argmaxinc / mlx-bench Goto Github PK
View Code? Open in Web Editor NEWBenchmark MLX Performance across commits
Benchmark MLX Performance across commits
I ran the bench command recommended in the README to test out MLX and saw an error in test_utils.py
possibly due to the way sw_vers
result is being unpacked.
Please consider my recommendation to change test_utils.py
to be
os_type, os_version, os_build_number, *spurious = [
line.rsplit("\t\t")[1]
for line in sw_vers.rsplit("\n")
]
instead of what it is currently
os_type, os_version, os_build_number = [....
Logs:
↪ python bench_mistral.py \
--repo-a ml-explore/mlx --commit-a cbcf44a4caf3fb504ed29ef78091126134e197a3 \
--repo-b ml-explore/mlx --commit-b 14b4e51a7c6455a61a74d24da9f47dfeb161023f \
--output-dir external --hub-model-name mlx-community/Mistral-7B-Instruct-v0.2-4-bit \
--max-context-length 800 \
--fail-for-mismatch-before-n-tokens 800
WARNING:coremltools:scikit-learn version 1.3.2 is not supported. Minimum required version: 0.17. Maximum required version: 1.1.2. Disabling scikit-learn conversion API.
WARNING:coremltools:Torch version 2.1.2 has not been tested with coremltools. You may run into unexpected errors. Torch 2.1.0 is the most recent version that has been tested.
INFO:__main__:Test configuration: Namespace(commit_a='cbcf44a4caf3fb504ed29ef78091126134e197a3', commit_b='14b4e51a7c6455a61a74d24da9f47dfeb161023f', fail_for_mismatch_before_n_tokens=800, hub_model_name='mlx-community/Mistral-7B-Instruct-v0.2-4-bit', hub_url='github.com', max_context_length=800, measure_every_n_tokens=100, output_dir='external', repo_a='ml-explore/mlx', repo_b='ml-explore/mlx')
INFO:__main__:Cloning repo A: ml-explore/mlx@cbcf44a4caf3fb504ed29ef78091126134e197a3
Cloning into 'mlx'...
remote: Enumerating objects: 11239, done.
remote: Counting objects: 100% (1202/1202), done.
remote: Compressing objects: 100% (339/339), done.
remote: Total 11239 (delta 933), reused 1078 (delta 855), pack-reused 10037
Receiving objects: 100% (11239/11239), 11.98 MiB | 19.86 MiB/s, done.
Resolving deltas: 100% (8794/8794), done.
INFO:argmaxtools.utils:Successfuly cloned mlx repo
HEAD is now at cbcf44a Some fixes in cache / thread safety (#777)
INFO:argmaxtools.utils:Successfuly checked out cbcf44a4caf3fb504ed29ef78091126134e197a3 in mlx
INFO:__main__:Cloning repo B: ml-explore/mlx@14b4e51a7c6455a61a74d24da9f47dfeb161023f
Cloning into 'mlx'...
remote: Enumerating objects: 11239, done.
remote: Counting objects: 100% (1202/1202), done.
remote: Compressing objects: 100% (338/338), done.
remote: Total 11239 (delta 933), reused 1079 (delta 856), pack-reused 10037
Receiving objects: 100% (11239/11239), 11.98 MiB | 28.60 MiB/s, done.
Resolving deltas: 100% (8794/8794), done.
INFO:argmaxtools.utils:Successfuly cloned mlx repo
HEAD is now at 14b4e51 Improved quantized matrix vector product (#786)
INFO:argmaxtools.utils:Successfuly checked out 14b4e51a7c6455a61a74d24da9f47dfeb161023f in mlx
README.md: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 4.48k/4.48k [00:00<00:00, 1.20MB/s]
config.json: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 304/304 [00:00<00:00, 1.33MB/s]
.gitattributes: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.52k/1.52k [00:00<00:00, 3.59MB/s]
tokenizer.model: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 493k/493k [00:00<00:00, 6.81MB/s]
weights.npz: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 4.26G/4.26G [01:43<00:00, 41.2MB/s]
Fetching 5 files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5/5 [01:44<00:00, 20.82s/it]
E
======================================================================
ERROR: setUpClass (__main__.MLXMistral7bRegressionTest)
----------------------------------------------------------------------
Traceback (most recent call last):
File "bench_mistral.py", line 40, in setUpClass
cls.inference_ctx = BenchContext().spec_dict()
File "/Users/somename/miniconda3/envs/mlx-src/lib/python3.8/site-packages/argmaxtools/test_utils.py", line 311, in spec_dict
"os_spec": self.os_spec(),
File "/Users/somename/miniconda3/envs/mlx-src/lib/python3.8/site-packages/argmaxtools/test_utils.py", line 370, in os_spec
os_type, os_version, os_build_number = [
ValueError: too many values to unpack (expected 3)
----------------------------------------------------------------------
Ran 0 tests in 108.467s
Perhaps, consider updating argmaxtools/test_utils.py:370
to be
def os_spec(self):
sw_vers = _term_exec("sw_vers")
"""
% sw_vers
ProductName: xOS
ProductVersion: d.d
BuildVersion: dXd
- d. -> digit(s)
- x,X -> letter(s)
"""
os_type, os_version, os_build_number, *spurious = [
line.rsplit("\t\t")[1]
for line in sw_vers.rsplit("\n")
]
return {
"os_version": os_version,
"os_type": os_type,
"os_build_number": os_build_number,
}
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.