test new proposed outfile naming convention changes to llama.cpp conversion script
Browse files- .args +1 -1
- .gitmodules +3 -3
- TinyLLama-v0.1-5M-F16.gguf → Tinyllama-5M-v0.2-F16.gguf +2 -2
- TinyLLama-v0.1-5M-F16.llamafile → Tinyllama-5M-v0.2-F16.llamafile +2 -2
- llama.cpp +1 -1
- llamafile +1 -1
- llamafile-creation-legacy.sh +52 -0
- llamafile-creation.sh +2 -2
- maykeye_tinyllama-metadata.json +2 -2
.args
CHANGED
@@ -1,2 +1,2 @@
|
|
1 |
-m
|
2 |
-
|
|
|
1 |
-m
|
2 |
+
Tinyllama-5M-v0.2-F16.gguf
|
.gitmodules
CHANGED
@@ -1,9 +1,9 @@
|
|
1 |
[submodule "maykeye_tinyllama"]
|
2 |
path = maykeye_tinyllama
|
3 |
url = https://huggingface.co/Maykeye/TinyLLama-v0
|
4 |
-
[submodule "llama.cpp"]
|
5 |
-
path = llama.cpp
|
6 |
-
url = git@github.com:ggerganov/llama.cpp.git
|
7 |
[submodule "llamafile"]
|
8 |
path = llamafile
|
9 |
url = git@github.com:Mozilla-Ocho/llamafile.git
|
|
|
|
|
|
|
|
1 |
[submodule "maykeye_tinyllama"]
|
2 |
path = maykeye_tinyllama
|
3 |
url = https://huggingface.co/Maykeye/TinyLLama-v0
|
|
|
|
|
|
|
4 |
[submodule "llamafile"]
|
5 |
path = llamafile
|
6 |
url = git@github.com:Mozilla-Ocho/llamafile.git
|
7 |
+
[submodule "llama.cpp"]
|
8 |
+
path = llama.cpp
|
9 |
+
url = git@github.com:mofosyne/llama.cpp.git
|
TinyLLama-v0.1-5M-F16.gguf → Tinyllama-5M-v0.2-F16.gguf
RENAMED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e0422a7e84b0d8a6ebc77513ec4fe53979850b9f6235265eba76e3e954072f86
|
3 |
+
size 10008256
|
TinyLLama-v0.1-5M-F16.llamafile → Tinyllama-5M-v0.2-F16.llamafile
RENAMED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0098e4ca8bbf84474fc65105d9149047d73ad964481182b954efae4ab5a9bfe9
|
3 |
+
size 19281727
|
llama.cpp
CHANGED
@@ -1 +1 @@
|
|
1 |
-
Subproject commit
|
|
|
1 |
+
Subproject commit 00ff73a90101c76108131a5867a3c3c78a42ee8c
|
llamafile
CHANGED
@@ -1 +1 @@
|
|
1 |
-
Subproject commit
|
|
|
1 |
+
Subproject commit 9cd8d70942a049ba3c3bddd12e87e1fb599fbd49
|
llamafile-creation-legacy.sh
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
|
3 |
+
MODEL_DIR="maykeye_tinyllama"
|
4 |
+
METADATA_FILE="maykeye_tinyllama-metadata.json"
|
5 |
+
|
6 |
+
###############################################################################
|
7 |
+
# Pull both model folder, llamafile (for the engine) and llama.cpp (for the conversion script)
|
8 |
+
echo == Prep Enviroment ==
|
9 |
+
git submodule update --init
|
10 |
+
|
11 |
+
###############################################################################
|
12 |
+
echo == Build and prep the llamafile engine execuable ==
|
13 |
+
pushd llamafile
|
14 |
+
make -j8
|
15 |
+
make
|
16 |
+
# This is where each executables is located for reference purpose for now as of 2024-04-05
|
17 |
+
# and was determined by running `sudo make install PREFIX=/usr/local`
|
18 |
+
# ./o/llamafile/zipalign --> /usr/local/bin/zipalign
|
19 |
+
# ./o/llama.cpp/main/main --> /usr/local/bin/llamafile
|
20 |
+
# ./o/llama.cpp/imatrix/imatrix --> /usr/local/bin/llamafile-imatrix
|
21 |
+
# ./o/llama.cpp/quantize/quantize --> /usr/local/bin/llamafile-quantize
|
22 |
+
# ./build/llamafile-convert --> /usr/local/bin/llamafile-convert
|
23 |
+
# ./o/llama.cpp/perplexity/perplexity --> /usr/local/bin/llamafile-perplexity
|
24 |
+
# ./o/llama.cpp/llava/llava-quantize --> /usr/local/bin/llava-quantize
|
25 |
+
popd
|
26 |
+
|
27 |
+
###############################################################################
|
28 |
+
echo == What is our llamafile name going to be? ==
|
29 |
+
OUTFILE=$(./llama.cpp/examples/convert-legacy-llama.py ${MODEL_DIR} --metadata ${METADATA_FILE} --outtype f16 --get-outfile)
|
30 |
+
echo We will be aiming to generate $OUTFILE.llamafile
|
31 |
+
|
32 |
+
###############################################################################
|
33 |
+
echo == Convert from safetensor to gguf ==
|
34 |
+
./llama.cpp/examples/convert-legacy-llama.py ${MODEL_DIR} --metadata ${METADATA_FILE} --outtype f16
|
35 |
+
mv ${MODEL_DIR}/${OUTFILE}.gguf ${OUTFILE}.gguf
|
36 |
+
|
37 |
+
###############################################################################
|
38 |
+
echo == Generating Llamafile ==
|
39 |
+
cp ./llamafile/o/llama.cpp/main/main ${OUTFILE}.llamafile
|
40 |
+
|
41 |
+
# Create an .args file with settings defaults
|
42 |
+
cat >.args <<EOF
|
43 |
+
-m
|
44 |
+
${OUTFILE}.gguf
|
45 |
+
EOF
|
46 |
+
|
47 |
+
# zip align engine, gguf and default args
|
48 |
+
./llamafile/o/llamafile/zipalign -j0 ${OUTFILE}.llamafile ${OUTFILE}.gguf .args
|
49 |
+
|
50 |
+
###############################################################################
|
51 |
+
echo == Test Output ==
|
52 |
+
./${OUTFILE}.llamafile --cli -p "hello world the gruff man said"
|
llamafile-creation.sh
CHANGED
@@ -26,12 +26,12 @@ popd
|
|
26 |
|
27 |
###############################################################################
|
28 |
echo == What is our llamafile name going to be? ==
|
29 |
-
OUTFILE=$(./llama.cpp/convert.py ${MODEL_DIR} --metadata ${METADATA_FILE} --outtype f16 --get-outfile)
|
30 |
echo We will be aiming to generate $OUTFILE.llamafile
|
31 |
|
32 |
###############################################################################
|
33 |
echo == Convert from safetensor to gguf ==
|
34 |
-
./llama.cpp/convert.py ${MODEL_DIR} --metadata ${METADATA_FILE} --outtype f16
|
35 |
mv ${MODEL_DIR}/${OUTFILE}.gguf ${OUTFILE}.gguf
|
36 |
|
37 |
###############################################################################
|
|
|
26 |
|
27 |
###############################################################################
|
28 |
echo == What is our llamafile name going to be? ==
|
29 |
+
OUTFILE=$(./llama.cpp/convert-hf-to-gguf.py ${MODEL_DIR} --metadata ${METADATA_FILE} --outtype f16 --get-outfile)
|
30 |
echo We will be aiming to generate $OUTFILE.llamafile
|
31 |
|
32 |
###############################################################################
|
33 |
echo == Convert from safetensor to gguf ==
|
34 |
+
./llama.cpp/convert-hf-to-gguf.py ${MODEL_DIR} --metadata ${METADATA_FILE} --outtype f16 --verbose
|
35 |
mv ${MODEL_DIR}/${OUTFILE}.gguf ${OUTFILE}.gguf
|
36 |
|
37 |
###############################################################################
|
maykeye_tinyllama-metadata.json
CHANGED
@@ -1,10 +1,10 @@
|
|
1 |
{
|
2 |
"general.name": "TinyLLama",
|
3 |
-
"general.version": "v0.
|
4 |
"general.author": "mofosyne",
|
5 |
"general.url": "https://huggingface.co/mofosyne/TinyLLama-v0-llamafile",
|
6 |
"general.description": "This gguf is ported from a first version of Maykeye attempt at recreating roneneldan/TinyStories-1M but using Llama architecture",
|
7 |
"general.license": "apache-2.0",
|
8 |
"general.source.url": "https://huggingface.co/Maykeye/TinyLLama-v0",
|
9 |
-
"general.source.huggingface.repository": "
|
10 |
}
|
|
|
1 |
{
|
2 |
"general.name": "TinyLLama",
|
3 |
+
"general.version": "v0.2",
|
4 |
"general.author": "mofosyne",
|
5 |
"general.url": "https://huggingface.co/mofosyne/TinyLLama-v0-llamafile",
|
6 |
"general.description": "This gguf is ported from a first version of Maykeye attempt at recreating roneneldan/TinyStories-1M but using Llama architecture",
|
7 |
"general.license": "apache-2.0",
|
8 |
"general.source.url": "https://huggingface.co/Maykeye/TinyLLama-v0",
|
9 |
+
"general.source.huggingface.repository": "Maykeye/TinyLLama-v0"
|
10 |
}
|