|
25 | 25 | submission_checker_dir = os.path.dirname(os.path.realpath(__file__)) |
26 | 26 |
|
27 | 27 | MODEL_CONFIG = { |
28 | | - "v5.0": { |
| 28 | + "v6.0": { |
29 | 29 | "models": [ |
30 | 30 | "resnet", |
31 | 31 | "retinanet", |
32 | 32 | "bert-99", |
33 | 33 | "bert-99.9", |
34 | | - "dlrm-v2-99", |
35 | | - "dlrm-v2-99.9", |
36 | 34 | "3d-unet-99", |
37 | 35 | "3d-unet-99.9", |
38 | | - "gptj-99", |
39 | | - "gptj-99.9", |
| 36 | + "llama3.1-8b", |
| 37 | + "llama3.1-8b-edge", |
40 | 38 | "llama2-70b-99", |
41 | 39 | "llama2-70b-99.9", |
42 | | - "llama2-70b-interactive-99", |
43 | | - "llama2-70b-interactive-99.9", |
44 | 40 | "stable-diffusion-xl", |
45 | 41 | "mixtral-8x7b", |
46 | 42 | "llama3.1-405b", |
47 | 43 | "rgat", |
48 | 44 | "pointpainting", |
| 45 | + "deepseek-r1", |
| 46 | + "whisper", |
| 47 | + "gpt-oss-120b", |
| 48 | + "wan-2.2-t2v-a14b", |
| 49 | + "qwen3-vl-235b-a22b", |
| 50 | + "dlrm-v3", |
49 | 51 | ], |
50 | 52 | "required-scenarios-datacenter": { |
51 | | - "resnet": ["Server", "Offline"], |
52 | 53 | "retinanet": ["Server", "Offline"], |
53 | | - "dlrm-v2-99": ["Server", "Offline"], |
54 | | - "dlrm-v2-99.9": ["Server", "Offline"], |
| 54 | + "dlrm-v3": ["Server", "Offline"], |
55 | 55 | "3d-unet-99": ["Offline"], |
56 | 56 | "3d-unet-99.9": ["Offline"], |
57 | | - "gptj-99": ["Server", "Offline"], |
58 | | - "gptj-99.9": ["Server", "Offline"], |
59 | | - "llama2-70b-99": ["Server", "Offline"], |
60 | | - "llama2-70b-99.9": ["Server", "Offline"], |
61 | | - "llama2-70b-interactive-99": ["Server", "Offline"], |
62 | | - "llama2-70b-interactive-99.9": ["Server", "Offline"], |
| 57 | + "llama3.1-8b": ["Offline"], |
| 58 | + "llama2-70b-99": ["Offline"], |
| 59 | + "llama2-70b-99.9": ["Offline"], |
63 | 60 | "stable-diffusion-xl": ["Server", "Offline"], |
64 | 61 | "mixtral-8x7b": ["Server", "Offline"], |
65 | | - "llama3.1-405b": ["Server", "Offline"], |
| 62 | + "llama3.1-405b": ["Offline"], |
66 | 63 | "rgat": ["Offline"], |
| 64 | + "whisper": ["Offline"], |
| 65 | + "deepseek-r1": ["Offline"], |
| 66 | + "gpt-oss-120b": ["Offline"], |
| 67 | + "qwen3-vl-235b-a22b": ["Server", "Offline"], |
| 68 | + "dlrm-v3": ["Server", "Offline"], |
| 69 | + }, |
| 70 | + "optional-scenarios-datacenter": { |
| 71 | + "llama2-70b-99": ["Interactive", "Server"], |
| 72 | + "llama2-70b-99.9": ["Interactive", "Server"], |
| 73 | + "llama3.1-405b": ["Interactive", "Server"], |
| 74 | + "llama3.1-8b": ["Interactive", "Server"], |
| 75 | + "deepseek-r1": ["Interactive", "Server"], |
| 76 | + "gpt-oss-120b": ["Interactive", "Server"], |
67 | 77 | }, |
68 | | - "optional-scenarios-datacenter": {}, |
69 | 78 | "required-scenarios-edge": { |
70 | 79 | "resnet": ["SingleStream", "MultiStream", "Offline"], |
71 | 80 | "retinanet": ["SingleStream", "MultiStream", "Offline"], |
72 | 81 | "bert-99": ["SingleStream", "Offline"], |
73 | 82 | "bert-99.9": ["SingleStream", "Offline"], |
74 | 83 | "3d-unet-99": ["SingleStream", "Offline"], |
75 | 84 | "3d-unet-99.9": ["SingleStream", "Offline"], |
76 | | - "gptj-99": ["SingleStream", "Offline"], |
77 | | - "gptj-99.9": ["SingleStream", "Offline"], |
| 85 | + "llama3.1-8b-edge": ["SingleStream", "Offline"], |
78 | 86 | "stable-diffusion-xl": ["SingleStream", "Offline"], |
79 | 87 | "pointpainting": ["SingleStream"], |
| 88 | + "whisper": ["Offline"], |
80 | 89 | }, |
81 | 90 | "optional-scenarios-edge": {}, |
82 | 91 | "required-scenarios-datacenter-edge": { |
83 | | - "resnet": ["SingleStream", "Offline", "MultiStream", "Server"], |
| 92 | + "resnet": ["SingleStream", "MultiStream", "Offline", "Server"], |
84 | 93 | "retinanet": ["SingleStream", "Offline", "MultiStream", "Server"], |
85 | 94 | "bert-99": ["SingleStream", "Offline"], |
86 | 95 | "bert-99.9": ["SingleStream", "Offline"], |
87 | | - "dlrm-v2-99": ["Offline", "Server"], |
88 | | - "dlrm-v2-99.9": ["Offline", "Server"], |
89 | 96 | "3d-unet-99": ["SingleStream", "Offline"], |
90 | 97 | "3d-unet-99.9": ["SingleStream", "Offline"], |
91 | | - "gptj-99": ["SingleStream", "Offline", "Server"], |
92 | | - "gptj-99.9": ["SingleStream", "Offline", "Server"], |
93 | | - "llama2-70b-99": ["Server", "Offline"], |
94 | | - "llama2-70b-99.9": ["Server", "Offline"], |
95 | | - "llama2-70b-interactive-99": ["Server", "Offline"], |
96 | | - "llama2-70b-interactive-99.9": ["Server", "Offline"], |
| 98 | + "llama3.1-8b": ["Offline"], |
| 99 | + "llama3.1-8b-edge": ["SingleStream", "Offline"], |
| 100 | + "llama2-70b-99": ["Offline"], |
| 101 | + "llama2-70b-99.9": ["Offline"], |
97 | 102 | "stable-diffusion-xl": ["SingleStream", "Offline", "Server"], |
98 | 103 | "mixtral-8x7b": ["Server", "Offline"], |
99 | | - "llama3.1-405b": ["Server", "Offline"], |
| 104 | + "llama3.1-405b": ["Offline"], |
100 | 105 | "rgat": ["Offline"], |
101 | 106 | "pointpainting": ["SingleStream"], |
| 107 | + "deepseek-r1": ["Offline"], |
| 108 | + "whisper": ["Offline"], |
| 109 | + "gpt-oss-120b": ["Offline"], |
| 110 | + "qwen3-vl-235b-a22b": ["Offline"], |
| 111 | + "dlrm-v3": ["Offline", "Server"], |
| 112 | + }, |
| 113 | + "optional-scenarios-datacenter-edge": { |
| 114 | + "llama2-70b-99": ["Interactive", "Server"], |
| 115 | + "llama2-70b-99.9": ["Interactive", "Server"], |
| 116 | + "llama3.1-405b": ["Interactive", "Server"], |
| 117 | + "llama3.1-8b": ["Interactive", "Server"], |
| 118 | + "deepseek-r1": ["Interactive", "Server"], |
| 119 | + "gpt-oss-120b": ["Interactive", "Server"], |
| 120 | + "qwen3-vl-235b-a22b": ["Interactive", "Server"], |
102 | 121 | }, |
103 | | - "optional-scenarios-datacenter-edge": {}, |
104 | 122 | "accuracy-target": { |
105 | 123 | "resnet": ("acc", 76.46 * 0.99), |
106 | 124 | "retinanet": ("mAP", 37.55 * 0.99), |
|
111 | 129 | "3d-unet-99": ("DICE", 0.86170 * 0.99), |
112 | 130 | "3d-unet-99.9": ("DICE", 0.86170 * 0.999), |
113 | 131 |
|
114 | | - "gptj-99": ( |
| 132 | + "llama3.1-8b": ( |
115 | 133 | "ROUGE1", |
116 | | - 42.9865 * 0.99, |
| 134 | + 38.7792 * 0.99, |
117 | 135 | "ROUGE2", |
118 | | - 20.1235 * 0.99, |
| 136 | + 15.9075 * 0.99, |
119 | 137 | "ROUGEL", |
120 | | - 29.9881 * 0.99, |
| 138 | + 24.4957 * 0.99, |
| 139 | + "ROUGELSUM", |
| 140 | + 35.793 * 0.99, |
121 | 141 | "GEN_LEN", |
122 | | - 4016878 * 0.9, |
| 142 | + 8167644 * 0.9, |
123 | 143 | ), |
124 | | - "gptj-99.9": ( |
| 144 | + "llama3.1-8b-edge": ( |
125 | 145 | "ROUGE1", |
126 | | - 42.9865 * 0.999, |
| 146 | + 39.06 * 0.99, |
127 | 147 | "ROUGE2", |
128 | | - 20.1235 * 0.999, |
| 148 | + 16.1147 * 0.99, |
129 | 149 | "ROUGEL", |
130 | | - 29.9881 * 0.999, |
| 150 | + 24.6375 * 0.99, |
| 151 | + "ROUGELSUM", |
| 152 | + 36.124 * 0.99, |
131 | 153 | "GEN_LEN", |
132 | | - 4016878 * 0.9, |
| 154 | + 3051113 * 0.9, |
133 | 155 | ), |
134 | 156 | "llama2-70b-99": ( |
135 | 157 | "ROUGE1", |
|
151 | 173 | "TOKENS_PER_SAMPLE", |
152 | 174 | 294.45 * 0.9, |
153 | 175 | ), |
154 | | - "llama2-70b-interactive-99": ( |
155 | | - "ROUGE1", |
156 | | - 44.4312 * 0.99, |
157 | | - "ROUGE2", |
158 | | - 22.0352 * 0.99, |
159 | | - "ROUGEL", |
160 | | - 28.6162 * 0.99, |
161 | | - "TOKENS_PER_SAMPLE", |
162 | | - 294.45 * 0.9, |
163 | | - ), |
164 | | - "llama2-70b-interactive-99.9": ( |
165 | | - "ROUGE1", |
166 | | - 44.4312 * 0.999, |
167 | | - "ROUGE2", |
168 | | - 22.0352 * 0.999, |
169 | | - "ROUGEL", |
170 | | - 28.6162 * 0.999, |
171 | | - "TOKENS_PER_SAMPLE", |
172 | | - 294.45 * 0.9, |
173 | | - ), |
174 | 176 | "stable-diffusion-xl": ( |
175 | 177 | "CLIP_SCORE", |
176 | 178 | 31.68631873, |
|
201 | 203 | ), |
202 | 204 | "rgat": ("acc", 0.7286 * 0.99), |
203 | 205 | "pointpainting": ("mAP", 0.5425 * 0.999), |
| 206 | + "deepseek-r1": ("exact_match", 0.99 * 81.3582, "TOKENS_PER_SAMPLE", 0.9 * 3886.2274), |
| 207 | + "whisper": ("ACCURACY", (100.0 - 2.0671) * 0.99), |
| 208 | + # TODO: Placeholder for now |
| 209 | + "gpt-oss-120b": ("exact_match", 83.13 * 0.99), |
| 210 | + # TODO: Placeholder for now |
| 211 | + "qwen3-vl-235b-a22b": ("F1", 0.7903 * 0.99), |
| 212 | + "dlrm-v3": ("AUC", 78.663 * 0.99), # TODO: Placeholder for now |
204 | 213 | }, |
205 | 214 | "accuracy-upper-limit": { |
206 | 215 | "stable-diffusion-xl": ( |
|
211 | 220 | ), |
212 | 221 | "llama2-70b-99": ("TOKENS_PER_SAMPLE", 294.45 * 1.1), |
213 | 222 | "llama2-70b-99.9": ("TOKENS_PER_SAMPLE", 294.45 * 1.1), |
214 | | - "llama2-70b-interactive-99": ("TOKENS_PER_SAMPLE", 294.45 * 1.1), |
215 | | - "llama2-70b-interactive-99.9": ("TOKENS_PER_SAMPLE", 294.45 * 1.1), |
216 | 223 | "mixtral-8x7b": ("TOKENS_PER_SAMPLE", 145.9 * 1.1), |
217 | 224 | "llama3.1-405b": ("TOKENS_PER_SAMPLE", 684.68 * 1.1), |
| 225 | + "llama3.1-8b": ("GEN_LEN", 8167644 * 1.1), |
| 226 | + "llama3.1-8b-edge": ("GEN_LEN", 3051113 * 1.1), |
| 227 | + "deepseek-r1": ("TOKENS_PER_SAMPLE", 1.1 * 3886.2274), |
| 228 | + # TODO: Placeholder for now |
| 229 | + "gpt-oss-120b": ("TOKENS_PER_SAMPLE", 1.1 * 9999), |
218 | 230 | }, |
219 | 231 | "accuracy-delta-perc": { |
220 | 232 | "stable-diffusion-xl": {"CLIP_SCORE": 1, "FID_SCORE": 2} |
|
228 | 240 | "dlrm-v2-99.9": 204800, |
229 | 241 | "3d-unet-99": 43, |
230 | 242 | "3d-unet-99.9": 43, |
231 | | - "gptj-99": 13368, |
232 | | - "gptj-99.9": 13368, |
| 243 | + "llama3.1-8b": 13368, |
| 244 | + "llama3.1-8b-edge": 5000, |
233 | 245 | "llama2-70b-99": 24576, |
234 | 246 | "llama2-70b-99.9": 24576, |
235 | | - "llama2-70b-interactive-99": 24576, |
236 | | - "llama2-70b-interactive-99.9": 24576, |
237 | 247 | "stable-diffusion-xl": 5000, |
238 | 248 | "mixtral-8x7b": 15000, |
239 | 249 | "llama3.1-405b": 8313, |
240 | 250 | "rgat": 788379, |
241 | 251 | "pointpainting": 1024, |
| 252 | + "deepseek-r1": 4388, |
| 253 | + "whisper": 1633, |
| 254 | + # TODO: Need to add accuracy sample count checkers as well (4395) |
| 255 | + "gpt-oss-120b": 6396, |
| 256 | + "qwen3-vl-235b-a22b": 48289, |
| 257 | + "dlrm-v3": 34996, |
242 | 258 | }, |
243 | 259 | "dataset-size": { |
244 | 260 | "resnet": 50000, |
|
249 | 265 | "dlrm-v2-99.9": 330067, |
250 | 266 | "3d-unet-99": 43, |
251 | 267 | "3d-unet-99.9": 43, |
252 | | - "gptj-99": 13368, |
253 | | - "gptj-99.9": 13368, |
| 268 | + "llama3.1-8b": 13368, |
| 269 | + "llama3.1-8b-edge": 5000, |
254 | 270 | "llama2-70b-99": 24576, |
255 | 271 | "llama2-70b-99.9": 24576, |
256 | | - "llama2-70b-interactive-99": 24576, |
257 | | - "llama2-70b-interactive-99.9": 24576, |
258 | 272 | "stable-diffusion-xl": 5000, |
259 | 273 | "mixtral-8x7b": 15000, |
260 | 274 | "llama3.1-405b": 8313, |
261 | 275 | "rgat": 788379, |
262 | 276 | "pointpainting": 39987, |
| 277 | + "deepseek-r1": 4388, |
| 278 | + "whisper": 1633, |
| 279 | + # TODO: Need to add accuracy sample count checkers as well (4395) |
| 280 | + "gpt-oss-120b": 6396, |
| 281 | + "qwen3-vl-235b-a22b": 48289, |
| 282 | + "dlrm-v3": 34996, |
263 | 283 | }, |
264 | 284 | # model_mapping.json is expected in the root directory of the |
265 | 285 | # submission folder for open submissions and so the below dictionary is |
|
270 | 290 | "mobilenet": "resnet", |
271 | 291 | "resnet50": "resnet", |
272 | 292 | "llama3_1-405b": "llama3.1-405b", |
| 293 | + "llama3_1-8b": "llama3.1-8b", |
| 294 | + "llama3_1-8b-edge": "llama3.1-8b-edge", |
273 | 295 | }, |
274 | 296 | "seeds": { |
275 | 297 | # TODO: Update random seeds |
276 | | - "qsl_rng_seed": 6023615788873153749, |
277 | | - "sample_index_rng_seed": 15036839855038426416, |
278 | | - "schedule_rng_seed": 9933818062894767841, |
| 298 | + "qsl_rng_seed": 1780908523862526354, |
| 299 | + "sample_index_rng_seed": 14771362308971278857, |
| 300 | + "schedule_rng_seed": 18209322760996052031, |
279 | 301 | }, |
280 | 302 | "ignore_errors": [], |
281 | 303 | "latency-constraint": { |
282 | 304 | "resnet": {"Server": 15000000}, |
283 | 305 | "retinanet": {"Server": 100000000}, |
284 | 306 | "dlrm-v2-99": {"Server": 60000000}, |
285 | 307 | "dlrm-v2-99.9": {"Server": 60000000}, |
286 | | - "gptj-99": {"Server": 20000000000}, |
287 | | - "gptj-99.9": {"Server": 20000000000}, |
| 308 | + "llama3.1-8b": {"Server": 20000000000}, |
288 | 309 | "stable-diffusion-xl": {"Server": 20000000000}, |
289 | 310 | "llama2-70b-99": {"Server": 20000000000}, |
290 | 311 | "llama2-70b-99.9": {"Server": 20000000000}, |
291 | | - "llama2-70b-interactive-99": {"Server": 20000000000}, |
292 | | - "llama2-70b-interactive-99.9": {"Server": 20000000000}, |
293 | 312 | "mixtral-8x7b": {"Server": 20000000000}, |
294 | | - "llama3.1-405b": {"Server": 60000000000} |
| 313 | + "llama3.1-405b": {"Server": 60000000000}, |
| 314 | + "deepseek-r1": {"Server": 60000000000}, |
| 315 | + "gpt-oss-120b": {"Server": 60000000000}, |
| 316 | + "qwen3-vl-235b-a22b": {"Server": 60000000000}, |
| 317 | + "dlrm-v3": {"Server": 60000000000}, |
295 | 318 | }, |
296 | 319 | "min-queries": { |
297 | 320 | "resnet": { |
|
312 | 335 | "dlrm-v2-99.9": {"Server": 270336, "Offline": 1}, |
313 | 336 | "3d-unet-99": {"SingleStream": 1024, "Offline": 1}, |
314 | 337 | "3d-unet-99.9": {"SingleStream": 1024, "Offline": 1}, |
315 | | - "gptj-99": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
316 | | - "gptj-99.9": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
| 338 | + "llama3.1-8b": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
| 339 | + "llama3.1-8b-edge": {"SingleStream": 1024, "Offline": 1}, |
317 | 340 | "llama2-70b-99": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
318 | 341 | "llama2-70b-99.9": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
319 | | - "llama2-70b-interactive-99": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
320 | | - "llama2-70b-interactive-99.9": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
321 | 342 | "stable-diffusion-xl": { |
322 | 343 | "SingleStream": 1024, |
323 | 344 | "Server": 270336, |
|
327 | 348 | "llama3.1-405b": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
328 | 349 | "rgat": {"SingleStream": 1024, "Offline": 1}, |
329 | 350 | "pointpainting": {"SingleStream": 1024}, |
| 351 | + "deepseek-r1": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
| 352 | + "whisper": {"SingleStream": 1024, "Offline": 1}, |
| 353 | + "gpt-oss-120b": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
| 354 | + "qwen3-vl-235b-a22b": {"SingleStream": 1024, "Server": 270336, "Offline": 1}, |
| 355 | + "dlrm-v3": {"Server": 270336, "Offline": 1}, |
330 | 356 | }, |
331 | 357 | }, |
332 | 358 | "v5.1": { |
|
719 | 745 | "rgat": 788379, |
720 | 746 | "deepseek-r1": 4388, |
721 | 747 | "whisper": 1633, |
| 748 | + # TODO: Need to add accuracy sample count checkers as well (4395) |
| 749 | + "gpt-oss-120b": 6396, |
| 750 | + "qwen3-vl-235b-a22b": 9660, |
| 751 | + "dlrm-v3": 34996, |
722 | 752 | } |
723 | 753 |
|
724 | 754 | SCENARIO_MAPPING = { |
|
873 | 903 | "deepseek-r1": { |
874 | 904 | "Server": { |
875 | 905 | "ttft": 2000 * 1000000, "tpot": 80 * 1000000 |
| 906 | + }, |
| 907 | + "Interactive": { |
| 908 | + "ttft": 1500 * 1000000, "tpot": 15 * 1000000 |
876 | 909 | } |
877 | | - } |
| 910 | + }, |
| 911 | + "gpt-oss-120b": { |
| 912 | + "Server": { |
| 913 | + "ttft": 3000 * 1000000, "tpot": 75 * 1000000 |
| 914 | + }, |
| 915 | + "Interactive": { |
| 916 | + "ttft": 1500 * 1000000, "tpot": 15 * 1000000 |
| 917 | + } |
| 918 | + }, |
878 | 919 |
|
879 | 920 | } |
880 | 921 |
|
|
0 commit comments