hf-transformers-bot commited on
Commit
3a643ea
·
verified ·
1 Parent(s): b7ec3d2

Upload benchmark results for run 18169101732

Browse files
2025-10-01/18169101732/benchmark_results/Llama-2-7b-hf/Llama-2-7b-hf_benchmark_20251001_164841.json ADDED
@@ -0,0 +1,1121 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_name": "Llama-2-7b-hf",
3
+ "benchmark_scenarios": [
4
+ {
5
+ "scenario_name": "eager_eager_attn",
6
+ "metadata": {
7
+ "timestamp": "2025-10-01T16:45:40.779464",
8
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
9
+ "hardware_info": {
10
+ "gpu_name": "unknown",
11
+ "gpu_memory_total_mb": 0,
12
+ "cpu_count": 160,
13
+ "memory_total_mb": 1285717,
14
+ "python_version": "3.12.10",
15
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
16
+ "cuda_version": null
17
+ },
18
+ "config": {
19
+ "name": "eager",
20
+ "model_id": "meta-llama/Llama-2-7b-hf",
21
+ "variant": "eager",
22
+ "warmup_iterations": 3,
23
+ "measurement_iterations": 5,
24
+ "num_tokens_to_generate": 100,
25
+ "device": "cuda",
26
+ "torch_dtype": "float16",
27
+ "compile_mode": null,
28
+ "compile_options": {},
29
+ "use_cache": true,
30
+ "batch_size": 1,
31
+ "sequence_length": null,
32
+ "attn_implementation": "eager",
33
+ "sdpa_backend": null,
34
+ "custom_params": {}
35
+ }
36
+ },
37
+ "measurements": {
38
+ "latency_seconds": {
39
+ "name": "latency_seconds",
40
+ "measurements": [
41
+ 1.61267529296875,
42
+ 1.6179969482421874,
43
+ 1.6089134521484374,
44
+ 1.62017041015625,
45
+ 1.623338623046875
46
+ ],
47
+ "mean": 1.6166189453125,
48
+ "median": 1.6179969482421874,
49
+ "std": 0.005188311520605786,
50
+ "min": 1.6089134521484374,
51
+ "max": 1.623338623046875,
52
+ "p25": 1.61267529296875,
53
+ "p75": 1.62017041015625,
54
+ "p90": 1.6220713378906249,
55
+ "p95": 1.62270498046875,
56
+ "p99": 1.62321189453125,
57
+ "unit": "seconds"
58
+ },
59
+ "time_to_first_token_seconds": {
60
+ "name": "time_to_first_token_seconds",
61
+ "measurements": [
62
+ 0.016998748779296875,
63
+ 0.01708210754394531,
64
+ 0.017180065155029296,
65
+ 0.0168929500579834,
66
+ 0.017088066101074217
67
+ ],
68
+ "mean": 0.01704838752746582,
69
+ "median": 0.01708210754394531,
70
+ "std": 9.661805064748314e-05,
71
+ "min": 0.0168929500579834,
72
+ "max": 0.017180065155029296,
73
+ "p25": 0.016998748779296875,
74
+ "p75": 0.017088066101074217,
75
+ "p90": 0.017143265533447263,
76
+ "p95": 0.01716166534423828,
77
+ "p99": 0.017176385192871093,
78
+ "unit": "seconds"
79
+ },
80
+ "tokens_per_second": {
81
+ "name": "tokens_per_second",
82
+ "measurements": [
83
+ 62.008762976650736,
84
+ 61.8048137288771,
85
+ 62.153747217705565,
86
+ 61.72190244503722,
87
+ 61.60144197906664
88
+ ],
89
+ "mean": 61.85813366946745,
90
+ "median": 61.8048137288771,
91
+ "std": 0.1986802177266707,
92
+ "min": 61.60144197906664,
93
+ "max": 62.153747217705565,
94
+ "p25": 61.72190244503722,
95
+ "p75": 62.008762976650736,
96
+ "p90": 62.09575352128363,
97
+ "p95": 62.1247503694946,
98
+ "p99": 62.147947848063374,
99
+ "unit": "tokens/sec"
100
+ },
101
+ "time_per_output_token_seconds": {
102
+ "name": "time_per_output_token_seconds",
103
+ "measurements": [
104
+ 0.016126752929687502,
105
+ 0.016179969482421876,
106
+ 0.016089134521484375,
107
+ 0.0162017041015625,
108
+ 0.01623338623046875
109
+ ],
110
+ "mean": 0.016166189453125,
111
+ "median": 0.016179969482421876,
112
+ "std": 5.188311520605776e-05,
113
+ "min": 0.016089134521484375,
114
+ "max": 0.01623338623046875,
115
+ "p25": 0.016126752929687502,
116
+ "p75": 0.0162017041015625,
117
+ "p90": 0.01622071337890625,
118
+ "p95": 0.0162270498046875,
119
+ "p99": 0.016232118945312502,
120
+ "unit": "seconds/token"
121
+ }
122
+ },
123
+ "gpu_metrics": {
124
+ "gpu_monitoring_status": "disabled",
125
+ "gpu_monitoring_reason": "no_gpus_available"
126
+ }
127
+ },
128
+ {
129
+ "scenario_name": "compiled_compile_max-autotune_eager_attn",
130
+ "metadata": {
131
+ "timestamp": "2025-10-01T16:45:58.960892",
132
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
133
+ "hardware_info": {
134
+ "gpu_name": "unknown",
135
+ "gpu_memory_total_mb": 0,
136
+ "cpu_count": 160,
137
+ "memory_total_mb": 1285717,
138
+ "python_version": "3.12.10",
139
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
140
+ "cuda_version": null
141
+ },
142
+ "config": {
143
+ "name": "compiled",
144
+ "model_id": "meta-llama/Llama-2-7b-hf",
145
+ "variant": "compiled",
146
+ "warmup_iterations": 3,
147
+ "measurement_iterations": 5,
148
+ "num_tokens_to_generate": 100,
149
+ "device": "cuda",
150
+ "torch_dtype": "float16",
151
+ "compile_mode": "max-autotune",
152
+ "compile_options": {},
153
+ "use_cache": true,
154
+ "batch_size": 1,
155
+ "sequence_length": null,
156
+ "attn_implementation": "eager",
157
+ "sdpa_backend": null,
158
+ "custom_params": {}
159
+ }
160
+ },
161
+ "measurements": {
162
+ "latency_seconds": {
163
+ "name": "latency_seconds",
164
+ "measurements": [
165
+ 2.947924072265625,
166
+ 2.977679443359375,
167
+ 3.043441162109375,
168
+ 3.088385009765625,
169
+ 3.13058154296875
170
+ ],
171
+ "mean": 3.0376022460937504,
172
+ "median": 3.043441162109375,
173
+ "std": 0.06766246939397137,
174
+ "min": 2.947924072265625,
175
+ "max": 3.13058154296875,
176
+ "p25": 2.977679443359375,
177
+ "p75": 3.088385009765625,
178
+ "p90": 3.1137029296875,
179
+ "p95": 3.122142236328125,
180
+ "p99": 3.128893681640625,
181
+ "unit": "seconds"
182
+ },
183
+ "time_to_first_token_seconds": {
184
+ "name": "time_to_first_token_seconds",
185
+ "measurements": [
186
+ 0.01813968849182129,
187
+ 0.017759176254272462,
188
+ 0.01792597198486328,
189
+ 0.017920732498168945,
190
+ 0.017965572357177734
191
+ ],
192
+ "mean": 0.017942228317260743,
193
+ "median": 0.01792597198486328,
194
+ "std": 0.00012146593027086832,
195
+ "min": 0.017759176254272462,
196
+ "max": 0.01813968849182129,
197
+ "p25": 0.017920732498168945,
198
+ "p75": 0.017965572357177734,
199
+ "p90": 0.018070042037963866,
200
+ "p95": 0.018104865264892578,
201
+ "p99": 0.018132723846435547,
202
+ "unit": "seconds"
203
+ },
204
+ "tokens_per_second": {
205
+ "name": "tokens_per_second",
206
+ "measurements": [
207
+ 33.922176266617704,
208
+ 33.58319856189135,
209
+ 32.85754337721158,
210
+ 32.3793826494414,
211
+ 31.942946902181433
212
+ ],
213
+ "mean": 32.93704955146869,
214
+ "median": 32.85754337721158,
215
+ "std": 0.733966875444952,
216
+ "min": 31.942946902181433,
217
+ "max": 33.922176266617704,
218
+ "p25": 32.3793826494414,
219
+ "p75": 33.58319856189135,
220
+ "p90": 33.786585184727166,
221
+ "p95": 33.854380725672435,
222
+ "p99": 33.90861715842865,
223
+ "unit": "tokens/sec"
224
+ },
225
+ "time_per_output_token_seconds": {
226
+ "name": "time_per_output_token_seconds",
227
+ "measurements": [
228
+ 0.02947924072265625,
229
+ 0.029776794433593753,
230
+ 0.030434411621093747,
231
+ 0.03088385009765625,
232
+ 0.031305815429687495
233
+ ],
234
+ "mean": 0.030376022460937502,
235
+ "median": 0.030434411621093747,
236
+ "std": 0.0006766246939397121,
237
+ "min": 0.02947924072265625,
238
+ "max": 0.031305815429687495,
239
+ "p25": 0.029776794433593753,
240
+ "p75": 0.03088385009765625,
241
+ "p90": 0.031137029296874997,
242
+ "p95": 0.031221422363281244,
243
+ "p99": 0.031288936816406245,
244
+ "unit": "seconds/token"
245
+ }
246
+ },
247
+ "gpu_metrics": {
248
+ "gpu_monitoring_status": "disabled",
249
+ "gpu_monitoring_reason": "no_gpus_available"
250
+ }
251
+ },
252
+ {
253
+ "scenario_name": "eager_sdpa_default",
254
+ "metadata": {
255
+ "timestamp": "2025-10-01T16:46:38.995037",
256
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
257
+ "hardware_info": {
258
+ "gpu_name": "unknown",
259
+ "gpu_memory_total_mb": 0,
260
+ "cpu_count": 160,
261
+ "memory_total_mb": 1285717,
262
+ "python_version": "3.12.10",
263
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
264
+ "cuda_version": null
265
+ },
266
+ "config": {
267
+ "name": "eager",
268
+ "model_id": "meta-llama/Llama-2-7b-hf",
269
+ "variant": "eager",
270
+ "warmup_iterations": 3,
271
+ "measurement_iterations": 5,
272
+ "num_tokens_to_generate": 100,
273
+ "device": "cuda",
274
+ "torch_dtype": "float16",
275
+ "compile_mode": null,
276
+ "compile_options": {},
277
+ "use_cache": true,
278
+ "batch_size": 1,
279
+ "sequence_length": null,
280
+ "attn_implementation": "sdpa",
281
+ "sdpa_backend": null,
282
+ "custom_params": {}
283
+ }
284
+ },
285
+ "measurements": {
286
+ "latency_seconds": {
287
+ "name": "latency_seconds",
288
+ "measurements": [
289
+ 1.37264013671875,
290
+ 1.378490966796875,
291
+ 1.3829903564453125,
292
+ 1.384458740234375,
293
+ 1.38162060546875
294
+ ],
295
+ "mean": 1.3800401611328126,
296
+ "median": 1.38162060546875,
297
+ "std": 0.004192509090673094,
298
+ "min": 1.37264013671875,
299
+ "max": 1.384458740234375,
300
+ "p25": 1.378490966796875,
301
+ "p75": 1.3829903564453125,
302
+ "p90": 1.38387138671875,
303
+ "p95": 1.3841650634765625,
304
+ "p99": 1.3844000048828127,
305
+ "unit": "seconds"
306
+ },
307
+ "time_to_first_token_seconds": {
308
+ "name": "time_to_first_token_seconds",
309
+ "measurements": [
310
+ 0.014690963745117188,
311
+ 0.014972758293151856,
312
+ 0.014907959938049317,
313
+ 0.014942638397216797,
314
+ 0.014998477935791016
315
+ ],
316
+ "mean": 0.014902559661865234,
317
+ "median": 0.014942638397216797,
318
+ "std": 0.0001100331432815653,
319
+ "min": 0.014690963745117188,
320
+ "max": 0.014998477935791016,
321
+ "p25": 0.014907959938049317,
322
+ "p75": 0.014972758293151856,
323
+ "p90": 0.014988190078735352,
324
+ "p95": 0.014993334007263184,
325
+ "p99": 0.014997449150085449,
326
+ "unit": "seconds"
327
+ },
328
+ "tokens_per_second": {
329
+ "name": "tokens_per_second",
330
+ "measurements": [
331
+ 72.85230653319422,
332
+ 72.54309415778371,
333
+ 72.3070840906144,
334
+ 72.23039379495773,
335
+ 72.37876997793649
336
+ ],
337
+ "mean": 72.46232971089731,
338
+ "median": 72.37876997793649,
339
+ "std": 0.22067284314201688,
340
+ "min": 72.23039379495773,
341
+ "max": 72.85230653319422,
342
+ "p25": 72.3070840906144,
343
+ "p75": 72.54309415778371,
344
+ "p90": 72.72862158303002,
345
+ "p95": 72.79046405811212,
346
+ "p99": 72.8399380381778,
347
+ "unit": "tokens/sec"
348
+ },
349
+ "time_per_output_token_seconds": {
350
+ "name": "time_per_output_token_seconds",
351
+ "measurements": [
352
+ 0.0137264013671875,
353
+ 0.01378490966796875,
354
+ 0.013829903564453126,
355
+ 0.01384458740234375,
356
+ 0.0138162060546875
357
+ ],
358
+ "mean": 0.013800401611328126,
359
+ "median": 0.0138162060546875,
360
+ "std": 4.1925090906730734e-05,
361
+ "min": 0.0137264013671875,
362
+ "max": 0.01384458740234375,
363
+ "p25": 0.01378490966796875,
364
+ "p75": 0.013829903564453126,
365
+ "p90": 0.013838713867187501,
366
+ "p95": 0.013841650634765626,
367
+ "p99": 0.013844000048828126,
368
+ "unit": "seconds/token"
369
+ }
370
+ },
371
+ "gpu_metrics": {
372
+ "gpu_monitoring_status": "disabled",
373
+ "gpu_monitoring_reason": "no_gpus_available"
374
+ }
375
+ },
376
+ {
377
+ "scenario_name": "eager_sdpa_math",
378
+ "metadata": {
379
+ "timestamp": "2025-10-01T16:46:55.410116",
380
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
381
+ "hardware_info": {
382
+ "gpu_name": "unknown",
383
+ "gpu_memory_total_mb": 0,
384
+ "cpu_count": 160,
385
+ "memory_total_mb": 1285717,
386
+ "python_version": "3.12.10",
387
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
388
+ "cuda_version": null
389
+ },
390
+ "config": {
391
+ "name": "eager",
392
+ "model_id": "meta-llama/Llama-2-7b-hf",
393
+ "variant": "eager",
394
+ "warmup_iterations": 3,
395
+ "measurement_iterations": 5,
396
+ "num_tokens_to_generate": 100,
397
+ "device": "cuda",
398
+ "torch_dtype": "float16",
399
+ "compile_mode": null,
400
+ "compile_options": {},
401
+ "use_cache": true,
402
+ "batch_size": 1,
403
+ "sequence_length": null,
404
+ "attn_implementation": "sdpa",
405
+ "sdpa_backend": "math",
406
+ "custom_params": {}
407
+ }
408
+ },
409
+ "measurements": {
410
+ "latency_seconds": {
411
+ "name": "latency_seconds",
412
+ "measurements": [
413
+ 1.7312880859375,
414
+ 1.7339990234375,
415
+ 1.703171875,
416
+ 1.7000650634765626,
417
+ 1.700549560546875
418
+ ],
419
+ "mean": 1.7138147216796873,
420
+ "median": 1.703171875,
421
+ "std": 0.015433808875104062,
422
+ "min": 1.7000650634765626,
423
+ "max": 1.7339990234375,
424
+ "p25": 1.700549560546875,
425
+ "p75": 1.7312880859375,
426
+ "p90": 1.7329146484375,
427
+ "p95": 1.7334568359375,
428
+ "p99": 1.7338905859375,
429
+ "unit": "seconds"
430
+ },
431
+ "time_to_first_token_seconds": {
432
+ "name": "time_to_first_token_seconds",
433
+ "measurements": [
434
+ 0.01870307922363281,
435
+ 0.019614580154418946,
436
+ 0.018965032577514647,
437
+ 0.019222827911376954,
438
+ 0.018900074005126954
439
+ ],
440
+ "mean": 0.01908111877441406,
441
+ "median": 0.018965032577514647,
442
+ "std": 0.00031427046918447964,
443
+ "min": 0.01870307922363281,
444
+ "max": 0.019614580154418946,
445
+ "p25": 0.018900074005126954,
446
+ "p75": 0.019222827911376954,
447
+ "p90": 0.01945787925720215,
448
+ "p95": 0.019536229705810548,
449
+ "p99": 0.019598910064697266,
450
+ "unit": "seconds"
451
+ },
452
+ "tokens_per_second": {
453
+ "name": "tokens_per_second",
454
+ "measurements": [
455
+ 57.76046217394811,
456
+ 57.670159353238176,
457
+ 58.7139803491647,
458
+ 58.82127816655684,
459
+ 58.80451962119897
460
+ ],
461
+ "mean": 58.35407993282136,
462
+ "median": 58.7139803491647,
463
+ "std": 0.523608215172847,
464
+ "min": 57.670159353238176,
465
+ "max": 58.82127816655684,
466
+ "p25": 57.76046217394811,
467
+ "p75": 58.80451962119897,
468
+ "p90": 58.81457474841369,
469
+ "p95": 58.817926457485264,
470
+ "p99": 58.820607824742524,
471
+ "unit": "tokens/sec"
472
+ },
473
+ "time_per_output_token_seconds": {
474
+ "name": "time_per_output_token_seconds",
475
+ "measurements": [
476
+ 0.017312880859375,
477
+ 0.017339990234375,
478
+ 0.01703171875,
479
+ 0.017000650634765625,
480
+ 0.01700549560546875
481
+ ],
482
+ "mean": 0.017138147216796875,
483
+ "median": 0.01703171875,
484
+ "std": 0.00015433808875104144,
485
+ "min": 0.017000650634765625,
486
+ "max": 0.017339990234375,
487
+ "p25": 0.01700549560546875,
488
+ "p75": 0.017312880859375,
489
+ "p90": 0.017329146484375002,
490
+ "p95": 0.017334568359375,
491
+ "p99": 0.017338905859375,
492
+ "unit": "seconds/token"
493
+ }
494
+ },
495
+ "gpu_metrics": {
496
+ "gpu_monitoring_status": "disabled",
497
+ "gpu_monitoring_reason": "no_gpus_available"
498
+ }
499
+ },
500
+ {
501
+ "scenario_name": "eager_sdpa_flash_attention",
502
+ "metadata": {
503
+ "timestamp": "2025-10-01T16:47:14.637601",
504
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
505
+ "hardware_info": {
506
+ "gpu_name": "unknown",
507
+ "gpu_memory_total_mb": 0,
508
+ "cpu_count": 160,
509
+ "memory_total_mb": 1285717,
510
+ "python_version": "3.12.10",
511
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
512
+ "cuda_version": null
513
+ },
514
+ "config": {
515
+ "name": "eager",
516
+ "model_id": "meta-llama/Llama-2-7b-hf",
517
+ "variant": "eager",
518
+ "warmup_iterations": 3,
519
+ "measurement_iterations": 5,
520
+ "num_tokens_to_generate": 100,
521
+ "device": "cuda",
522
+ "torch_dtype": "float16",
523
+ "compile_mode": null,
524
+ "compile_options": {},
525
+ "use_cache": true,
526
+ "batch_size": 1,
527
+ "sequence_length": null,
528
+ "attn_implementation": "sdpa",
529
+ "sdpa_backend": "flash_attention",
530
+ "custom_params": {}
531
+ }
532
+ },
533
+ "measurements": {
534
+ "latency_seconds": {
535
+ "name": "latency_seconds",
536
+ "measurements": [
537
+ 1.37681396484375,
538
+ 1.3858720703125,
539
+ 1.3967607421875,
540
+ 1.382825927734375,
541
+ 1.38182861328125
542
+ ],
543
+ "mean": 1.384820263671875,
544
+ "median": 1.382825927734375,
545
+ "std": 0.006643936983565205,
546
+ "min": 1.37681396484375,
547
+ "max": 1.3967607421875,
548
+ "p25": 1.38182861328125,
549
+ "p75": 1.3858720703125,
550
+ "p90": 1.3924052734374999,
551
+ "p95": 1.3945830078124999,
552
+ "p99": 1.3963251953125,
553
+ "unit": "seconds"
554
+ },
555
+ "time_to_first_token_seconds": {
556
+ "name": "time_to_first_token_seconds",
557
+ "measurements": [
558
+ 0.015094555854797364,
559
+ 0.01502303695678711,
560
+ 0.014739044189453125,
561
+ 0.01515163516998291,
562
+ 0.01506391716003418
563
+ ],
564
+ "mean": 0.015014437866210937,
565
+ "median": 0.01506391716003418,
566
+ "std": 0.00014394847295584915,
567
+ "min": 0.014739044189453125,
568
+ "max": 0.01515163516998291,
569
+ "p25": 0.01502303695678711,
570
+ "p75": 0.015094555854797364,
571
+ "p90": 0.01512880344390869,
572
+ "p95": 0.0151402193069458,
573
+ "p99": 0.015149351997375488,
574
+ "unit": "seconds"
575
+ },
576
+ "tokens_per_second": {
577
+ "name": "tokens_per_second",
578
+ "measurements": [
579
+ 72.63145388807024,
580
+ 72.15673231473019,
581
+ 71.59422296146987,
582
+ 72.31568196283405,
583
+ 72.36787474138555
584
+ ],
585
+ "mean": 72.21319317369799,
586
+ "median": 72.31568196283405,
587
+ "std": 0.3451620827058354,
588
+ "min": 71.59422296146987,
589
+ "max": 72.63145388807024,
590
+ "p25": 72.15673231473019,
591
+ "p75": 72.36787474138555,
592
+ "p90": 72.52602222939636,
593
+ "p95": 72.57873805873331,
594
+ "p99": 72.62091072220285,
595
+ "unit": "tokens/sec"
596
+ },
597
+ "time_per_output_token_seconds": {
598
+ "name": "time_per_output_token_seconds",
599
+ "measurements": [
600
+ 0.0137681396484375,
601
+ 0.013858720703125,
602
+ 0.013967607421874998,
603
+ 0.013828259277343751,
604
+ 0.013818286132812501
605
+ ],
606
+ "mean": 0.013848202636718748,
607
+ "median": 0.013828259277343751,
608
+ "std": 6.643936983565183e-05,
609
+ "min": 0.0137681396484375,
610
+ "max": 0.013967607421874998,
611
+ "p25": 0.013818286132812501,
612
+ "p75": 0.013858720703125,
613
+ "p90": 0.013924052734375,
614
+ "p95": 0.013945830078124999,
615
+ "p99": 0.013963251953124999,
616
+ "unit": "seconds/token"
617
+ }
618
+ },
619
+ "gpu_metrics": {
620
+ "gpu_monitoring_status": "disabled",
621
+ "gpu_monitoring_reason": "no_gpus_available"
622
+ }
623
+ },
624
+ {
625
+ "scenario_name": "eager_sdpa_efficient_attention",
626
+ "metadata": {
627
+ "timestamp": "2025-10-01T16:47:31.214692",
628
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
629
+ "hardware_info": {
630
+ "gpu_name": "unknown",
631
+ "gpu_memory_total_mb": 0,
632
+ "cpu_count": 160,
633
+ "memory_total_mb": 1285717,
634
+ "python_version": "3.12.10",
635
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
636
+ "cuda_version": null
637
+ },
638
+ "config": {
639
+ "name": "eager",
640
+ "model_id": "meta-llama/Llama-2-7b-hf",
641
+ "variant": "eager",
642
+ "warmup_iterations": 3,
643
+ "measurement_iterations": 5,
644
+ "num_tokens_to_generate": 100,
645
+ "device": "cuda",
646
+ "torch_dtype": "float16",
647
+ "compile_mode": null,
648
+ "compile_options": {},
649
+ "use_cache": true,
650
+ "batch_size": 1,
651
+ "sequence_length": null,
652
+ "attn_implementation": "sdpa",
653
+ "sdpa_backend": "efficient_attention",
654
+ "custom_params": {}
655
+ }
656
+ },
657
+ "measurements": {
658
+ "latency_seconds": {
659
+ "name": "latency_seconds",
660
+ "measurements": [
661
+ 1.3455224609375,
662
+ 1.2517255859375,
663
+ 1.247614990234375,
664
+ 1.25335205078125,
665
+ 1.251693115234375
666
+ ],
667
+ "mean": 1.2699816406250002,
668
+ "median": 1.2517255859375,
669
+ "std": 0.03781793102555915,
670
+ "min": 1.247614990234375,
671
+ "max": 1.3455224609375,
672
+ "p25": 1.251693115234375,
673
+ "p75": 1.25335205078125,
674
+ "p90": 1.308654296875,
675
+ "p95": 1.32708837890625,
676
+ "p99": 1.34183564453125,
677
+ "unit": "seconds"
678
+ },
679
+ "time_to_first_token_seconds": {
680
+ "name": "time_to_first_token_seconds",
681
+ "measurements": [
682
+ 0.01485392189025879,
683
+ 0.013823463439941406,
684
+ 0.013686627388000487,
685
+ 0.013939021110534668,
686
+ 0.01376742458343506
687
+ ],
688
+ "mean": 0.014014091682434083,
689
+ "median": 0.013823463439941406,
690
+ "std": 0.0004278707789586778,
691
+ "min": 0.013686627388000487,
692
+ "max": 0.01485392189025879,
693
+ "p25": 0.01376742458343506,
694
+ "p75": 0.013939021110534668,
695
+ "p90": 0.014487961578369141,
696
+ "p95": 0.014670941734313965,
697
+ "p99": 0.014817325859069825,
698
+ "unit": "seconds"
699
+ },
700
+ "tokens_per_second": {
701
+ "name": "tokens_per_second",
702
+ "measurements": [
703
+ 74.32057279097701,
704
+ 79.88971474534763,
705
+ 80.15293242125455,
706
+ 79.78604250710498,
707
+ 79.89178719839435
708
+ ],
709
+ "mean": 78.80820993261571,
710
+ "median": 79.88971474534763,
711
+ "std": 2.2470921022990193,
712
+ "min": 74.32057279097701,
713
+ "max": 80.15293242125455,
714
+ "p25": 79.78604250710498,
715
+ "p75": 79.89178719839435,
716
+ "p90": 80.04847433211047,
717
+ "p95": 80.1007033766825,
718
+ "p99": 80.14248661234014,
719
+ "unit": "tokens/sec"
720
+ },
721
+ "time_per_output_token_seconds": {
722
+ "name": "time_per_output_token_seconds",
723
+ "measurements": [
724
+ 0.013455224609375,
725
+ 0.012517255859375,
726
+ 0.012476149902343749,
727
+ 0.0125335205078125,
728
+ 0.01251693115234375
729
+ ],
730
+ "mean": 0.01269981640625,
731
+ "median": 0.012517255859375,
732
+ "std": 0.0003781793102555914,
733
+ "min": 0.012476149902343749,
734
+ "max": 0.013455224609375,
735
+ "p25": 0.01251693115234375,
736
+ "p75": 0.0125335205078125,
737
+ "p90": 0.01308654296875,
738
+ "p95": 0.0132708837890625,
739
+ "p99": 0.0134183564453125,
740
+ "unit": "seconds/token"
741
+ }
742
+ },
743
+ "gpu_metrics": {
744
+ "gpu_monitoring_status": "disabled",
745
+ "gpu_monitoring_reason": "no_gpus_available"
746
+ }
747
+ },
748
+ {
749
+ "scenario_name": "compiled_compile_max-autotune_sdpa_default",
750
+ "metadata": {
751
+ "timestamp": "2025-10-01T16:47:47.235435",
752
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
753
+ "hardware_info": {
754
+ "gpu_name": "unknown",
755
+ "gpu_memory_total_mb": 0,
756
+ "cpu_count": 160,
757
+ "memory_total_mb": 1285717,
758
+ "python_version": "3.12.10",
759
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
760
+ "cuda_version": null
761
+ },
762
+ "config": {
763
+ "name": "compiled",
764
+ "model_id": "meta-llama/Llama-2-7b-hf",
765
+ "variant": "compiled",
766
+ "warmup_iterations": 3,
767
+ "measurement_iterations": 5,
768
+ "num_tokens_to_generate": 100,
769
+ "device": "cuda",
770
+ "torch_dtype": "float16",
771
+ "compile_mode": "max-autotune",
772
+ "compile_options": {},
773
+ "use_cache": true,
774
+ "batch_size": 1,
775
+ "sequence_length": null,
776
+ "attn_implementation": "sdpa",
777
+ "sdpa_backend": null,
778
+ "custom_params": {}
779
+ }
780
+ },
781
+ "measurements": {
782
+ "latency_seconds": {
783
+ "name": "latency_seconds",
784
+ "measurements": [
785
+ 1.4993692626953126,
786
+ 1.5045772705078124,
787
+ 1.499740234375,
788
+ 1.49411669921875,
789
+ 1.4866534423828126
790
+ ],
791
+ "mean": 1.4968913818359377,
792
+ "median": 1.4993692626953126,
793
+ "std": 0.00609660572599227,
794
+ "min": 1.4866534423828126,
795
+ "max": 1.5045772705078124,
796
+ "p25": 1.49411669921875,
797
+ "p75": 1.499740234375,
798
+ "p90": 1.5026424560546874,
799
+ "p95": 1.50360986328125,
800
+ "p99": 1.5043837890625,
801
+ "unit": "seconds"
802
+ },
803
+ "time_to_first_token_seconds": {
804
+ "name": "time_to_first_token_seconds",
805
+ "measurements": [
806
+ 0.015594386100769042,
807
+ 0.01569214344024658,
808
+ 0.015617666244506835,
809
+ 0.01567450428009033,
810
+ 0.015572305679321288
811
+ ],
812
+ "mean": 0.015630201148986812,
813
+ "median": 0.015617666244506835,
814
+ "std": 4.602464945882928e-05,
815
+ "min": 0.015572305679321288,
816
+ "max": 0.01569214344024658,
817
+ "p25": 0.015594386100769042,
818
+ "p75": 0.01567450428009033,
819
+ "p90": 0.015685087776184083,
820
+ "p95": 0.015688615608215332,
821
+ "p99": 0.015691437873840332,
822
+ "unit": "seconds"
823
+ },
824
+ "tokens_per_second": {
825
+ "name": "tokens_per_second",
826
+ "measurements": [
827
+ 66.69471122826468,
828
+ 66.46385131569137,
829
+ 66.67821380525534,
830
+ 66.92917631687567,
831
+ 67.265172332107
832
+ ],
833
+ "mean": 66.80622499963881,
834
+ "median": 66.69471122826468,
835
+ "std": 0.27268722263547673,
836
+ "min": 66.46385131569137,
837
+ "max": 67.265172332107,
838
+ "p25": 66.67821380525534,
839
+ "p75": 66.92917631687567,
840
+ "p90": 67.13077392601447,
841
+ "p95": 67.19797312906073,
842
+ "p99": 67.25173249149775,
843
+ "unit": "tokens/sec"
844
+ },
845
+ "time_per_output_token_seconds": {
846
+ "name": "time_per_output_token_seconds",
847
+ "measurements": [
848
+ 0.014993692626953126,
849
+ 0.015045772705078125,
850
+ 0.014997402343749999,
851
+ 0.0149411669921875,
852
+ 0.014866534423828126
853
+ ],
854
+ "mean": 0.014968913818359375,
855
+ "median": 0.014993692626953126,
856
+ "std": 6.096605725992296e-05,
857
+ "min": 0.014866534423828126,
858
+ "max": 0.015045772705078125,
859
+ "p25": 0.0149411669921875,
860
+ "p75": 0.014997402343749999,
861
+ "p90": 0.015026424560546875,
862
+ "p95": 0.0150360986328125,
863
+ "p99": 0.015043837890625,
864
+ "unit": "seconds/token"
865
+ }
866
+ },
867
+ "gpu_metrics": {
868
+ "gpu_monitoring_status": "disabled",
869
+ "gpu_monitoring_reason": "no_gpus_available"
870
+ }
871
+ },
872
+ {
873
+ "scenario_name": "compiled_compile_max-autotune_sdpa_math",
874
+ "metadata": {
875
+ "timestamp": "2025-10-01T16:48:04.754905",
876
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
877
+ "hardware_info": {
878
+ "gpu_name": "unknown",
879
+ "gpu_memory_total_mb": 0,
880
+ "cpu_count": 160,
881
+ "memory_total_mb": 1285717,
882
+ "python_version": "3.12.10",
883
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
884
+ "cuda_version": null
885
+ },
886
+ "config": {
887
+ "name": "compiled",
888
+ "model_id": "meta-llama/Llama-2-7b-hf",
889
+ "variant": "compiled",
890
+ "warmup_iterations": 3,
891
+ "measurement_iterations": 5,
892
+ "num_tokens_to_generate": 100,
893
+ "device": "cuda",
894
+ "torch_dtype": "float16",
895
+ "compile_mode": "max-autotune",
896
+ "compile_options": {},
897
+ "use_cache": true,
898
+ "batch_size": 1,
899
+ "sequence_length": null,
900
+ "attn_implementation": "sdpa",
901
+ "sdpa_backend": "math",
902
+ "custom_params": {}
903
+ }
904
+ },
905
+ "measurements": {
906
+ "latency_seconds": {
907
+ "name": "latency_seconds",
908
+ "measurements": [
909
+ 1.812379150390625,
910
+ 1.784843017578125,
911
+ 1.772970458984375,
912
+ 1.777419189453125,
913
+ 1.77984521484375
914
+ ],
915
+ "mean": 1.78549140625,
916
+ "median": 1.77984521484375,
917
+ "std": 0.013979881792646545,
918
+ "min": 1.772970458984375,
919
+ "max": 1.812379150390625,
920
+ "p25": 1.777419189453125,
921
+ "p75": 1.784843017578125,
922
+ "p90": 1.801364697265625,
923
+ "p95": 1.806871923828125,
924
+ "p99": 1.811277705078125,
925
+ "unit": "seconds"
926
+ },
927
+ "time_to_first_token_seconds": {
928
+ "name": "time_to_first_token_seconds",
929
+ "measurements": [
930
+ 0.019036151885986327,
931
+ 0.018907756805419923,
932
+ 0.018819875717163086,
933
+ 0.01859356117248535,
934
+ 0.018770078659057617
935
+ ],
936
+ "mean": 0.01882548484802246,
937
+ "median": 0.018819875717163086,
938
+ "std": 0.00014699603753688561,
939
+ "min": 0.01859356117248535,
940
+ "max": 0.019036151885986327,
941
+ "p25": 0.018770078659057617,
942
+ "p75": 0.018907756805419923,
943
+ "p90": 0.018984793853759764,
944
+ "p95": 0.019010472869873046,
945
+ "p99": 0.019031016082763672,
946
+ "unit": "seconds"
947
+ },
948
+ "tokens_per_second": {
949
+ "name": "tokens_per_second",
950
+ "measurements": [
951
+ 55.17609269475807,
952
+ 56.027336306411534,
953
+ 56.40251900039203,
954
+ 56.26134824771861,
955
+ 56.184660984005205
956
+ ],
957
+ "mean": 56.010391446657096,
958
+ "median": 56.184660984005205,
959
+ "std": 0.4343857137929538,
960
+ "min": 55.17609269475807,
961
+ "max": 56.40251900039203,
962
+ "p25": 56.027336306411534,
963
+ "p75": 56.26134824771861,
964
+ "p90": 56.34605069932266,
965
+ "p95": 56.37428484985735,
966
+ "p99": 56.39687217028509,
967
+ "unit": "tokens/sec"
968
+ },
969
+ "time_per_output_token_seconds": {
970
+ "name": "time_per_output_token_seconds",
971
+ "measurements": [
972
+ 0.01812379150390625,
973
+ 0.01784843017578125,
974
+ 0.017729704589843753,
975
+ 0.01777419189453125,
976
+ 0.0177984521484375
977
+ ],
978
+ "mean": 0.0178549140625,
979
+ "median": 0.0177984521484375,
980
+ "std": 0.00013979881792646522,
981
+ "min": 0.017729704589843753,
982
+ "max": 0.01812379150390625,
983
+ "p25": 0.01777419189453125,
984
+ "p75": 0.01784843017578125,
985
+ "p90": 0.01801364697265625,
986
+ "p95": 0.01806871923828125,
987
+ "p99": 0.01811277705078125,
988
+ "unit": "seconds/token"
989
+ }
990
+ },
991
+ "gpu_metrics": {
992
+ "gpu_monitoring_status": "disabled",
993
+ "gpu_monitoring_reason": "no_gpus_available"
994
+ }
995
+ },
996
+ {
997
+ "scenario_name": "compiled_compile_max-autotune_sdpa_efficient_attention",
998
+ "metadata": {
999
+ "timestamp": "2025-10-01T16:48:29.071170",
1000
+ "commit_id": "94bbf8e19911a3aed089d47094c836a18717c9c0",
1001
+ "hardware_info": {
1002
+ "gpu_name": "unknown",
1003
+ "gpu_memory_total_mb": 0,
1004
+ "cpu_count": 160,
1005
+ "memory_total_mb": 1285717,
1006
+ "python_version": "3.12.10",
1007
+ "torch_version": "2.7.1+rocm6.4.1.git2a215e4a",
1008
+ "cuda_version": null
1009
+ },
1010
+ "config": {
1011
+ "name": "compiled",
1012
+ "model_id": "meta-llama/Llama-2-7b-hf",
1013
+ "variant": "compiled",
1014
+ "warmup_iterations": 3,
1015
+ "measurement_iterations": 5,
1016
+ "num_tokens_to_generate": 100,
1017
+ "device": "cuda",
1018
+ "torch_dtype": "float16",
1019
+ "compile_mode": "max-autotune",
1020
+ "compile_options": {},
1021
+ "use_cache": true,
1022
+ "batch_size": 1,
1023
+ "sequence_length": null,
1024
+ "attn_implementation": "sdpa",
1025
+ "sdpa_backend": "efficient_attention",
1026
+ "custom_params": {}
1027
+ }
1028
+ },
1029
+ "measurements": {
1030
+ "latency_seconds": {
1031
+ "name": "latency_seconds",
1032
+ "measurements": [
1033
+ 1.310001953125,
1034
+ 1.427109375,
1035
+ 1.46913525390625,
1036
+ 1.4190069580078124,
1037
+ 1.3986259765625
1038
+ ],
1039
+ "mean": 1.4047759033203124,
1040
+ "median": 1.4190069580078124,
1041
+ "std": 0.05265616035637173,
1042
+ "min": 1.310001953125,
1043
+ "max": 1.46913525390625,
1044
+ "p25": 1.3986259765625,
1045
+ "p75": 1.427109375,
1046
+ "p90": 1.4523249023437501,
1047
+ "p95": 1.460730078125,
1048
+ "p99": 1.4674542187500001,
1049
+ "unit": "seconds"
1050
+ },
1051
+ "time_to_first_token_seconds": {
1052
+ "name": "time_to_first_token_seconds",
1053
+ "measurements": [
1054
+ 0.013458991050720214,
1055
+ 0.015611665725708008,
1056
+ 0.015620426177978515,
1057
+ 0.015665143966674806,
1058
+ 0.013521150588989258
1059
+ ],
1060
+ "mean": 0.014775475502014159,
1061
+ "median": 0.015611665725708008,
1062
+ "std": 0.0010498693261505518,
1063
+ "min": 0.013458991050720214,
1064
+ "max": 0.015665143966674806,
1065
+ "p25": 0.013521150588989258,
1066
+ "p75": 0.015620426177978515,
1067
+ "p90": 0.01564725685119629,
1068
+ "p95": 0.015656200408935547,
1069
+ "p99": 0.015663355255126955,
1070
+ "unit": "seconds"
1071
+ },
1072
+ "tokens_per_second": {
1073
+ "name": "tokens_per_second",
1074
+ "measurements": [
1075
+ 76.3357640509243,
1076
+ 70.07171401981716,
1077
+ 68.06725230649273,
1078
+ 70.47181793977464,
1079
+ 71.49874353526376
1080
+ ],
1081
+ "mean": 71.28905837045451,
1082
+ "median": 70.47181793977464,
1083
+ "std": 2.758370881407156,
1084
+ "min": 68.06725230649273,
1085
+ "max": 76.3357640509243,
1086
+ "p25": 70.07171401981716,
1087
+ "p75": 71.49874353526376,
1088
+ "p90": 74.40095584466007,
1089
+ "p95": 75.36835994779219,
1090
+ "p99": 76.14228323029788,
1091
+ "unit": "tokens/sec"
1092
+ },
1093
+ "time_per_output_token_seconds": {
1094
+ "name": "time_per_output_token_seconds",
1095
+ "measurements": [
1096
+ 0.01310001953125,
1097
+ 0.014271093749999998,
1098
+ 0.0146913525390625,
1099
+ 0.014190069580078124,
1100
+ 0.013986259765625
1101
+ ],
1102
+ "mean": 0.014047759033203124,
1103
+ "median": 0.014190069580078124,
1104
+ "std": 0.0005265616035637173,
1105
+ "min": 0.01310001953125,
1106
+ "max": 0.0146913525390625,
1107
+ "p25": 0.013986259765625,
1108
+ "p75": 0.014271093749999998,
1109
+ "p90": 0.0145232490234375,
1110
+ "p95": 0.01460730078125,
1111
+ "p99": 0.0146745421875,
1112
+ "unit": "seconds/token"
1113
+ }
1114
+ },
1115
+ "gpu_metrics": {
1116
+ "gpu_monitoring_status": "disabled",
1117
+ "gpu_monitoring_reason": "no_gpus_available"
1118
+ }
1119
+ }
1120
+ ]
1121
+ }