LLM-Pilot: Characterize and Optimize Performance of your LLM Inference Services
Paper • 2410.02425 • Published
Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
source: string
This is similar to the Apple Silicon benchmark thread, but for Vulkan! We'll be testing the Llama 2 (... 1555 chars omitted): string
pp512 t/s: string
tg128 t/s: string
Commit: string
Comments: string
Summary
LLaMA 7B: string
BW [GB/s]: string
GPU Cores: string
F16 PP [t/s]: string
F16 TG [t/s]: string
Q8_0 PP [t/s]: string
Q8_0 TG [t/s]: string
Q4_0 PP [t/s]: string
Q4_0 TG [t/s]: string
This is similar to the Performance of llama.cpp on Apple Silicon M-series, Performance of llama.cpp (... 1360 chars omitted): string
Memory: string
Thanks to: string
sbc: struct<Raspberry Pi 5 (8 GB): struct<ram: double, bw: double>, Raspberry Pi 5 (4 GB): struct<ram: do (... 130 chars omitted)
child 0, Raspberry Pi 5 (8 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
child 1, Raspberry Pi 5 (4 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
child 2, Raspberry Pi 4 (8 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
child 3, Raspberry Pi 4 (4 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
apple: struct<m5 max: struct<ram: double, bw: double, cpuCores: double, gpuCores: double>, m5 pro: struct<r (... 1278 chars omitted)
child 0, m5 max: struct<ram: double, bw: double, cpuCores: double, gpuCores: double>
child 0, ram: double
child 1, bw: double
child 2, cpuCores: double
...
cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 203, Arc A770M: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 204, Arc A550M: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 205, Arc A370M: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 206, Iris Xe: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 207, Iris Plus: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 208, UHD 770: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 209, UHD 730: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 210, UHD Graphics 630: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 211, UHD Graphics 620: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
to
{'generated_at': Value('timestamp[s]'), 'source': {'repo': Value('string'), 'license': Value('string'), 'note': Value('string')}, 'gpus': {'RTX 5090': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5080': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5070 Ti': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5070': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5060 Ti 16GB': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5060 Ti': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5060': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5050': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4090': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4080 SUPER': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4080': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070 Ti SUPER': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070 Ti': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070 SUPER': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070': {'vram': Value('float64'), 'bw': Valu
...
': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm3': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2 ultra': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2 max': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2 pro': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1 ultra': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1 max': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1 pro': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}}, 'sbc': {'Raspberry Pi 5 (8 GB)': {'ram': Value('float64'), 'bw': Value('float64')}, 'Raspberry Pi 5 (4 GB)': {'ram': Value('float64'), 'bw': Value('float64')}, 'Raspberry Pi 4 (8 GB)': {'ram': Value('float64'), 'bw': Value('float64')}, 'Raspberry Pi 4 (4 GB)': {'ram': Value('float64'), 'bw': Value('float64')}}}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
return get_rows(
^^^^^^^^^
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/utils.py", line 77, in get_rows
rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2815, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2352, in __iter__
for key, pa_table in self._iter_arrow():
^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2377, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 310, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 130, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2369, in table_cast
return cast_table_to_schema(table, schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
raise CastError(
datasets.table.CastError: Couldn't cast
source: string
This is similar to the Apple Silicon benchmark thread, but for Vulkan! We'll be testing the Llama 2 (... 1555 chars omitted): string
pp512 t/s: string
tg128 t/s: string
Commit: string
Comments: string
Summary
LLaMA 7B: string
BW [GB/s]: string
GPU Cores: string
F16 PP [t/s]: string
F16 TG [t/s]: string
Q8_0 PP [t/s]: string
Q8_0 TG [t/s]: string
Q4_0 PP [t/s]: string
Q4_0 TG [t/s]: string
This is similar to the Performance of llama.cpp on Apple Silicon M-series, Performance of llama.cpp (... 1360 chars omitted): string
Memory: string
Thanks to: string
sbc: struct<Raspberry Pi 5 (8 GB): struct<ram: double, bw: double>, Raspberry Pi 5 (4 GB): struct<ram: do (... 130 chars omitted)
child 0, Raspberry Pi 5 (8 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
child 1, Raspberry Pi 5 (4 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
child 2, Raspberry Pi 4 (8 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
child 3, Raspberry Pi 4 (4 GB): struct<ram: double, bw: double>
child 0, ram: double
child 1, bw: double
apple: struct<m5 max: struct<ram: double, bw: double, cpuCores: double, gpuCores: double>, m5 pro: struct<r (... 1278 chars omitted)
child 0, m5 max: struct<ram: double, bw: double, cpuCores: double, gpuCores: double>
child 0, ram: double
child 1, bw: double
child 2, cpuCores: double
...
cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 203, Arc A770M: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 204, Arc A550M: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 205, Arc A370M: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 206, Iris Xe: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 207, Iris Plus: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 208, UHD 770: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 209, UHD 730: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 210, UHD Graphics 630: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
child 211, UHD Graphics 620: struct<vram: double, bw: double, cores: double>
child 0, vram: double
child 1, bw: double
child 2, cores: double
to
{'generated_at': Value('timestamp[s]'), 'source': {'repo': Value('string'), 'license': Value('string'), 'note': Value('string')}, 'gpus': {'RTX 5090': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5080': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5070 Ti': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5070': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5060 Ti 16GB': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5060 Ti': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5060': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 5050': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4090': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4080 SUPER': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4080': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070 Ti SUPER': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070 Ti': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070 SUPER': {'vram': Value('float64'), 'bw': Value('float64'), 'cores': Value('float64')}, 'RTX 4070': {'vram': Value('float64'), 'bw': Valu
...
': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm3': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2 ultra': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2 max': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2 pro': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm2': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1 ultra': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1 max': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1 pro': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}, 'm1': {'ram': Value('float64'), 'bw': Value('float64'), 'cpuCores': Value('float64'), 'gpuCores': Value('float64')}}, 'sbc': {'Raspberry Pi 5 (8 GB)': {'ram': Value('float64'), 'bw': Value('float64')}, 'Raspberry Pi 5 (4 GB)': {'ram': Value('float64'), 'bw': Value('float64')}, 'Raspberry Pi 4 (8 GB)': {'ram': Value('float64'), 'bw': Value('float64')}, 'Raspberry Pi 4 (4 GB)': {'ram': Value('float64'), 'bw': Value('float64')}}}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Local LLM speed benchmarks (community union)
Real measured local-LLM inference performance on consumer hardware: (accelerator, model, quant, context) -> prompt tok/s, decode tok/s, TTFT. Collected to train FitCheck's honest speed predictor (methodology: LLM-Pilot, leave-one-accelerator-out validation).
Sources & attribution
- LocalScore (https://www.localscore.ai) — Mozilla Builders project by cjpais. The bulk of this dataset. Scraped politely from public pages; the upstream data license is unstated — if you are the maintainer and want this mirror changed or removed, open a discussion and it will be honoured fast.
- llama.cpp community benchmarks — GitHub discussions #4167 (Apple Silicon), #15013 (CUDA), #10879 (Vulkan/AMD/Intel), llama-bench tables.
- Hardware bandwidth specs joined from vendor product pages via the MIT spec tables of https://github.com/midudev/canirun.ai.
Files
localscore_runs.jsonl— run-level: hardware + model + averaged metrics.localscore_tests.jsonl— per-scenario rows (n_prompt 16-4096, n_gen 16-3072) for context-dependence.llamacpp_bench.jsonl— fixed-workload community tables (validation set).gpu_specs.json— device -> memory bandwidth / VRAM (vendor specs).
- Downloads last month
- -