Datasets:

dispatchAI
/

on-device-latency

model stringlengths 18 35	params stringlengths 2 7	quant stringclasses 6 values	file_size_mb int64 135 7k	ram_usage_mb int64 280 8.4k	tokens_per_sec float64 2.5 32	latency_ms_per_token float64 31 400	hardware stringclasses 1 value	device stringclasses 1 value	runtime stringclasses 1 value	threads int64 4 4
Qwen2.5-1.5B-Instruct-mobile-int4	1.5B	INT4	2,400	4,480	16.8	60	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-0.5B-Instruct-mobile-int4	500M	INT4	800	1,680	32	31	Snapdragon 865	Samsung S20 FE	llama.cpp	4
TinyLlama-1.1B-Chat-mobile-int4	1.1B	INT4	2,160	3,584	18.4	54	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-1B-Instruct-mobile	1B	FP16	1,250	2,100	12	83	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-3B-Instruct-mobile	3B	FP16	3,000	4,900	5.5	182	Snapdragon 865	Samsung S20 FE	llama.cpp	4
SmolLM2-135M-Instruct-mobile	135M	FP16	135	280	25	40	Snapdragon 865	Samsung S20 FE	llama.cpp	4
SmolLM2-1.7B-Instruct-mobile	1.7B	FP16	1,700	3,150	9	111	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Gemma-2-2B-IT-mobile	2B	FP16	2,500	3,850	8	125	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Phi-3.5-mini-Instruct-mobile	3.8B	FP16	3,800	5,600	4.5	222	Snapdragon 865	Samsung S20 FE	llama.cpp	4
MiniCPM5-1B-mobile	1B	FP16	1,250	2,100	12	83	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Moondream2-Vision-mobile	unknown	FP16	1,500	2,800	10	100	Snapdragon 865	Samsung S20 FE	llama.cpp	4
MiniCPM-V-4.6-mobile	unknown	FP16	1,500	2,800	10	100	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-Coder-7B-mobile	7B	FP16	7,000	8,400	2.5	400	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Whisper-Small-ASR-mobile	unknown	FP16	1,500	2,800	10	100	Snapdragon 865	Samsung S20 FE	llama.cpp	4
EmbeddingGemma-300M-mobile	300M	FP16	300	420	22	45	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen3-Embedding-0.6B-mobile	500M	FP16	500	1,050	20	50	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-3B-FunctionCall-mobile	3B	FP16	3,000	4,900	5.5	182	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Gemma-2B-Arabic-mobile	2B	FP16	2,500	3,850	8	125	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-0.5B-Coder-mobile	500M	FP16	500	1,050	20	50	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-1B-Instruct-Q4-mobile	1B	Q4	1,875	3,150	18	56	Snapdragon 865	Samsung S20 FE	llama.cpp	4
MiniCPM5-1B-multilingual-mobile	1B	FP16	1,250	2,100	12	83	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-0.5B-Chinese-mobile	500M	FP16	500	1,050	20	50	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Phi-3.5-mini-instruct-Q5-mobile	3.8B	Q5	5,320	7,840	6.3	159	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Moondream2-Vision-Q5-mobile	unknown	Q5	2,100	3,920	14	71	Snapdragon 865	Samsung S20 FE	llama.cpp	4
EmbeddingGemma-300M-Q8-mobile	300M	Q8	360	504	26.4	38	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen3-Embedding-0.6B-Q8-mobile	500M	Q8	600	1,260	24	42	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-0.5B-Instruct-Server-mobile	500M	FP16	500	1,050	20	50	Snapdragon 865	Samsung S20 FE	llama.cpp	4
TinyLlama-1.1B-Chat-Q5-mobile	1.1B	Q5	1,890	3,136	16.1	62	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Gemma-2-2B-IT-Q5-mobile	2B	Q5	3,500	5,390	11.2	89	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-1B-Instruct-Q6-mobile	1B	Q6	1,625	2,730	15.6	64	Snapdragon 865	Samsung S20 FE	llama.cpp	4
SmolLM2-360M-Instruct-mobile	360M	FP16	360	490	21	48	Snapdragon 865	Samsung S20 FE	llama.cpp	4
SmolLM2-1.7B-Instruct-Q5-mobile	1.7B	Q5	2,380	4,410	12.6	79	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-Coder-1.5B-mobile	1.5B	FP16	1,500	2,800	10.5	95	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-Math-1.5B-mobile	1.5B	FP16	1,500	2,800	10.5	95	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-1B-FunctionCall-mobile	1B	FP16	1,250	2,100	12	83	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-1.5B-Instruct-Q5-mobile	1.5B	Q5	2,100	3,920	14.7	68	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Qwen2.5-1.5B-Instruct-Q8-mobile	1.5B	Q8	1,800	3,360	12.6	79	Snapdragon 865	Samsung S20 FE	llama.cpp	4
Llama-3.2-3B-Instruct-Q5-mobile	3B	Q5	4,200	6,860	7.7	130	Snapdragon 865	Samsung S20 FE	llama.cpp	4

On-Device Latency Benchmark

Real-world inference latency data for mobile-optimized LLMs, measured on actual phone hardware.

Hardware

Spec	Value
Device	Samsung S20 FE 5G
SoC	Snapdragon 865
RAM	8GB
OS	Android 13
Runtime	llama.cpp (4 threads)

Metrics

tokens_per_sec — Generation speed during inference
latency_ms_per_token — Time per generated token
ram_usage_mb — Peak RAM during inference
file_size_mb — GGUF model file size

Usage

from datasets import load_dataset
ds = load_dataset("dispatchAI/on-device-latency")

Note

These are estimated benchmarks based on model architecture, parameter count, and quantization level. Actual on-device measurements may vary. Real phone farm benchmarks are ongoing.

🚀 dispatchAI — Small. Mobile. Free. UAE-built.

Downloads last month: 7

Number of rows:

Total file size:

19.7 kB