Skip to content
This repository was archived by the owner on Sep 10, 2025. It is now read-only.

Commit 3f747e4

Browse files
mikekgfbmalfet
mikekgfb
authored and
malfet
committed
Group size2groupsize (#189)
* convert group_size to groupsize * group_size to groupsize in REAADME.md
1 parent 2548f93 commit 3f747e4

File tree

10 files changed

+164
-164
lines changed

10 files changed

+164
-164
lines changed

‎.github/workflows/compile-bf16.yml‎

Lines changed: 15 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -58,44 +58,44 @@ jobs:
5858
echo "******************************************"
5959
echo "******* Emb: channel-wise quantized ******"
6060
echo "******************************************"
61-
python generate.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
61+
python generate.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
6262
cat ./output_eager
63-
python generate.py --dtype ${DTYPE} --compile --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
63+
python generate.py --dtype ${DTYPE} --compile --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
6464
cat ./output_compiled
65-
python export.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
65+
python export.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
6666
python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
6767
cat ./output_aoti
6868

6969
echo "******************************************"
7070
echo "******** Emb: group-wise quantized *******"
7171
echo "******************************************"
72-
python generate.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
72+
python generate.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
7373
cat ./output_eager
74-
python generate.py --dtype ${DTYPE} --compile --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
74+
python generate.py --dtype ${DTYPE} --compile --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
7575
cat ./output_compiled
76-
python export.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
76+
python export.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
7777
python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
7878
cat ./output_aoti
7979

8080
echo "******************************************"
8181
echo "******* INT8 channel-wise quantized ******"
8282
echo "******************************************"
83-
python generate.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
83+
python generate.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
8484
cat ./output_eager
85-
python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
85+
python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
8686
cat ./output_compiled
87-
python export.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
87+
python export.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
8888
python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
8989
cat ./output_aoti
9090

9191
echo "******************************************"
9292
echo "******** INT8 group-wise quantized *******"
9393
echo "******************************************"
94-
python generate.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
94+
python generate.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
9595
cat ./output_eager
96-
python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
96+
python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
9797
cat ./output_compiled
98-
python export.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
98+
python export.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
9999
python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
100100
cat ./output_aoti
101101

@@ -106,11 +106,11 @@ jobs:
106106
DTYPE=bfloat16
107107
fi
108108

109-
python generate.py --dtype ${DTYPE} --quant '{"linear:int4" : {"group_size": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
109+
python generate.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
110110
cat ./output_eager
111-
python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int4" : {"group_size": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
111+
python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
112112
cat ./output_compiled
113-
python export.py --dtype ${DTYPE} --quant '{"linear:int4" : {"group_size": 32}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
113+
python export.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
114114
python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
115115
cat ./output_aoti
116116

‎.github/workflows/compile-gguf.yml‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -56,22 +56,22 @@ jobs:
5656
echo "******************************************"
5757
echo "******* Emb: channel-wise quantized ******"
5858
echo "******************************************"
59-
python generate.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_eager
59+
python generate.py --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_eager
6060
cat ./output_eager
61-
python generate.py --compile --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_compiled
61+
python generate.py --compile --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_compiled
6262
cat ./output_compiled
63-
python export.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --gguf-path ${GGUF_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
63+
python export.py --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --gguf-path ${GGUF_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
6464
python generate.py --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
6565
cat ./output_aoti
6666
6767
echo "******************************************"
6868
echo "******** Emb: group-wise quantized *******"
6969
echo "******************************************"
70-
python generate.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_eager
70+
python generate.py --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_eager
7171
cat ./output_eager
72-
python generate.py --compile --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_compiled
72+
python generate.py --compile --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 > ./output_compiled
7373
cat ./output_compiled
74-
python export.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --gguf-path ${GGUF_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
74+
python export.py --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --gguf-path ${GGUF_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
7575
python generate.py --gguf-path ${GGUF_PATH} --tokenizer-path ${TOKENIZER_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti
7676
cat ./output_aoti
7777

0 commit comments

Comments
 (0)