Foke Singh
111afcdd95
Instead of giving config, model weights and tokenizer paths, rely on `huggingface-cli` download.
134 lines
3.2 KiB
Python
134 lines
3.2 KiB
Python
load("@aspect_bazel_lib//lib:expand_template.bzl", "expand_template")
|
|
load("@aspect_bazel_lib//lib:tar.bzl", "mtree_spec", "tar")
|
|
load("@aspect_bazel_lib//lib:transitions.bzl", "platform_transition_filegroup")
|
|
load("@bazel_skylib//rules:native_binary.bzl", "native_test")
|
|
load("@bazel_skylib//rules:write_file.bzl", "write_file")
|
|
load("@rules_cc//cc:cc_binary.bzl", "cc_binary")
|
|
load("@rules_oci//oci:defs.bzl", "oci_image", "oci_load", "oci_push")
|
|
load("@zml//bazel:zig.bzl", "zig_cc_binary")
|
|
|
|
zig_cc_binary(
|
|
name = "llama",
|
|
srcs = [
|
|
"llama.zig",
|
|
],
|
|
main = "main.zig",
|
|
deps = [
|
|
"@com_github_hejsil_clap//:clap",
|
|
"@zml//async",
|
|
"@zml//stdx",
|
|
"@zml//zml",
|
|
],
|
|
)
|
|
|
|
zig_cc_binary(
|
|
name = "test-implementation",
|
|
srcs = ["llama.zig"],
|
|
args = [
|
|
"--weights=$(location @Meta-Llama-3.1-8B-Instruct//:model.safetensors.index.json)",
|
|
"--config=$(location @Meta-Llama-3.1-8B-Instruct//:config.json)",
|
|
],
|
|
data = [
|
|
"@Meta-Llama-3.1-8B-Instruct//:config.json",
|
|
"@Meta-Llama-3.1-8B-Instruct//:model.safetensors.index.json",
|
|
],
|
|
main = "test.zig",
|
|
tags = [
|
|
"manual",
|
|
],
|
|
deps = [
|
|
"@zml//async",
|
|
"@zml//stdx",
|
|
"@zml//zml",
|
|
],
|
|
)
|
|
|
|
native_test(
|
|
name = "test_tokenizer",
|
|
src = "@zml//zml/tokenizer:main",
|
|
# Note: all Llama-3.x tokenizers are the same,
|
|
# but using the 3.2-1B version because downloading the tokenizer triggers downloading the model.
|
|
args = [
|
|
"--tokenizer=$(location @Meta-Llama-3.2-1B-Instruct//:tokenizer.json)",
|
|
"""--prompt='Examples of titles:
|
|
📉 Stock Market Trends
|
|
🍪 Perfect Chocolate Chip Recipe
|
|
Evolution of Music Streaming
|
|
Remote Work Productivity Tips
|
|
Artificial Intelligence in Healthcare
|
|
🎮 Video Game Development Insights
|
|
'""",
|
|
# this correspond to encoding with HF tokenizers, with bos=False
|
|
"--expected=41481,315,15671,512,9468,241,231,12937,8152,50730,198,9468,235,103,24118,39520,32013,26371,198,35212,3294,315,10948,45910,198,25732,5664,5761,1968,26788,198,9470,16895,22107,304,39435,198,9468,236,106,8519,4140,11050,73137,198",
|
|
],
|
|
data = ["@Meta-Llama-3.2-1B-Instruct//:tokenizer.json"],
|
|
tags = [
|
|
"manual",
|
|
],
|
|
)
|
|
|
|
mtree_spec(
|
|
name = "mtree",
|
|
srcs = [":llama"],
|
|
tags = [
|
|
"manual",
|
|
],
|
|
)
|
|
|
|
tar(
|
|
name = "archive",
|
|
srcs = [":llama"],
|
|
args = [
|
|
"--options",
|
|
"zstd:compression-level=9",
|
|
],
|
|
compress = "zstd",
|
|
mtree = ":mtree",
|
|
tags = [
|
|
"manual",
|
|
],
|
|
)
|
|
|
|
oci_image(
|
|
name = "image_",
|
|
base = "@distroless_cc_debian12_debug",
|
|
entrypoint = ["./{}/llama".format(package_name())],
|
|
tags = [
|
|
"manual",
|
|
],
|
|
tars = [
|
|
"@zml//runtimes:layers",
|
|
":archive",
|
|
],
|
|
)
|
|
|
|
platform_transition_filegroup(
|
|
name = "image",
|
|
srcs = [":image_"],
|
|
tags = [
|
|
"manual",
|
|
],
|
|
target_platform = "@zml//platforms:linux_amd64",
|
|
)
|
|
|
|
oci_load(
|
|
name = "load",
|
|
image = ":image",
|
|
repo_tags = [
|
|
"distroless/llama:latest",
|
|
],
|
|
tags = [
|
|
"manual",
|
|
],
|
|
)
|
|
|
|
oci_push(
|
|
name = "push",
|
|
image = ":image",
|
|
remote_tags = ["latest"],
|
|
repository = "index.docker.io/steeve/llama",
|
|
tags = [
|
|
"manual",
|
|
],
|
|
)
|