ameforge commited on
Commit
77a439e
·
verified ·
1 Parent(s): 9ea1407

Upload benchmark_results.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. benchmark_results.json +1300 -0
benchmark_results.json ADDED
@@ -0,0 +1,1300 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "suite_size": 200,
3
+ "seed": 42,
4
+ "models": [
5
+ {
6
+ "name": "SAM",
7
+ "is_sam": true,
8
+ "params": "35.9M",
9
+ "params_raw": 35911302,
10
+ "size_mb": 137.0,
11
+ "json_valid": 84.0,
12
+ "op_correct": 31.5,
13
+ "exact_match": 8.5,
14
+ "latency_ms": 691.6,
15
+ "n_cases": 200,
16
+ "per_tier": [
17
+ {
18
+ "tier": 1,
19
+ "n": 40,
20
+ "json_valid": 92.5,
21
+ "op_correct": 75.0,
22
+ "exact_match": 20.0
23
+ },
24
+ {
25
+ "tier": 2,
26
+ "n": 40,
27
+ "json_valid": 80.0,
28
+ "op_correct": 32.5,
29
+ "exact_match": 0.0
30
+ },
31
+ {
32
+ "tier": 3,
33
+ "n": 40,
34
+ "json_valid": 77.5,
35
+ "op_correct": 50.0,
36
+ "exact_match": 22.5
37
+ },
38
+ {
39
+ "tier": 4,
40
+ "n": 40,
41
+ "json_valid": 70.0,
42
+ "op_correct": 0.0,
43
+ "exact_match": 0.0
44
+ },
45
+ {
46
+ "tier": 5,
47
+ "n": 40,
48
+ "json_valid": 100.0,
49
+ "op_correct": 0.0,
50
+ "exact_match": 0.0
51
+ }
52
+ ],
53
+ "per_domain": [
54
+ {
55
+ "domain": "CAL",
56
+ "n": 2,
57
+ "json_valid": 100.0,
58
+ "op_correct": 100.0,
59
+ "exact_match": 50.0
60
+ },
61
+ {
62
+ "domain": "DB",
63
+ "n": 27,
64
+ "json_valid": 77.8,
65
+ "op_correct": 48.1,
66
+ "exact_match": 3.7
67
+ },
68
+ {
69
+ "domain": "ECOMMERCE",
70
+ "n": 8,
71
+ "json_valid": 100.0,
72
+ "op_correct": 87.5,
73
+ "exact_match": 0.0
74
+ },
75
+ {
76
+ "domain": "FILE",
77
+ "n": 4,
78
+ "json_valid": 100.0,
79
+ "op_correct": 50.0,
80
+ "exact_match": 0.0
81
+ },
82
+ {
83
+ "domain": "HOME",
84
+ "n": 7,
85
+ "json_valid": 100.0,
86
+ "op_correct": 28.6,
87
+ "exact_match": 0.0
88
+ },
89
+ {
90
+ "domain": "HTTP",
91
+ "n": 25,
92
+ "json_valid": 96.0,
93
+ "op_correct": 60.0,
94
+ "exact_match": 40.0
95
+ },
96
+ {
97
+ "domain": "MQTT",
98
+ "n": 15,
99
+ "json_valid": 33.3,
100
+ "op_correct": 33.3,
101
+ "exact_match": 0.0
102
+ },
103
+ {
104
+ "domain": "ROS",
105
+ "n": 46,
106
+ "json_valid": 87.0,
107
+ "op_correct": 26.1,
108
+ "exact_match": 4.3
109
+ },
110
+ {
111
+ "domain": "VEHICLE",
112
+ "n": 17,
113
+ "json_valid": 100.0,
114
+ "op_correct": 29.4,
115
+ "exact_match": 17.6
116
+ },
117
+ {
118
+ "domain": "WORKFLOW",
119
+ "n": 49,
120
+ "json_valid": 81.6,
121
+ "op_correct": 0.0,
122
+ "exact_match": 0.0
123
+ }
124
+ ]
125
+ },
126
+ {
127
+ "name": "SmolLM2-360M",
128
+ "is_sam": false,
129
+ "params": "361.8M",
130
+ "params_raw": 361821120,
131
+ "size_mb": 690.1,
132
+ "json_valid": 98.5,
133
+ "op_correct": 0.0,
134
+ "exact_match": 0.0,
135
+ "latency_ms": 1027.1,
136
+ "n_cases": 200,
137
+ "per_tier": [
138
+ {
139
+ "tier": 1,
140
+ "n": 40,
141
+ "json_valid": 100.0,
142
+ "op_correct": 0.0,
143
+ "exact_match": 0.0
144
+ },
145
+ {
146
+ "tier": 2,
147
+ "n": 40,
148
+ "json_valid": 100.0,
149
+ "op_correct": 0.0,
150
+ "exact_match": 0.0
151
+ },
152
+ {
153
+ "tier": 3,
154
+ "n": 40,
155
+ "json_valid": 100.0,
156
+ "op_correct": 0.0,
157
+ "exact_match": 0.0
158
+ },
159
+ {
160
+ "tier": 4,
161
+ "n": 40,
162
+ "json_valid": 92.5,
163
+ "op_correct": 0.0,
164
+ "exact_match": 0.0
165
+ },
166
+ {
167
+ "tier": 5,
168
+ "n": 40,
169
+ "json_valid": 100.0,
170
+ "op_correct": 0.0,
171
+ "exact_match": 0.0
172
+ }
173
+ ],
174
+ "per_domain": [
175
+ {
176
+ "domain": "CAL",
177
+ "n": 2,
178
+ "json_valid": 100.0,
179
+ "op_correct": 0.0,
180
+ "exact_match": 0.0
181
+ },
182
+ {
183
+ "domain": "DB",
184
+ "n": 27,
185
+ "json_valid": 100.0,
186
+ "op_correct": 0.0,
187
+ "exact_match": 0.0
188
+ },
189
+ {
190
+ "domain": "ECOMMERCE",
191
+ "n": 8,
192
+ "json_valid": 100.0,
193
+ "op_correct": 0.0,
194
+ "exact_match": 0.0
195
+ },
196
+ {
197
+ "domain": "FILE",
198
+ "n": 4,
199
+ "json_valid": 100.0,
200
+ "op_correct": 0.0,
201
+ "exact_match": 0.0
202
+ },
203
+ {
204
+ "domain": "HOME",
205
+ "n": 7,
206
+ "json_valid": 100.0,
207
+ "op_correct": 0.0,
208
+ "exact_match": 0.0
209
+ },
210
+ {
211
+ "domain": "HTTP",
212
+ "n": 25,
213
+ "json_valid": 100.0,
214
+ "op_correct": 0.0,
215
+ "exact_match": 0.0
216
+ },
217
+ {
218
+ "domain": "MQTT",
219
+ "n": 15,
220
+ "json_valid": 100.0,
221
+ "op_correct": 0.0,
222
+ "exact_match": 0.0
223
+ },
224
+ {
225
+ "domain": "ROS",
226
+ "n": 46,
227
+ "json_valid": 93.5,
228
+ "op_correct": 0.0,
229
+ "exact_match": 0.0
230
+ },
231
+ {
232
+ "domain": "VEHICLE",
233
+ "n": 17,
234
+ "json_valid": 100.0,
235
+ "op_correct": 0.0,
236
+ "exact_match": 0.0
237
+ },
238
+ {
239
+ "domain": "WORKFLOW",
240
+ "n": 49,
241
+ "json_valid": 100.0,
242
+ "op_correct": 0.0,
243
+ "exact_match": 0.0
244
+ }
245
+ ]
246
+ },
247
+ {
248
+ "name": "Qwen2.5-0.5B",
249
+ "is_sam": false,
250
+ "params": "494.0M",
251
+ "params_raw": 494032768,
252
+ "size_mb": 942.3,
253
+ "json_valid": 71.5,
254
+ "op_correct": 28.5,
255
+ "exact_match": 2.5,
256
+ "latency_ms": 2108.7,
257
+ "n_cases": 200,
258
+ "per_tier": [
259
+ {
260
+ "tier": 1,
261
+ "n": 40,
262
+ "json_valid": 60.0,
263
+ "op_correct": 42.5,
264
+ "exact_match": 12.5
265
+ },
266
+ {
267
+ "tier": 2,
268
+ "n": 40,
269
+ "json_valid": 65.0,
270
+ "op_correct": 35.0,
271
+ "exact_match": 0.0
272
+ },
273
+ {
274
+ "tier": 3,
275
+ "n": 40,
276
+ "json_valid": 67.5,
277
+ "op_correct": 45.0,
278
+ "exact_match": 0.0
279
+ },
280
+ {
281
+ "tier": 4,
282
+ "n": 40,
283
+ "json_valid": 100.0,
284
+ "op_correct": 20.0,
285
+ "exact_match": 0.0
286
+ },
287
+ {
288
+ "tier": 5,
289
+ "n": 40,
290
+ "json_valid": 65.0,
291
+ "op_correct": 0.0,
292
+ "exact_match": 0.0
293
+ }
294
+ ],
295
+ "per_domain": [
296
+ {
297
+ "domain": "CAL",
298
+ "n": 2,
299
+ "json_valid": 100.0,
300
+ "op_correct": 50.0,
301
+ "exact_match": 0.0
302
+ },
303
+ {
304
+ "domain": "DB",
305
+ "n": 27,
306
+ "json_valid": 51.9,
307
+ "op_correct": 7.4,
308
+ "exact_match": 0.0
309
+ },
310
+ {
311
+ "domain": "ECOMMERCE",
312
+ "n": 8,
313
+ "json_valid": 25.0,
314
+ "op_correct": 25.0,
315
+ "exact_match": 0.0
316
+ },
317
+ {
318
+ "domain": "FILE",
319
+ "n": 4,
320
+ "json_valid": 100.0,
321
+ "op_correct": 100.0,
322
+ "exact_match": 0.0
323
+ },
324
+ {
325
+ "domain": "HOME",
326
+ "n": 7,
327
+ "json_valid": 100.0,
328
+ "op_correct": 42.9,
329
+ "exact_match": 0.0
330
+ },
331
+ {
332
+ "domain": "HTTP",
333
+ "n": 25,
334
+ "json_valid": 72.0,
335
+ "op_correct": 36.0,
336
+ "exact_match": 0.0
337
+ },
338
+ {
339
+ "domain": "MQTT",
340
+ "n": 15,
341
+ "json_valid": 86.7,
342
+ "op_correct": 26.7,
343
+ "exact_match": 0.0
344
+ },
345
+ {
346
+ "domain": "ROS",
347
+ "n": 46,
348
+ "json_valid": 73.9,
349
+ "op_correct": 43.5,
350
+ "exact_match": 4.3
351
+ },
352
+ {
353
+ "domain": "VEHICLE",
354
+ "n": 17,
355
+ "json_valid": 82.4,
356
+ "op_correct": 64.7,
357
+ "exact_match": 17.6
358
+ },
359
+ {
360
+ "domain": "WORKFLOW",
361
+ "n": 49,
362
+ "json_valid": 71.4,
363
+ "op_correct": 2.0,
364
+ "exact_match": 0.0
365
+ }
366
+ ]
367
+ },
368
+ {
369
+ "name": "TinyLlama-1.1B",
370
+ "is_sam": false,
371
+ "params": "1100.0M",
372
+ "params_raw": 1100048384,
373
+ "size_mb": 2098.2,
374
+ "json_valid": 93.5,
375
+ "op_correct": 57.0,
376
+ "exact_match": 9.0,
377
+ "latency_ms": 4484.0,
378
+ "n_cases": 200,
379
+ "per_tier": [
380
+ {
381
+ "tier": 1,
382
+ "n": 40,
383
+ "json_valid": 100.0,
384
+ "op_correct": 95.0,
385
+ "exact_match": 27.5
386
+ },
387
+ {
388
+ "tier": 2,
389
+ "n": 40,
390
+ "json_valid": 90.0,
391
+ "op_correct": 90.0,
392
+ "exact_match": 12.5
393
+ },
394
+ {
395
+ "tier": 3,
396
+ "n": 40,
397
+ "json_valid": 100.0,
398
+ "op_correct": 77.5,
399
+ "exact_match": 5.0
400
+ },
401
+ {
402
+ "tier": 4,
403
+ "n": 40,
404
+ "json_valid": 77.5,
405
+ "op_correct": 22.5,
406
+ "exact_match": 0.0
407
+ },
408
+ {
409
+ "tier": 5,
410
+ "n": 40,
411
+ "json_valid": 100.0,
412
+ "op_correct": 0.0,
413
+ "exact_match": 0.0
414
+ }
415
+ ],
416
+ "per_domain": [
417
+ {
418
+ "domain": "CAL",
419
+ "n": 2,
420
+ "json_valid": 100.0,
421
+ "op_correct": 100.0,
422
+ "exact_match": 0.0
423
+ },
424
+ {
425
+ "domain": "DB",
426
+ "n": 27,
427
+ "json_valid": 100.0,
428
+ "op_correct": 66.7,
429
+ "exact_match": 18.5
430
+ },
431
+ {
432
+ "domain": "ECOMMERCE",
433
+ "n": 8,
434
+ "json_valid": 100.0,
435
+ "op_correct": 100.0,
436
+ "exact_match": 0.0
437
+ },
438
+ {
439
+ "domain": "FILE",
440
+ "n": 4,
441
+ "json_valid": 100.0,
442
+ "op_correct": 100.0,
443
+ "exact_match": 25.0
444
+ },
445
+ {
446
+ "domain": "HOME",
447
+ "n": 7,
448
+ "json_valid": 100.0,
449
+ "op_correct": 100.0,
450
+ "exact_match": 0.0
451
+ },
452
+ {
453
+ "domain": "HTTP",
454
+ "n": 25,
455
+ "json_valid": 84.0,
456
+ "op_correct": 80.0,
457
+ "exact_match": 0.0
458
+ },
459
+ {
460
+ "domain": "MQTT",
461
+ "n": 15,
462
+ "json_valid": 100.0,
463
+ "op_correct": 100.0,
464
+ "exact_match": 0.0
465
+ },
466
+ {
467
+ "domain": "ROS",
468
+ "n": 46,
469
+ "json_valid": 80.4,
470
+ "op_correct": 54.3,
471
+ "exact_match": 17.4
472
+ },
473
+ {
474
+ "domain": "VEHICLE",
475
+ "n": 17,
476
+ "json_valid": 100.0,
477
+ "op_correct": 35.3,
478
+ "exact_match": 23.5
479
+ },
480
+ {
481
+ "domain": "WORKFLOW",
482
+ "n": 49,
483
+ "json_valid": 100.0,
484
+ "op_correct": 18.4,
485
+ "exact_match": 0.0
486
+ }
487
+ ]
488
+ },
489
+ {
490
+ "name": "Llama-3.2-1B",
491
+ "is_sam": false,
492
+ "params": "?",
493
+ "params_raw": 0,
494
+ "size_mb": 0,
495
+ "json_valid": 0,
496
+ "op_correct": 0,
497
+ "exact_match": 0,
498
+ "latency_ms": 0,
499
+ "error": "load_failed"
500
+ },
501
+ {
502
+ "name": "Qwen2.5-1.5B",
503
+ "is_sam": false,
504
+ "params": "1543.7M",
505
+ "params_raw": 1543714304,
506
+ "size_mb": 2944.4,
507
+ "json_valid": 100.0,
508
+ "op_correct": 75.0,
509
+ "exact_match": 9.5,
510
+ "latency_ms": 1976.9,
511
+ "n_cases": 200,
512
+ "per_tier": [
513
+ {
514
+ "tier": 1,
515
+ "n": 40,
516
+ "json_valid": 100.0,
517
+ "op_correct": 95.0,
518
+ "exact_match": 30.0
519
+ },
520
+ {
521
+ "tier": 2,
522
+ "n": 40,
523
+ "json_valid": 100.0,
524
+ "op_correct": 90.0,
525
+ "exact_match": 15.0
526
+ },
527
+ {
528
+ "tier": 3,
529
+ "n": 40,
530
+ "json_valid": 100.0,
531
+ "op_correct": 100.0,
532
+ "exact_match": 2.5
533
+ },
534
+ {
535
+ "tier": 4,
536
+ "n": 40,
537
+ "json_valid": 100.0,
538
+ "op_correct": 50.0,
539
+ "exact_match": 0.0
540
+ },
541
+ {
542
+ "tier": 5,
543
+ "n": 40,
544
+ "json_valid": 100.0,
545
+ "op_correct": 40.0,
546
+ "exact_match": 0.0
547
+ }
548
+ ],
549
+ "per_domain": [
550
+ {
551
+ "domain": "CAL",
552
+ "n": 2,
553
+ "json_valid": 100.0,
554
+ "op_correct": 100.0,
555
+ "exact_match": 0.0
556
+ },
557
+ {
558
+ "domain": "DB",
559
+ "n": 27,
560
+ "json_valid": 100.0,
561
+ "op_correct": 66.7,
562
+ "exact_match": 18.5
563
+ },
564
+ {
565
+ "domain": "ECOMMERCE",
566
+ "n": 8,
567
+ "json_valid": 100.0,
568
+ "op_correct": 100.0,
569
+ "exact_match": 0.0
570
+ },
571
+ {
572
+ "domain": "FILE",
573
+ "n": 4,
574
+ "json_valid": 100.0,
575
+ "op_correct": 100.0,
576
+ "exact_match": 0.0
577
+ },
578
+ {
579
+ "domain": "HOME",
580
+ "n": 7,
581
+ "json_valid": 100.0,
582
+ "op_correct": 85.7,
583
+ "exact_match": 0.0
584
+ },
585
+ {
586
+ "domain": "HTTP",
587
+ "n": 25,
588
+ "json_valid": 100.0,
589
+ "op_correct": 84.0,
590
+ "exact_match": 4.0
591
+ },
592
+ {
593
+ "domain": "MQTT",
594
+ "n": 15,
595
+ "json_valid": 100.0,
596
+ "op_correct": 93.3,
597
+ "exact_match": 0.0
598
+ },
599
+ {
600
+ "domain": "ROS",
601
+ "n": 46,
602
+ "json_valid": 100.0,
603
+ "op_correct": 76.1,
604
+ "exact_match": 21.7
605
+ },
606
+ {
607
+ "domain": "VEHICLE",
608
+ "n": 17,
609
+ "json_valid": 100.0,
610
+ "op_correct": 100.0,
611
+ "exact_match": 17.6
612
+ },
613
+ {
614
+ "domain": "WORKFLOW",
615
+ "n": 49,
616
+ "json_valid": 100.0,
617
+ "op_correct": 51.0,
618
+ "exact_match": 0.0
619
+ }
620
+ ]
621
+ },
622
+ {
623
+ "name": "SmolLM2-1.7B",
624
+ "is_sam": false,
625
+ "params": "1711.4M",
626
+ "params_raw": 1711376384,
627
+ "size_mb": 3264.2,
628
+ "json_valid": 85.0,
629
+ "op_correct": 47.0,
630
+ "exact_match": 2.5,
631
+ "latency_ms": 1884.5,
632
+ "n_cases": 200,
633
+ "per_tier": [
634
+ {
635
+ "tier": 1,
636
+ "n": 40,
637
+ "json_valid": 92.5,
638
+ "op_correct": 75.0,
639
+ "exact_match": 12.5
640
+ },
641
+ {
642
+ "tier": 2,
643
+ "n": 40,
644
+ "json_valid": 72.5,
645
+ "op_correct": 62.5,
646
+ "exact_match": 0.0
647
+ },
648
+ {
649
+ "tier": 3,
650
+ "n": 40,
651
+ "json_valid": 100.0,
652
+ "op_correct": 55.0,
653
+ "exact_match": 0.0
654
+ },
655
+ {
656
+ "tier": 4,
657
+ "n": 40,
658
+ "json_valid": 60.0,
659
+ "op_correct": 2.5,
660
+ "exact_match": 0.0
661
+ },
662
+ {
663
+ "tier": 5,
664
+ "n": 40,
665
+ "json_valid": 100.0,
666
+ "op_correct": 40.0,
667
+ "exact_match": 0.0
668
+ }
669
+ ],
670
+ "per_domain": [
671
+ {
672
+ "domain": "CAL",
673
+ "n": 2,
674
+ "json_valid": 100.0,
675
+ "op_correct": 100.0,
676
+ "exact_match": 0.0
677
+ },
678
+ {
679
+ "domain": "DB",
680
+ "n": 27,
681
+ "json_valid": 66.7,
682
+ "op_correct": 59.3,
683
+ "exact_match": 0.0
684
+ },
685
+ {
686
+ "domain": "ECOMMERCE",
687
+ "n": 8,
688
+ "json_valid": 100.0,
689
+ "op_correct": 100.0,
690
+ "exact_match": 0.0
691
+ },
692
+ {
693
+ "domain": "FILE",
694
+ "n": 4,
695
+ "json_valid": 100.0,
696
+ "op_correct": 100.0,
697
+ "exact_match": 0.0
698
+ },
699
+ {
700
+ "domain": "HOME",
701
+ "n": 7,
702
+ "json_valid": 85.7,
703
+ "op_correct": 57.1,
704
+ "exact_match": 0.0
705
+ },
706
+ {
707
+ "domain": "HTTP",
708
+ "n": 25,
709
+ "json_valid": 100.0,
710
+ "op_correct": 4.0,
711
+ "exact_match": 0.0
712
+ },
713
+ {
714
+ "domain": "MQTT",
715
+ "n": 15,
716
+ "json_valid": 100.0,
717
+ "op_correct": 100.0,
718
+ "exact_match": 0.0
719
+ },
720
+ {
721
+ "domain": "ROS",
722
+ "n": 46,
723
+ "json_valid": 89.1,
724
+ "op_correct": 47.8,
725
+ "exact_match": 4.3
726
+ },
727
+ {
728
+ "domain": "VEHICLE",
729
+ "n": 17,
730
+ "json_valid": 58.8,
731
+ "op_correct": 29.4,
732
+ "exact_match": 17.6
733
+ },
734
+ {
735
+ "domain": "WORKFLOW",
736
+ "n": 49,
737
+ "json_valid": 83.7,
738
+ "op_correct": 34.7,
739
+ "exact_match": 0.0
740
+ }
741
+ ]
742
+ },
743
+ {
744
+ "name": "Gemma-2-2B",
745
+ "is_sam": false,
746
+ "params": "?",
747
+ "params_raw": 0,
748
+ "size_mb": 0,
749
+ "json_valid": 0,
750
+ "op_correct": 0,
751
+ "exact_match": 0,
752
+ "latency_ms": 0,
753
+ "error": "load_failed"
754
+ },
755
+ {
756
+ "name": "Qwen2.5-3B",
757
+ "is_sam": false,
758
+ "params": "3085.9M",
759
+ "params_raw": 3085938688,
760
+ "size_mb": 5886.0,
761
+ "json_valid": 100.0,
762
+ "op_correct": 81.5,
763
+ "exact_match": 17.5,
764
+ "latency_ms": 2716.7,
765
+ "n_cases": 200,
766
+ "per_tier": [
767
+ {
768
+ "tier": 1,
769
+ "n": 40,
770
+ "json_valid": 100.0,
771
+ "op_correct": 97.5,
772
+ "exact_match": 40.0
773
+ },
774
+ {
775
+ "tier": 2,
776
+ "n": 40,
777
+ "json_valid": 100.0,
778
+ "op_correct": 100.0,
779
+ "exact_match": 20.0
780
+ },
781
+ {
782
+ "tier": 3,
783
+ "n": 40,
784
+ "json_valid": 100.0,
785
+ "op_correct": 100.0,
786
+ "exact_match": 27.5
787
+ },
788
+ {
789
+ "tier": 4,
790
+ "n": 40,
791
+ "json_valid": 100.0,
792
+ "op_correct": 70.0,
793
+ "exact_match": 0.0
794
+ },
795
+ {
796
+ "tier": 5,
797
+ "n": 40,
798
+ "json_valid": 100.0,
799
+ "op_correct": 40.0,
800
+ "exact_match": 0.0
801
+ }
802
+ ],
803
+ "per_domain": [
804
+ {
805
+ "domain": "CAL",
806
+ "n": 2,
807
+ "json_valid": 100.0,
808
+ "op_correct": 100.0,
809
+ "exact_match": 50.0
810
+ },
811
+ {
812
+ "domain": "DB",
813
+ "n": 27,
814
+ "json_valid": 100.0,
815
+ "op_correct": 66.7,
816
+ "exact_match": 18.5
817
+ },
818
+ {
819
+ "domain": "ECOMMERCE",
820
+ "n": 8,
821
+ "json_valid": 100.0,
822
+ "op_correct": 100.0,
823
+ "exact_match": 0.0
824
+ },
825
+ {
826
+ "domain": "FILE",
827
+ "n": 4,
828
+ "json_valid": 100.0,
829
+ "op_correct": 75.0,
830
+ "exact_match": 0.0
831
+ },
832
+ {
833
+ "domain": "HOME",
834
+ "n": 7,
835
+ "json_valid": 100.0,
836
+ "op_correct": 100.0,
837
+ "exact_match": 0.0
838
+ },
839
+ {
840
+ "domain": "HTTP",
841
+ "n": 25,
842
+ "json_valid": 100.0,
843
+ "op_correct": 100.0,
844
+ "exact_match": 8.0
845
+ },
846
+ {
847
+ "domain": "MQTT",
848
+ "n": 15,
849
+ "json_valid": 100.0,
850
+ "op_correct": 100.0,
851
+ "exact_match": 73.3
852
+ },
853
+ {
854
+ "domain": "ROS",
855
+ "n": 46,
856
+ "json_valid": 100.0,
857
+ "op_correct": 93.5,
858
+ "exact_match": 28.3
859
+ },
860
+ {
861
+ "domain": "VEHICLE",
862
+ "n": 17,
863
+ "json_valid": 100.0,
864
+ "op_correct": 100.0,
865
+ "exact_match": 17.6
866
+ },
867
+ {
868
+ "domain": "WORKFLOW",
869
+ "n": 49,
870
+ "json_valid": 100.0,
871
+ "op_correct": 51.0,
872
+ "exact_match": 0.0
873
+ }
874
+ ]
875
+ },
876
+ {
877
+ "name": "Llama-3.2-3B",
878
+ "is_sam": false,
879
+ "params": "?",
880
+ "params_raw": 0,
881
+ "size_mb": 0,
882
+ "json_valid": 0,
883
+ "op_correct": 0,
884
+ "exact_match": 0,
885
+ "latency_ms": 0,
886
+ "error": "load_failed"
887
+ },
888
+ {
889
+ "name": "Phi-3.5-mini-3.8B",
890
+ "is_sam": false,
891
+ "params": "?",
892
+ "params_raw": 0,
893
+ "size_mb": 0,
894
+ "json_valid": 0,
895
+ "op_correct": 0,
896
+ "exact_match": 0,
897
+ "latency_ms": 0,
898
+ "error": "load_failed"
899
+ },
900
+ {
901
+ "name": "Llama-3.2-3B-uns",
902
+ "is_sam": false,
903
+ "params": "?",
904
+ "params_raw": 0,
905
+ "size_mb": 0,
906
+ "json_valid": 0,
907
+ "op_correct": 0,
908
+ "exact_match": 0,
909
+ "latency_ms": 0,
910
+ "error": "load_failed"
911
+ },
912
+ {
913
+ "name": "Gemma-3-270M-GGUF",
914
+ "is_sam": false,
915
+ "params": "270.0M",
916
+ "params_raw": 270000000,
917
+ "size_mb": 241.4,
918
+ "json_valid": 92.0,
919
+ "op_correct": 2.5,
920
+ "exact_match": 0.0,
921
+ "latency_ms": 6871.5,
922
+ "n_cases": 200,
923
+ "per_tier": [
924
+ {
925
+ "tier": 1,
926
+ "n": 40,
927
+ "json_valid": 85.0,
928
+ "op_correct": 5.0,
929
+ "exact_match": 0.0
930
+ },
931
+ {
932
+ "tier": 2,
933
+ "n": 40,
934
+ "json_valid": 97.5,
935
+ "op_correct": 7.5,
936
+ "exact_match": 0.0
937
+ },
938
+ {
939
+ "tier": 3,
940
+ "n": 40,
941
+ "json_valid": 90.0,
942
+ "op_correct": 0.0,
943
+ "exact_match": 0.0
944
+ },
945
+ {
946
+ "tier": 4,
947
+ "n": 40,
948
+ "json_valid": 87.5,
949
+ "op_correct": 0.0,
950
+ "exact_match": 0.0
951
+ },
952
+ {
953
+ "tier": 5,
954
+ "n": 40,
955
+ "json_valid": 100.0,
956
+ "op_correct": 0.0,
957
+ "exact_match": 0.0
958
+ }
959
+ ],
960
+ "per_domain": [
961
+ {
962
+ "domain": "CAL",
963
+ "n": 2,
964
+ "json_valid": 100.0,
965
+ "op_correct": 0.0,
966
+ "exact_match": 0.0
967
+ },
968
+ {
969
+ "domain": "DB",
970
+ "n": 27,
971
+ "json_valid": 85.2,
972
+ "op_correct": 0.0,
973
+ "exact_match": 0.0
974
+ },
975
+ {
976
+ "domain": "ECOMMERCE",
977
+ "n": 8,
978
+ "json_valid": 100.0,
979
+ "op_correct": 12.5,
980
+ "exact_match": 0.0
981
+ },
982
+ {
983
+ "domain": "FILE",
984
+ "n": 4,
985
+ "json_valid": 100.0,
986
+ "op_correct": 0.0,
987
+ "exact_match": 0.0
988
+ },
989
+ {
990
+ "domain": "HOME",
991
+ "n": 7,
992
+ "json_valid": 100.0,
993
+ "op_correct": 42.9,
994
+ "exact_match": 0.0
995
+ },
996
+ {
997
+ "domain": "HTTP",
998
+ "n": 25,
999
+ "json_valid": 92.0,
1000
+ "op_correct": 0.0,
1001
+ "exact_match": 0.0
1002
+ },
1003
+ {
1004
+ "domain": "MQTT",
1005
+ "n": 15,
1006
+ "json_valid": 66.7,
1007
+ "op_correct": 0.0,
1008
+ "exact_match": 0.0
1009
+ },
1010
+ {
1011
+ "domain": "ROS",
1012
+ "n": 46,
1013
+ "json_valid": 89.1,
1014
+ "op_correct": 2.2,
1015
+ "exact_match": 0.0
1016
+ },
1017
+ {
1018
+ "domain": "VEHICLE",
1019
+ "n": 17,
1020
+ "json_valid": 100.0,
1021
+ "op_correct": 0.0,
1022
+ "exact_match": 0.0
1023
+ },
1024
+ {
1025
+ "domain": "WORKFLOW",
1026
+ "n": 49,
1027
+ "json_valid": 100.0,
1028
+ "op_correct": 0.0,
1029
+ "exact_match": 0.0
1030
+ }
1031
+ ]
1032
+ },
1033
+ {
1034
+ "name": "Ministral-3-8B-GGUF",
1035
+ "is_sam": false,
1036
+ "params": "?",
1037
+ "params_raw": 0,
1038
+ "size_mb": 0,
1039
+ "json_valid": 0,
1040
+ "op_correct": 0,
1041
+ "exact_match": 0,
1042
+ "latency_ms": 0,
1043
+ "error": "load_failed"
1044
+ },
1045
+ {
1046
+ "name": "DeepSeek-R1-Qwen3-8B-GGUF",
1047
+ "is_sam": false,
1048
+ "params": "8000.0M",
1049
+ "params_raw": 8000000000,
1050
+ "size_mb": 4794.9,
1051
+ "json_valid": 98.5,
1052
+ "op_correct": 79.5,
1053
+ "exact_match": 11.5,
1054
+ "latency_ms": 37642.4,
1055
+ "n_cases": 200,
1056
+ "per_tier": [
1057
+ {
1058
+ "tier": 1,
1059
+ "n": 40,
1060
+ "json_valid": 100.0,
1061
+ "op_correct": 90.0,
1062
+ "exact_match": 27.5
1063
+ },
1064
+ {
1065
+ "tier": 2,
1066
+ "n": 40,
1067
+ "json_valid": 95.0,
1068
+ "op_correct": 95.0,
1069
+ "exact_match": 7.5
1070
+ },
1071
+ {
1072
+ "tier": 3,
1073
+ "n": 40,
1074
+ "json_valid": 100.0,
1075
+ "op_correct": 77.5,
1076
+ "exact_match": 22.5
1077
+ },
1078
+ {
1079
+ "tier": 4,
1080
+ "n": 40,
1081
+ "json_valid": 97.5,
1082
+ "op_correct": 95.0,
1083
+ "exact_match": 0.0
1084
+ },
1085
+ {
1086
+ "tier": 5,
1087
+ "n": 40,
1088
+ "json_valid": 100.0,
1089
+ "op_correct": 40.0,
1090
+ "exact_match": 0.0
1091
+ }
1092
+ ],
1093
+ "per_domain": [
1094
+ {
1095
+ "domain": "CAL",
1096
+ "n": 2,
1097
+ "json_valid": 100.0,
1098
+ "op_correct": 100.0,
1099
+ "exact_match": 50.0
1100
+ },
1101
+ {
1102
+ "domain": "DB",
1103
+ "n": 27,
1104
+ "json_valid": 100.0,
1105
+ "op_correct": 100.0,
1106
+ "exact_match": 0.0
1107
+ },
1108
+ {
1109
+ "domain": "ECOMMERCE",
1110
+ "n": 8,
1111
+ "json_valid": 100.0,
1112
+ "op_correct": 87.5,
1113
+ "exact_match": 0.0
1114
+ },
1115
+ {
1116
+ "domain": "FILE",
1117
+ "n": 4,
1118
+ "json_valid": 100.0,
1119
+ "op_correct": 100.0,
1120
+ "exact_match": 25.0
1121
+ },
1122
+ {
1123
+ "domain": "HOME",
1124
+ "n": 7,
1125
+ "json_valid": 100.0,
1126
+ "op_correct": 85.7,
1127
+ "exact_match": 0.0
1128
+ },
1129
+ {
1130
+ "domain": "HTTP",
1131
+ "n": 25,
1132
+ "json_valid": 92.0,
1133
+ "op_correct": 92.0,
1134
+ "exact_match": 8.0
1135
+ },
1136
+ {
1137
+ "domain": "MQTT",
1138
+ "n": 15,
1139
+ "json_valid": 100.0,
1140
+ "op_correct": 100.0,
1141
+ "exact_match": 73.3
1142
+ },
1143
+ {
1144
+ "domain": "ROS",
1145
+ "n": 46,
1146
+ "json_valid": 97.8,
1147
+ "op_correct": 71.7,
1148
+ "exact_match": 10.9
1149
+ },
1150
+ {
1151
+ "domain": "VEHICLE",
1152
+ "n": 17,
1153
+ "json_valid": 100.0,
1154
+ "op_correct": 100.0,
1155
+ "exact_match": 17.6
1156
+ },
1157
+ {
1158
+ "domain": "WORKFLOW",
1159
+ "n": 49,
1160
+ "json_valid": 100.0,
1161
+ "op_correct": 51.0,
1162
+ "exact_match": 0.0
1163
+ }
1164
+ ]
1165
+ }
1166
+ ],
1167
+ "per_tier": [
1168
+ {
1169
+ "tier": 1,
1170
+ "n": 40,
1171
+ "json_valid": 92.5,
1172
+ "op_correct": 75.0,
1173
+ "exact_match": 20.0,
1174
+ "description": "Atomic \u2014 single op"
1175
+ },
1176
+ {
1177
+ "tier": 2,
1178
+ "n": 40,
1179
+ "json_valid": 80.0,
1180
+ "op_correct": 32.5,
1181
+ "exact_match": 0.0,
1182
+ "description": "Compound \u2014 2-3 ops same domain"
1183
+ },
1184
+ {
1185
+ "tier": 3,
1186
+ "n": 40,
1187
+ "json_valid": 77.5,
1188
+ "op_correct": 50.0,
1189
+ "exact_match": 22.5,
1190
+ "description": "Noisy \u2014 typos / casing / fillers"
1191
+ },
1192
+ {
1193
+ "tier": 4,
1194
+ "n": 40,
1195
+ "json_valid": 70.0,
1196
+ "op_correct": 0.0,
1197
+ "exact_match": 0.0,
1198
+ "description": "Long chain \u2014 4+ ops"
1199
+ },
1200
+ {
1201
+ "tier": 5,
1202
+ "n": 40,
1203
+ "json_valid": 100.0,
1204
+ "op_correct": 0.0,
1205
+ "exact_match": 0.0,
1206
+ "description": "Cross-domain workflow"
1207
+ }
1208
+ ],
1209
+ "per_domain": [
1210
+ {
1211
+ "domain": "CAL",
1212
+ "n": 2,
1213
+ "json_valid": 100.0,
1214
+ "op_correct": 100.0,
1215
+ "exact_match": 50.0
1216
+ },
1217
+ {
1218
+ "domain": "DB",
1219
+ "n": 27,
1220
+ "json_valid": 77.8,
1221
+ "op_correct": 48.1,
1222
+ "exact_match": 3.7
1223
+ },
1224
+ {
1225
+ "domain": "ECOMMERCE",
1226
+ "n": 8,
1227
+ "json_valid": 100.0,
1228
+ "op_correct": 87.5,
1229
+ "exact_match": 0.0
1230
+ },
1231
+ {
1232
+ "domain": "FILE",
1233
+ "n": 4,
1234
+ "json_valid": 100.0,
1235
+ "op_correct": 50.0,
1236
+ "exact_match": 0.0
1237
+ },
1238
+ {
1239
+ "domain": "HOME",
1240
+ "n": 7,
1241
+ "json_valid": 100.0,
1242
+ "op_correct": 28.6,
1243
+ "exact_match": 0.0
1244
+ },
1245
+ {
1246
+ "domain": "HTTP",
1247
+ "n": 25,
1248
+ "json_valid": 96.0,
1249
+ "op_correct": 60.0,
1250
+ "exact_match": 40.0
1251
+ },
1252
+ {
1253
+ "domain": "MQTT",
1254
+ "n": 15,
1255
+ "json_valid": 33.3,
1256
+ "op_correct": 33.3,
1257
+ "exact_match": 0.0
1258
+ },
1259
+ {
1260
+ "domain": "ROS",
1261
+ "n": 46,
1262
+ "json_valid": 87.0,
1263
+ "op_correct": 26.1,
1264
+ "exact_match": 4.3
1265
+ },
1266
+ {
1267
+ "domain": "VEHICLE",
1268
+ "n": 17,
1269
+ "json_valid": 100.0,
1270
+ "op_correct": 29.4,
1271
+ "exact_match": 17.6
1272
+ },
1273
+ {
1274
+ "domain": "WORKFLOW",
1275
+ "n": 49,
1276
+ "json_valid": 81.6,
1277
+ "op_correct": 0.0,
1278
+ "exact_match": 0.0
1279
+ }
1280
+ ],
1281
+ "tier_distribution": {
1282
+ "1": 40,
1283
+ "2": 40,
1284
+ "3": 40,
1285
+ "4": 40,
1286
+ "5": 40
1287
+ },
1288
+ "domain_distribution": {
1289
+ "CAL": 2,
1290
+ "DB": 27,
1291
+ "ECOMMERCE": 8,
1292
+ "FILE": 4,
1293
+ "HOME": 7,
1294
+ "HTTP": 25,
1295
+ "MQTT": 15,
1296
+ "ROS": 46,
1297
+ "VEHICLE": 17,
1298
+ "WORKFLOW": 49
1299
+ }
1300
+ }