ilessio-aiflowlab commited on
Commit
009d70e
·
verified ·
1 Parent(s): b27e755

Upload benchmarks/bench_12_full_pipeline_combos.json with huggingface_hub

Browse files
benchmarks/bench_12_full_pipeline_combos.json ADDED
@@ -0,0 +1,340 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "benchmark": "full_pipeline_combos",
3
+ "timestamp": "2026-03-19T12:50:46.264173+00:00",
4
+ "device": "cuda",
5
+ "gpu": "NVIDIA L4",
6
+ "n_gpus": 1,
7
+ "pipelines": {
8
+ "nano_diff_p75_q4": {
9
+ "name": "nano_diff_p75_q4",
10
+ "config": {
11
+ "variant": "nano",
12
+ "language_model": "Qwen/Qwen2.5-0.5B",
13
+ "lora_rank": 32,
14
+ "action_head_type": "diffusion"
15
+ },
16
+ "model": {
17
+ "total_params_m": 967.9,
18
+ "trainable_params_m": 495.6
19
+ },
20
+ "training": {
21
+ "steps": 30,
22
+ "loss_start": 2.8391,
23
+ "loss_end": 1.665,
24
+ "loss_reduction_pct": 41.4,
25
+ "speed_steps_per_s": 1.5
26
+ },
27
+ "pruning": {
28
+ "ratio": 0.75,
29
+ "layers_before": 24,
30
+ "layers_after": 15,
31
+ "layers_removed": [
32
+ 2,
33
+ 3,
34
+ 6,
35
+ 10,
36
+ 13,
37
+ 19,
38
+ 22,
39
+ 23,
40
+ 24
41
+ ],
42
+ "params_after_m": 830.8
43
+ },
44
+ "quantization": {
45
+ "error": "'FORGEStudent' object has no attribute 'get'"
46
+ },
47
+ "inference": {
48
+ "fp32_p50_ms": 99.22,
49
+ "fp32_fps": 10.0,
50
+ "fp16_p50_ms": 83.25,
51
+ "fp16_fps": 12.0,
52
+ "fp16_speedup": 1.19,
53
+ "gpu_mem_gb": 11.67
54
+ },
55
+ "timings": {
56
+ "build_s": 6.0,
57
+ "train_s": 19.8,
58
+ "prune_s": 20.2,
59
+ "quantize_s": 124.7
60
+ },
61
+ "total_time_s": 170.7
62
+ },
63
+ "nano_flow_p50_q4": {
64
+ "name": "nano_flow_p50_q4",
65
+ "config": {
66
+ "variant": "nano",
67
+ "language_model": "Qwen/Qwen2.5-0.5B",
68
+ "lora_rank": 32,
69
+ "action_head_type": "flow"
70
+ },
71
+ "model": {
72
+ "total_params_m": 967.9,
73
+ "trainable_params_m": 495.6
74
+ },
75
+ "training": {
76
+ "steps": 30,
77
+ "loss_start": 7.8611,
78
+ "loss_end": 1.8667,
79
+ "loss_reduction_pct": 76.3,
80
+ "speed_steps_per_s": 1.5
81
+ },
82
+ "pruning": {
83
+ "ratio": 0.5,
84
+ "layers_before": 24,
85
+ "layers_after": 9,
86
+ "layers_removed": [
87
+ 2,
88
+ 3,
89
+ 7,
90
+ 8,
91
+ 9,
92
+ 10,
93
+ 12,
94
+ 14,
95
+ 17,
96
+ 18,
97
+ 20,
98
+ 21,
99
+ 22,
100
+ 23,
101
+ 24
102
+ ],
103
+ "params_after_m": 739.3
104
+ },
105
+ "quantization": {
106
+ "error": "'FORGEStudent' object has no attribute 'get'"
107
+ },
108
+ "inference": {
109
+ "fp32_p50_ms": 70.83,
110
+ "fp32_fps": 14.1,
111
+ "fp16_p50_ms": 132.02,
112
+ "fp16_fps": 7.8,
113
+ "fp16_speedup": 0.56,
114
+ "gpu_mem_gb": 11.36
115
+ },
116
+ "timings": {
117
+ "build_s": 2.9,
118
+ "train_s": 19.5,
119
+ "prune_s": 19.1,
120
+ "quantize_s": 124.2
121
+ },
122
+ "total_time_s": 165.7
123
+ },
124
+ "nano_lora64_p90_q4": {
125
+ "name": "nano_lora64_p90_q4",
126
+ "config": {
127
+ "variant": "nano",
128
+ "language_model": "Qwen/Qwen2.5-0.5B",
129
+ "lora_rank": 64,
130
+ "action_head_type": "diffusion"
131
+ },
132
+ "model": {
133
+ "total_params_m": 972.3,
134
+ "trainable_params_m": 500.0
135
+ },
136
+ "training": {
137
+ "steps": 30,
138
+ "loss_start": 9.6341,
139
+ "loss_end": 1.3188,
140
+ "loss_reduction_pct": 86.3,
141
+ "speed_steps_per_s": 1.4
142
+ },
143
+ "pruning": {
144
+ "ratio": 0.9,
145
+ "layers_before": 24,
146
+ "layers_after": 18,
147
+ "layers_removed": [
148
+ 5,
149
+ 8,
150
+ 9,
151
+ 18,
152
+ 19,
153
+ 23
154
+ ],
155
+ "params_after_m": 880.8
156
+ },
157
+ "quantization": {
158
+ "error": "'FORGEStudent' object has no attribute 'get'"
159
+ },
160
+ "inference": {
161
+ "fp32_p50_ms": 109.74,
162
+ "fp32_fps": 9.1,
163
+ "fp16_p50_ms": 88.12,
164
+ "fp16_fps": 11.2,
165
+ "fp16_speedup": 1.23,
166
+ "gpu_mem_gb": 15.69
167
+ },
168
+ "timings": {
169
+ "build_s": 5.3,
170
+ "train_s": 21.3,
171
+ "prune_s": 19.8,
172
+ "quantize_s": 129.1
173
+ },
174
+ "total_time_s": 175.5
175
+ },
176
+ "nano_diff_p75_q8": {
177
+ "name": "nano_diff_p75_q8",
178
+ "config": {
179
+ "variant": "nano",
180
+ "language_model": "Qwen/Qwen2.5-0.5B",
181
+ "lora_rank": 32,
182
+ "action_head_type": "diffusion"
183
+ },
184
+ "model": {
185
+ "total_params_m": 967.9,
186
+ "trainable_params_m": 495.6
187
+ },
188
+ "training": {
189
+ "steps": 30,
190
+ "loss_start": 13.6661,
191
+ "loss_end": 1.0559,
192
+ "loss_reduction_pct": 92.3,
193
+ "speed_steps_per_s": 1.5
194
+ },
195
+ "pruning": {
196
+ "ratio": 0.75,
197
+ "layers_before": 24,
198
+ "layers_after": 15,
199
+ "layers_removed": [
200
+ 3,
201
+ 4,
202
+ 8,
203
+ 11,
204
+ 12,
205
+ 13,
206
+ 16,
207
+ 19,
208
+ 24
209
+ ],
210
+ "params_after_m": 830.8
211
+ },
212
+ "quantization": {
213
+ "error": "'FORGEStudent' object has no attribute 'get'"
214
+ },
215
+ "inference": {
216
+ "fp32_p50_ms": 100.25,
217
+ "fp32_fps": 10.0,
218
+ "fp16_p50_ms": 86.84,
219
+ "fp16_fps": 11.3,
220
+ "fp16_speedup": 1.14,
221
+ "gpu_mem_gb": 11.7
222
+ },
223
+ "timings": {
224
+ "build_s": 3.1,
225
+ "train_s": 19.7,
226
+ "prune_s": 19.8,
227
+ "quantize_s": 129.2
228
+ },
229
+ "total_time_s": 171.8
230
+ },
231
+ "nano_flow_lora64_p60_q4": {
232
+ "name": "nano_flow_lora64_p60_q4",
233
+ "config": {
234
+ "variant": "nano",
235
+ "language_model": "Qwen/Qwen2.5-0.5B",
236
+ "lora_rank": 64,
237
+ "action_head_type": "flow"
238
+ },
239
+ "model": {
240
+ "total_params_m": 972.3,
241
+ "trainable_params_m": 500.0
242
+ },
243
+ "training": {
244
+ "steps": 30,
245
+ "loss_start": 16.2305,
246
+ "loss_end": 3.939,
247
+ "loss_reduction_pct": 75.7,
248
+ "speed_steps_per_s": 1.5
249
+ },
250
+ "pruning": {
251
+ "ratio": 0.6,
252
+ "layers_before": 24,
253
+ "layers_after": 11,
254
+ "layers_removed": [
255
+ 2,
256
+ 4,
257
+ 6,
258
+ 7,
259
+ 9,
260
+ 10,
261
+ 11,
262
+ 16,
263
+ 18,
264
+ 19,
265
+ 20,
266
+ 22,
267
+ 24
268
+ ],
269
+ "params_after_m": 774.1
270
+ },
271
+ "quantization": {
272
+ "error": "'FORGEStudent' object has no attribute 'get'"
273
+ },
274
+ "inference": {
275
+ "fp32_p50_ms": 78.17,
276
+ "fp32_fps": 12.7,
277
+ "fp16_p50_ms": 69.98,
278
+ "fp16_fps": 14.1,
279
+ "fp16_speedup": 1.11,
280
+ "gpu_mem_gb": 11.54
281
+ },
282
+ "timings": {
283
+ "build_s": 3.1,
284
+ "train_s": 19.8,
285
+ "prune_s": 19.3,
286
+ "quantize_s": 125.4
287
+ },
288
+ "total_time_s": 167.6
289
+ },
290
+ "nano_diff_noprune_q8": {
291
+ "name": "nano_diff_noprune_q8",
292
+ "config": {
293
+ "variant": "nano",
294
+ "language_model": "Qwen/Qwen2.5-0.5B",
295
+ "lora_rank": 32,
296
+ "action_head_type": "diffusion"
297
+ },
298
+ "model": {
299
+ "total_params_m": 967.9,
300
+ "trainable_params_m": 495.6
301
+ },
302
+ "training": {
303
+ "steps": 30,
304
+ "loss_start": 3.97,
305
+ "loss_end": 1.6131,
306
+ "loss_reduction_pct": 59.4,
307
+ "speed_steps_per_s": 1.5
308
+ },
309
+ "pruning": {
310
+ "ratio": 1.0,
311
+ "layers_before": 24,
312
+ "layers_after": 21,
313
+ "layers_removed": [
314
+ 10,
315
+ 14,
316
+ 16
317
+ ],
318
+ "params_after_m": 922.2
319
+ },
320
+ "quantization": {
321
+ "error": "'FORGEStudent' object has no attribute 'get'"
322
+ },
323
+ "inference": {
324
+ "fp32_p50_ms": 123.55,
325
+ "fp32_fps": 8.1,
326
+ "fp16_p50_ms": 90.81,
327
+ "fp16_fps": 11.0,
328
+ "fp16_speedup": 1.37,
329
+ "gpu_mem_gb": 12.04
330
+ },
331
+ "timings": {
332
+ "build_s": 3.0,
333
+ "train_s": 19.6,
334
+ "prune_s": 19.8,
335
+ "quantize_s": 124.8
336
+ },
337
+ "total_time_s": 167.2
338
+ }
339
+ }
340
+ }