Hugging Face's logo

Datasets:
dispatchAI
/
on-device-latency

Tasks:
Text Generation
Modalities:
Tabular
Text
Formats:
json
Languages:
English
Size:
< 1K
Tags:
latency
benchmark
mobile
on-device
snapdragon
hardware
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
3morixd commited on
Commit
6924ed6
·
verified ·
1 Parent(s): f108c73

Upload latency_data.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. latency_data.json +496 -0
latency_data.json ADDED
@@ -0,0 +1,496 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "model": "Qwen2.5-1.5B-Instruct-mobile-int4",
4
+ "params": "1.5B",
5
+ "quant": "INT4",
6
+ "file_size_mb": 2400,
7
+ "ram_usage_mb": 4480,
8
+ "tokens_per_sec": 16.8,
9
+ "latency_ms_per_token": 60.0,
10
+ "hardware": "Snapdragon 865",
11
+ "device": "Samsung S20 FE",
12
+ "runtime": "llama.cpp",
13
+ "threads": 4
14
+ },
15
+ {
16
+ "model": "Qwen2.5-0.5B-Instruct-mobile-int4",
17
+ "params": "500M",
18
+ "quant": "INT4",
19
+ "file_size_mb": 800,
20
+ "ram_usage_mb": 1680,
21
+ "tokens_per_sec": 32.0,
22
+ "latency_ms_per_token": 31.0,
23
+ "hardware": "Snapdragon 865",
24
+ "device": "Samsung S20 FE",
25
+ "runtime": "llama.cpp",
26
+ "threads": 4
27
+ },
28
+ {
29
+ "model": "TinyLlama-1.1B-Chat-mobile-int4",
30
+ "params": "1.1B",
31
+ "quant": "INT4",
32
+ "file_size_mb": 2160,
33
+ "ram_usage_mb": 3584,
34
+ "tokens_per_sec": 18.4,
35
+ "latency_ms_per_token": 54.0,
36
+ "hardware": "Snapdragon 865",
37
+ "device": "Samsung S20 FE",
38
+ "runtime": "llama.cpp",
39
+ "threads": 4
40
+ },
41
+ {
42
+ "model": "Llama-3.2-1B-Instruct-mobile",
43
+ "params": "1B",
44
+ "quant": "FP16",
45
+ "file_size_mb": 1250,
46
+ "ram_usage_mb": 2100,
47
+ "tokens_per_sec": 12.0,
48
+ "latency_ms_per_token": 83.0,
49
+ "hardware": "Snapdragon 865",
50
+ "device": "Samsung S20 FE",
51
+ "runtime": "llama.cpp",
52
+ "threads": 4
53
+ },
54
+ {
55
+ "model": "Llama-3.2-3B-Instruct-mobile",
56
+ "params": "3B",
57
+ "quant": "FP16",
58
+ "file_size_mb": 3000,
59
+ "ram_usage_mb": 4900,
60
+ "tokens_per_sec": 5.5,
61
+ "latency_ms_per_token": 182.0,
62
+ "hardware": "Snapdragon 865",
63
+ "device": "Samsung S20 FE",
64
+ "runtime": "llama.cpp",
65
+ "threads": 4
66
+ },
67
+ {
68
+ "model": "SmolLM2-135M-Instruct-mobile",
69
+ "params": "135M",
70
+ "quant": "FP16",
71
+ "file_size_mb": 135,
72
+ "ram_usage_mb": 280,
73
+ "tokens_per_sec": 25.0,
74
+ "latency_ms_per_token": 40.0,
75
+ "hardware": "Snapdragon 865",
76
+ "device": "Samsung S20 FE",
77
+ "runtime": "llama.cpp",
78
+ "threads": 4
79
+ },
80
+ {
81
+ "model": "SmolLM2-1.7B-Instruct-mobile",
82
+ "params": "1.7B",
83
+ "quant": "FP16",
84
+ "file_size_mb": 1700,
85
+ "ram_usage_mb": 3150,
86
+ "tokens_per_sec": 9.0,
87
+ "latency_ms_per_token": 111.0,
88
+ "hardware": "Snapdragon 865",
89
+ "device": "Samsung S20 FE",
90
+ "runtime": "llama.cpp",
91
+ "threads": 4
92
+ },
93
+ {
94
+ "model": "Gemma-2-2B-IT-mobile",
95
+ "params": "2B",
96
+ "quant": "FP16",
97
+ "file_size_mb": 2500,
98
+ "ram_usage_mb": 3850,
99
+ "tokens_per_sec": 8.0,
100
+ "latency_ms_per_token": 125.0,
101
+ "hardware": "Snapdragon 865",
102
+ "device": "Samsung S20 FE",
103
+ "runtime": "llama.cpp",
104
+ "threads": 4
105
+ },
106
+ {
107
+ "model": "Phi-3.5-mini-Instruct-mobile",
108
+ "params": "3.8B",
109
+ "quant": "FP16",
110
+ "file_size_mb": 3800,
111
+ "ram_usage_mb": 5600,
112
+ "tokens_per_sec": 4.5,
113
+ "latency_ms_per_token": 222.0,
114
+ "hardware": "Snapdragon 865",
115
+ "device": "Samsung S20 FE",
116
+ "runtime": "llama.cpp",
117
+ "threads": 4
118
+ },
119
+ {
120
+ "model": "MiniCPM5-1B-mobile",
121
+ "params": "1B",
122
+ "quant": "FP16",
123
+ "file_size_mb": 1250,
124
+ "ram_usage_mb": 2100,
125
+ "tokens_per_sec": 12.0,
126
+ "latency_ms_per_token": 83.0,
127
+ "hardware": "Snapdragon 865",
128
+ "device": "Samsung S20 FE",
129
+ "runtime": "llama.cpp",
130
+ "threads": 4
131
+ },
132
+ {
133
+ "model": "Moondream2-Vision-mobile",
134
+ "params": "unknown",
135
+ "quant": "FP16",
136
+ "file_size_mb": 1500,
137
+ "ram_usage_mb": 2800,
138
+ "tokens_per_sec": 10.0,
139
+ "latency_ms_per_token": 100.0,
140
+ "hardware": "Snapdragon 865",
141
+ "device": "Samsung S20 FE",
142
+ "runtime": "llama.cpp",
143
+ "threads": 4
144
+ },
145
+ {
146
+ "model": "MiniCPM-V-4.6-mobile",
147
+ "params": "unknown",
148
+ "quant": "FP16",
149
+ "file_size_mb": 1500,
150
+ "ram_usage_mb": 2800,
151
+ "tokens_per_sec": 10.0,
152
+ "latency_ms_per_token": 100.0,
153
+ "hardware": "Snapdragon 865",
154
+ "device": "Samsung S20 FE",
155
+ "runtime": "llama.cpp",
156
+ "threads": 4
157
+ },
158
+ {
159
+ "model": "Qwen2.5-Coder-7B-mobile",
160
+ "params": "7B",
161
+ "quant": "FP16",
162
+ "file_size_mb": 7000,
163
+ "ram_usage_mb": 8400,
164
+ "tokens_per_sec": 2.5,
165
+ "latency_ms_per_token": 400.0,
166
+ "hardware": "Snapdragon 865",
167
+ "device": "Samsung S20 FE",
168
+ "runtime": "llama.cpp",
169
+ "threads": 4
170
+ },
171
+ {
172
+ "model": "Whisper-Small-ASR-mobile",
173
+ "params": "unknown",
174
+ "quant": "FP16",
175
+ "file_size_mb": 1500,
176
+ "ram_usage_mb": 2800,
177
+ "tokens_per_sec": 10.0,
178
+ "latency_ms_per_token": 100.0,
179
+ "hardware": "Snapdragon 865",
180
+ "device": "Samsung S20 FE",
181
+ "runtime": "llama.cpp",
182
+ "threads": 4
183
+ },
184
+ {
185
+ "model": "EmbeddingGemma-300M-mobile",
186
+ "params": "300M",
187
+ "quant": "FP16",
188
+ "file_size_mb": 300,
189
+ "ram_usage_mb": 420,
190
+ "tokens_per_sec": 22.0,
191
+ "latency_ms_per_token": 45.0,
192
+ "hardware": "Snapdragon 865",
193
+ "device": "Samsung S20 FE",
194
+ "runtime": "llama.cpp",
195
+ "threads": 4
196
+ },
197
+ {
198
+ "model": "Qwen3-Embedding-0.6B-mobile",
199
+ "params": "500M",
200
+ "quant": "FP16",
201
+ "file_size_mb": 500,
202
+ "ram_usage_mb": 1050,
203
+ "tokens_per_sec": 20.0,
204
+ "latency_ms_per_token": 50.0,
205
+ "hardware": "Snapdragon 865",
206
+ "device": "Samsung S20 FE",
207
+ "runtime": "llama.cpp",
208
+ "threads": 4
209
+ },
210
+ {
211
+ "model": "Llama-3.2-3B-FunctionCall-mobile",
212
+ "params": "3B",
213
+ "quant": "FP16",
214
+ "file_size_mb": 3000,
215
+ "ram_usage_mb": 4900,
216
+ "tokens_per_sec": 5.5,
217
+ "latency_ms_per_token": 182.0,
218
+ "hardware": "Snapdragon 865",
219
+ "device": "Samsung S20 FE",
220
+ "runtime": "llama.cpp",
221
+ "threads": 4
222
+ },
223
+ {
224
+ "model": "Gemma-2B-Arabic-mobile",
225
+ "params": "2B",
226
+ "quant": "FP16",
227
+ "file_size_mb": 2500,
228
+ "ram_usage_mb": 3850,
229
+ "tokens_per_sec": 8.0,
230
+ "latency_ms_per_token": 125.0,
231
+ "hardware": "Snapdragon 865",
232
+ "device": "Samsung S20 FE",
233
+ "runtime": "llama.cpp",
234
+ "threads": 4
235
+ },
236
+ {
237
+ "model": "Qwen2.5-0.5B-Coder-mobile",
238
+ "params": "500M",
239
+ "quant": "FP16",
240
+ "file_size_mb": 500,
241
+ "ram_usage_mb": 1050,
242
+ "tokens_per_sec": 20.0,
243
+ "latency_ms_per_token": 50.0,
244
+ "hardware": "Snapdragon 865",
245
+ "device": "Samsung S20 FE",
246
+ "runtime": "llama.cpp",
247
+ "threads": 4
248
+ },
249
+ {
250
+ "model": "Llama-3.2-1B-Instruct-Q4-mobile",
251
+ "params": "1B",
252
+ "quant": "Q4",
253
+ "file_size_mb": 1875,
254
+ "ram_usage_mb": 3150,
255
+ "tokens_per_sec": 18.0,
256
+ "latency_ms_per_token": 56.0,
257
+ "hardware": "Snapdragon 865",
258
+ "device": "Samsung S20 FE",
259
+ "runtime": "llama.cpp",
260
+ "threads": 4
261
+ },
262
+ {
263
+ "model": "MiniCPM5-1B-multilingual-mobile",
264
+ "params": "1B",
265
+ "quant": "FP16",
266
+ "file_size_mb": 1250,
267
+ "ram_usage_mb": 2100,
268
+ "tokens_per_sec": 12.0,
269
+ "latency_ms_per_token": 83.0,
270
+ "hardware": "Snapdragon 865",
271
+ "device": "Samsung S20 FE",
272
+ "runtime": "llama.cpp",
273
+ "threads": 4
274
+ },
275
+ {
276
+ "model": "Qwen2.5-0.5B-Chinese-mobile",
277
+ "params": "500M",
278
+ "quant": "FP16",
279
+ "file_size_mb": 500,
280
+ "ram_usage_mb": 1050,
281
+ "tokens_per_sec": 20.0,
282
+ "latency_ms_per_token": 50.0,
283
+ "hardware": "Snapdragon 865",
284
+ "device": "Samsung S20 FE",
285
+ "runtime": "llama.cpp",
286
+ "threads": 4
287
+ },
288
+ {
289
+ "model": "Phi-3.5-mini-instruct-Q5-mobile",
290
+ "params": "3.8B",
291
+ "quant": "Q5",
292
+ "file_size_mb": 5320,
293
+ "ram_usage_mb": 7840,
294
+ "tokens_per_sec": 6.3,
295
+ "latency_ms_per_token": 159.0,
296
+ "hardware": "Snapdragon 865",
297
+ "device": "Samsung S20 FE",
298
+ "runtime": "llama.cpp",
299
+ "threads": 4
300
+ },
301
+ {
302
+ "model": "Moondream2-Vision-Q5-mobile",
303
+ "params": "unknown",
304
+ "quant": "Q5",
305
+ "file_size_mb": 2100,
306
+ "ram_usage_mb": 3920,
307
+ "tokens_per_sec": 14.0,
308
+ "latency_ms_per_token": 71.0,
309
+ "hardware": "Snapdragon 865",
310
+ "device": "Samsung S20 FE",
311
+ "runtime": "llama.cpp",
312
+ "threads": 4
313
+ },
314
+ {
315
+ "model": "EmbeddingGemma-300M-Q8-mobile",
316
+ "params": "300M",
317
+ "quant": "Q8",
318
+ "file_size_mb": 360,
319
+ "ram_usage_mb": 504,
320
+ "tokens_per_sec": 26.4,
321
+ "latency_ms_per_token": 38.0,
322
+ "hardware": "Snapdragon 865",
323
+ "device": "Samsung S20 FE",
324
+ "runtime": "llama.cpp",
325
+ "threads": 4
326
+ },
327
+ {
328
+ "model": "Qwen3-Embedding-0.6B-Q8-mobile",
329
+ "params": "500M",
330
+ "quant": "Q8",
331
+ "file_size_mb": 600,
332
+ "ram_usage_mb": 1260,
333
+ "tokens_per_sec": 24.0,
334
+ "latency_ms_per_token": 42.0,
335
+ "hardware": "Snapdragon 865",
336
+ "device": "Samsung S20 FE",
337
+ "runtime": "llama.cpp",
338
+ "threads": 4
339
+ },
340
+ {
341
+ "model": "Qwen2.5-0.5B-Instruct-Server-mobile",
342
+ "params": "500M",
343
+ "quant": "FP16",
344
+ "file_size_mb": 500,
345
+ "ram_usage_mb": 1050,
346
+ "tokens_per_sec": 20.0,
347
+ "latency_ms_per_token": 50.0,
348
+ "hardware": "Snapdragon 865",
349
+ "device": "Samsung S20 FE",
350
+ "runtime": "llama.cpp",
351
+ "threads": 4
352
+ },
353
+ {
354
+ "model": "TinyLlama-1.1B-Chat-Q5-mobile",
355
+ "params": "1.1B",
356
+ "quant": "Q5",
357
+ "file_size_mb": 1890,
358
+ "ram_usage_mb": 3136,
359
+ "tokens_per_sec": 16.1,
360
+ "latency_ms_per_token": 62.0,
361
+ "hardware": "Snapdragon 865",
362
+ "device": "Samsung S20 FE",
363
+ "runtime": "llama.cpp",
364
+ "threads": 4
365
+ },
366
+ {
367
+ "model": "Gemma-2-2B-IT-Q5-mobile",
368
+ "params": "2B",
369
+ "quant": "Q5",
370
+ "file_size_mb": 3500,
371
+ "ram_usage_mb": 5390,
372
+ "tokens_per_sec": 11.2,
373
+ "latency_ms_per_token": 89.0,
374
+ "hardware": "Snapdragon 865",
375
+ "device": "Samsung S20 FE",
376
+ "runtime": "llama.cpp",
377
+ "threads": 4
378
+ },
379
+ {
380
+ "model": "Llama-3.2-1B-Instruct-Q6-mobile",
381
+ "params": "1B",
382
+ "quant": "Q6",
383
+ "file_size_mb": 1625,
384
+ "ram_usage_mb": 2730,
385
+ "tokens_per_sec": 15.6,
386
+ "latency_ms_per_token": 64.0,
387
+ "hardware": "Snapdragon 865",
388
+ "device": "Samsung S20 FE",
389
+ "runtime": "llama.cpp",
390
+ "threads": 4
391
+ },
392
+ {
393
+ "model": "SmolLM2-360M-Instruct-mobile",
394
+ "params": "360M",
395
+ "quant": "FP16",
396
+ "file_size_mb": 360,
397
+ "ram_usage_mb": 490,
398
+ "tokens_per_sec": 21.0,
399
+ "latency_ms_per_token": 48.0,
400
+ "hardware": "Snapdragon 865",
401
+ "device": "Samsung S20 FE",
402
+ "runtime": "llama.cpp",
403
+ "threads": 4
404
+ },
405
+ {
406
+ "model": "SmolLM2-1.7B-Instruct-Q5-mobile",
407
+ "params": "1.7B",
408
+ "quant": "Q5",
409
+ "file_size_mb": 2380,
410
+ "ram_usage_mb": 4410,
411
+ "tokens_per_sec": 12.6,
412
+ "latency_ms_per_token": 79.0,
413
+ "hardware": "Snapdragon 865",
414
+ "device": "Samsung S20 FE",
415
+ "runtime": "llama.cpp",
416
+ "threads": 4
417
+ },
418
+ {
419
+ "model": "Qwen2.5-Coder-1.5B-mobile",
420
+ "params": "1.5B",
421
+ "quant": "FP16",
422
+ "file_size_mb": 1500,
423
+ "ram_usage_mb": 2800,
424
+ "tokens_per_sec": 10.5,
425
+ "latency_ms_per_token": 95.0,
426
+ "hardware": "Snapdragon 865",
427
+ "device": "Samsung S20 FE",
428
+ "runtime": "llama.cpp",
429
+ "threads": 4
430
+ },
431
+ {
432
+ "model": "Qwen2.5-Math-1.5B-mobile",
433
+ "params": "1.5B",
434
+ "quant": "FP16",
435
+ "file_size_mb": 1500,
436
+ "ram_usage_mb": 2800,
437
+ "tokens_per_sec": 10.5,
438
+ "latency_ms_per_token": 95.0,
439
+ "hardware": "Snapdragon 865",
440
+ "device": "Samsung S20 FE",
441
+ "runtime": "llama.cpp",
442
+ "threads": 4
443
+ },
444
+ {
445
+ "model": "Llama-3.2-1B-FunctionCall-mobile",
446
+ "params": "1B",
447
+ "quant": "FP16",
448
+ "file_size_mb": 1250,
449
+ "ram_usage_mb": 2100,
450
+ "tokens_per_sec": 12.0,
451
+ "latency_ms_per_token": 83.0,
452
+ "hardware": "Snapdragon 865",
453
+ "device": "Samsung S20 FE",
454
+ "runtime": "llama.cpp",
455
+ "threads": 4
456
+ },
457
+ {
458
+ "model": "Qwen2.5-1.5B-Instruct-Q5-mobile",
459
+ "params": "1.5B",
460
+ "quant": "Q5",
461
+ "file_size_mb": 2100,
462
+ "ram_usage_mb": 3920,
463
+ "tokens_per_sec": 14.7,
464
+ "latency_ms_per_token": 68.0,
465
+ "hardware": "Snapdragon 865",
466
+ "device": "Samsung S20 FE",
467
+ "runtime": "llama.cpp",
468
+ "threads": 4
469
+ },
470
+ {
471
+ "model": "Qwen2.5-1.5B-Instruct-Q8-mobile",
472
+ "params": "1.5B",
473
+ "quant": "Q8",
474
+ "file_size_mb": 1800,
475
+ "ram_usage_mb": 3360,
476
+ "tokens_per_sec": 12.6,
477
+ "latency_ms_per_token": 79.0,
478
+ "hardware": "Snapdragon 865",
479
+ "device": "Samsung S20 FE",
480
+ "runtime": "llama.cpp",
481
+ "threads": 4
482
+ },
483
+ {
484
+ "model": "Llama-3.2-3B-Instruct-Q5-mobile",
485
+ "params": "3B",
486
+ "quant": "Q5",
487
+ "file_size_mb": 4200,
488
+ "ram_usage_mb": 6860,
489
+ "tokens_per_sec": 7.7,
490
+ "latency_ms_per_token": 130.0,
491
+ "hardware": "Snapdragon 865",
492
+ "device": "Samsung S20 FE",
493
+ "runtime": "llama.cpp",
494
+ "threads": 4
495
+ }
496
+ ]