devMartianLabs commited on
Commit
1ff2bbc
·
verified ·
1 Parent(s): ffaa051

Add head-to-head multi-model benchmark results and scripts (multi_model_head_to_head_benchmark.json)

Browse files
multi_model_head_to_head_benchmark.json ADDED
@@ -0,0 +1,368 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "timestamp_utc": "2026-11-08T11:31:53Z",
3
+ "elapsed_sec": 10.31,
4
+ "leaderboard": [
5
+ {
6
+ "model_id": "marsnext/Phobos-1 (Final Post-Trained)",
7
+ "org": "MarsNext",
8
+ "total_params_m": 96.26,
9
+ "active_params_m": 52.22,
10
+ "train_tokens": "1.56B",
11
+ "mc20_acc_pct": 35.0,
12
+ "mc20_correct": "7/20",
13
+ "gen25_score_pct": 80.0,
14
+ "gen25_passed": "20/25",
15
+ "python_pass_at_1": "3/5 (60%)",
16
+ "sql_synthesis": "4/5 (80%)",
17
+ "stem_qa": "4/5 (80%)",
18
+ "math_symbolic": "4/5 (80%)",
19
+ "think_planning": "5/5 (100%)",
20
+ "combined_avg_pct": 57.5,
21
+ "speed_tok_per_s": 17.49
22
+ },
23
+ {
24
+ "model_id": "marsnext/Phobos-1 (Base Step 33,977)",
25
+ "org": "MarsNext",
26
+ "total_params_m": 96.26,
27
+ "active_params_m": 52.22,
28
+ "train_tokens": "1.56B",
29
+ "mc20_acc_pct": 35.0,
30
+ "mc20_correct": "7/20",
31
+ "gen25_score_pct": 68.0,
32
+ "gen25_passed": "17/25",
33
+ "python_pass_at_1": "3/5 (60%)",
34
+ "sql_synthesis": "4/5 (80%)",
35
+ "stem_qa": "3/5 (60%)",
36
+ "math_symbolic": "2/5 (40%)",
37
+ "think_planning": "5/5 (100%)",
38
+ "combined_avg_pct": 51.5,
39
+ "speed_tok_per_s": 18.52
40
+ },
41
+ {
42
+ "model_id": "EleutherAI/pythia-70m-deduped",
43
+ "org": "EleutherAI",
44
+ "total_params_m": 70.43,
45
+ "active_params_m": 70.43,
46
+ "train_tokens": "300B",
47
+ "mc20_acc_pct": 30.0,
48
+ "mc20_correct": "6/20",
49
+ "gen25_score_pct": 12.0,
50
+ "gen25_passed": "3/25",
51
+ "python_pass_at_1": "0/5 (0%)",
52
+ "sql_synthesis": "0/5 (0%)",
53
+ "stem_qa": "1/5 (20%)",
54
+ "math_symbolic": "1/5 (20%)",
55
+ "think_planning": "1/5 (20%)",
56
+ "combined_avg_pct": 21.0,
57
+ "speed_tok_per_s": 101.6
58
+ },
59
+ {
60
+ "model_id": "openai-community/gpt2",
61
+ "org": "OpenAI",
62
+ "total_params_m": 124.44,
63
+ "active_params_m": 124.44,
64
+ "train_tokens": "40B",
65
+ "mc20_acc_pct": 40.0,
66
+ "mc20_correct": "8/20",
67
+ "gen25_score_pct": 20.0,
68
+ "gen25_passed": "5/25",
69
+ "python_pass_at_1": "0/5 (0%)",
70
+ "sql_synthesis": "1/5 (20%)",
71
+ "stem_qa": "1/5 (20%)",
72
+ "math_symbolic": "1/5 (20%)",
73
+ "think_planning": "2/5 (40%)",
74
+ "combined_avg_pct": 30.0,
75
+ "speed_tok_per_s": 70.3
76
+ },
77
+ {
78
+ "model_id": "EleutherAI/pythia-160m-deduped",
79
+ "org": "EleutherAI",
80
+ "total_params_m": 162.32,
81
+ "active_params_m": 162.32,
82
+ "train_tokens": "300B",
83
+ "mc20_acc_pct": 25.0,
84
+ "mc20_correct": "5/20",
85
+ "gen25_score_pct": 12.0,
86
+ "gen25_passed": "3/25",
87
+ "python_pass_at_1": "0/5 (0%)",
88
+ "sql_synthesis": "0/5 (0%)",
89
+ "stem_qa": "1/5 (20%)",
90
+ "math_symbolic": "0/5 (0%)",
91
+ "think_planning": "2/5 (40%)",
92
+ "combined_avg_pct": 18.5,
93
+ "speed_tok_per_s": 67.1
94
+ },
95
+ {
96
+ "model_id": "HuggingFaceTB/SmolLM2-135M",
97
+ "org": "HuggingFace",
98
+ "total_params_m": 134.52,
99
+ "active_params_m": 134.52,
100
+ "train_tokens": "2,000B",
101
+ "mc20_acc_pct": 60.0,
102
+ "mc20_correct": "12/20",
103
+ "gen25_score_pct": 36.0,
104
+ "gen25_passed": "9/25",
105
+ "python_pass_at_1": "0/5 (0%)",
106
+ "sql_synthesis": "2/5 (40%)",
107
+ "stem_qa": "4/5 (80%)",
108
+ "math_symbolic": "1/5 (20%)",
109
+ "think_planning": "2/5 (40%)",
110
+ "combined_avg_pct": 48.0,
111
+ "speed_tok_per_s": 25.8
112
+ }
113
+ ],
114
+ "phobos_mc_final": {
115
+ "correct": 7,
116
+ "total": 20,
117
+ "acc_pct": 35.0,
118
+ "details": [
119
+ {
120
+ "id": "ARC-01",
121
+ "pred": 0,
122
+ "ans": 1,
123
+ "correct": false
124
+ },
125
+ {
126
+ "id": "ARC-02",
127
+ "pred": 2,
128
+ "ans": 2,
129
+ "correct": true
130
+ },
131
+ {
132
+ "id": "ARC-03",
133
+ "pred": 0,
134
+ "ans": 0,
135
+ "correct": true
136
+ },
137
+ {
138
+ "id": "ARC-04",
139
+ "pred": 1,
140
+ "ans": 1,
141
+ "correct": true
142
+ },
143
+ {
144
+ "id": "ARC-05",
145
+ "pred": 1,
146
+ "ans": 2,
147
+ "correct": false
148
+ },
149
+ {
150
+ "id": "SCIQ-01",
151
+ "pred": 1,
152
+ "ans": 2,
153
+ "correct": false
154
+ },
155
+ {
156
+ "id": "SCIQ-02",
157
+ "pred": 1,
158
+ "ans": 1,
159
+ "correct": true
160
+ },
161
+ {
162
+ "id": "SCIQ-03",
163
+ "pred": 1,
164
+ "ans": 2,
165
+ "correct": false
166
+ },
167
+ {
168
+ "id": "SCIQ-04",
169
+ "pred": 2,
170
+ "ans": 1,
171
+ "correct": false
172
+ },
173
+ {
174
+ "id": "SCIQ-05",
175
+ "pred": 2,
176
+ "ans": 2,
177
+ "correct": true
178
+ },
179
+ {
180
+ "id": "PIQA-01",
181
+ "pred": 1,
182
+ "ans": 1,
183
+ "correct": true
184
+ },
185
+ {
186
+ "id": "PIQA-02",
187
+ "pred": 0,
188
+ "ans": 1,
189
+ "correct": false
190
+ },
191
+ {
192
+ "id": "PIQA-03",
193
+ "pred": 3,
194
+ "ans": 1,
195
+ "correct": false
196
+ },
197
+ {
198
+ "id": "CS-01",
199
+ "pred": 2,
200
+ "ans": 1,
201
+ "correct": false
202
+ },
203
+ {
204
+ "id": "CS-02",
205
+ "pred": 0,
206
+ "ans": 2,
207
+ "correct": false
208
+ },
209
+ {
210
+ "id": "CS-03",
211
+ "pred": 2,
212
+ "ans": 1,
213
+ "correct": false
214
+ },
215
+ {
216
+ "id": "CS-04",
217
+ "pred": 1,
218
+ "ans": 1,
219
+ "correct": true
220
+ },
221
+ {
222
+ "id": "MATH-MC1",
223
+ "pred": 0,
224
+ "ans": 1,
225
+ "correct": false
226
+ },
227
+ {
228
+ "id": "MATH-MC2",
229
+ "pred": 2,
230
+ "ans": 1,
231
+ "correct": false
232
+ },
233
+ {
234
+ "id": "MATH-MC3",
235
+ "pred": 3,
236
+ "ans": 0,
237
+ "correct": false
238
+ }
239
+ ]
240
+ },
241
+ "phobos_mc_base": {
242
+ "correct": 7,
243
+ "total": 20,
244
+ "acc_pct": 35.0,
245
+ "details": [
246
+ {
247
+ "id": "ARC-01",
248
+ "pred": 0,
249
+ "ans": 1,
250
+ "correct": false
251
+ },
252
+ {
253
+ "id": "ARC-02",
254
+ "pred": 2,
255
+ "ans": 2,
256
+ "correct": true
257
+ },
258
+ {
259
+ "id": "ARC-03",
260
+ "pred": 0,
261
+ "ans": 0,
262
+ "correct": true
263
+ },
264
+ {
265
+ "id": "ARC-04",
266
+ "pred": 1,
267
+ "ans": 1,
268
+ "correct": true
269
+ },
270
+ {
271
+ "id": "ARC-05",
272
+ "pred": 1,
273
+ "ans": 2,
274
+ "correct": false
275
+ },
276
+ {
277
+ "id": "SCIQ-01",
278
+ "pred": 1,
279
+ "ans": 2,
280
+ "correct": false
281
+ },
282
+ {
283
+ "id": "SCIQ-02",
284
+ "pred": 1,
285
+ "ans": 1,
286
+ "correct": true
287
+ },
288
+ {
289
+ "id": "SCIQ-03",
290
+ "pred": 1,
291
+ "ans": 2,
292
+ "correct": false
293
+ },
294
+ {
295
+ "id": "SCIQ-04",
296
+ "pred": 2,
297
+ "ans": 1,
298
+ "correct": false
299
+ },
300
+ {
301
+ "id": "SCIQ-05",
302
+ "pred": 2,
303
+ "ans": 2,
304
+ "correct": true
305
+ },
306
+ {
307
+ "id": "PIQA-01",
308
+ "pred": 1,
309
+ "ans": 1,
310
+ "correct": true
311
+ },
312
+ {
313
+ "id": "PIQA-02",
314
+ "pred": 0,
315
+ "ans": 1,
316
+ "correct": false
317
+ },
318
+ {
319
+ "id": "PIQA-03",
320
+ "pred": 3,
321
+ "ans": 1,
322
+ "correct": false
323
+ },
324
+ {
325
+ "id": "CS-01",
326
+ "pred": 2,
327
+ "ans": 1,
328
+ "correct": false
329
+ },
330
+ {
331
+ "id": "CS-02",
332
+ "pred": 0,
333
+ "ans": 2,
334
+ "correct": false
335
+ },
336
+ {
337
+ "id": "CS-03",
338
+ "pred": 2,
339
+ "ans": 1,
340
+ "correct": false
341
+ },
342
+ {
343
+ "id": "CS-04",
344
+ "pred": 1,
345
+ "ans": 1,
346
+ "correct": true
347
+ },
348
+ {
349
+ "id": "MATH-MC1",
350
+ "pred": 2,
351
+ "ans": 1,
352
+ "correct": false
353
+ },
354
+ {
355
+ "id": "MATH-MC2",
356
+ "pred": 2,
357
+ "ans": 1,
358
+ "correct": false
359
+ },
360
+ {
361
+ "id": "MATH-MC3",
362
+ "pred": 3,
363
+ "ans": 0,
364
+ "correct": false
365
+ }
366
+ ]
367
+ }
368
+ }