiez 2024-04-25 18:29:35 +08:00 • committed by GitHub
parent 784e11dea1
commit 1966eb2615
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 148 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

65
examples/quantize/test.sh Normal file
View file

@ -0,0 +1,65 @@
#!/bin/bash
set -eu
if [ $# -lt 1 ]
then
echo "usage: $0 path_to_build_binary [path_to_temp_folder]"
echo "example: $0 ../../build/bin ../../tmp"
exit 1
fi
if [ $# -gt 1 ]
then
TMP_DIR=$2
else
TMP_DIR=/tmp
fi
set -x
SPLIT=$1/gguf-split
QUANTIZE=$1/quantize
MAIN=$1/main
WORK_PATH=$TMP_DIR/quantize
CUR_DIR=$(pwd)
mkdir -p "$WORK_PATH"
# Clean up in case of previously failed test
rm -f $WORK_PATH/ggml-model-split*.gguf $WORK_PATH/ggml-model-requant*.gguf
# 1. Get a model
(
cd $WORK_PATH
"$CUR_DIR"/../../scripts/hf.sh --repo ggml-org/gemma-1.1-2b-it-Q8_0-GGUF --file gemma-1.1-2b-it.Q8_0.gguf
)
echo PASS
# 2. Split model
$SPLIT --split-max-tensors 28 $WORK_PATH/gemma-1.1-2b-it.Q8_0.gguf $WORK_PATH/ggml-model-split
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
echo PASS
echo
# 3a. Test the requanted model is loading properly
$MAIN --model $WORK_PATH/ggml-model-requant-00001-of-00006.gguf --random-prompt --n-predict 32
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
echo PASS
echo
# 4b. Test the requanted model is loading properly
$MAIN --model $WORK_PATH/ggml-model-requant-merge.gguf --random-prompt --n-predict 32
echo PASS
echo
# Clean up
rm -f $WORK_PATH/ggml-model-split*.gguf $WORK_PATH/ggml-model-requant*.gguf