Phi-3-mini-4k-instruct

MERA Created at 22.09.2024 21:12
0.399
The overall result
327
Place in the rating
Weak tasks:
162
RWSD
376
PARus
354
RCB
367
ruEthics
411
MultiQ
441
ruWorldTree
429
ruOpenBookQA
549
CheGeKa
436
ruMMLU
340
ruHateSpeech
450
ruDetox
376
ruHHH
366
ruTiE
52
ruHumanEval
365
USE
300
MathLogicQA
250
ruMultiAr
376
SimpleAr
593
LCS
304
BPS
247
ruModAr
460
MaMuRAMu
48
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.024 Accuracy
RCB 0.5 / 0.418 Accuracy F1 macro
USE 0.093 Grade norm
RWSD 0.573 Accuracy
PARus 0.666 Accuracy
ruTiE 0.577 Accuracy
MultiQ 0.259 / 0.164 F1 Exact match
CheGeKa 0.008 / 0 F1 Exact match
ruModAr 0.481 Exact match
MaMuRAMu 0.44 Accuracy
ruMultiAr 0.268 Exact match
ruCodeEval 0.396 / 0.645 / 0.72 Pass@k
MathLogicQA 0.393 Accuracy
ruWorldTree 0.636 / 0.637 Accuracy F1 macro
ruOpenBookQA 0.573 / 0.571 Accuracy F1 macro

Evaluation on open tasks:

Go to the ratings by subcategory

The table will scroll to the left

Task name Result Metric
BPS 0.909 Accuracy
ruMMLU 0.408 Accuracy
SimpleAr 0.879 Exact match
ruHumanEval 0.376 / 0.617 / 0.701 Pass@k
ruHHH 0.567
ruHateSpeech 0.638
ruDetox 0.089
ruEthics
Correct God Ethical
Virtue 0.179 0.184 0.193
Law 0.171 0.187 0.183
Moral 0.19 0.199 0.195
Justice 0.13 0.14 0.131
Utilitarianism 0.162 0.178 0.151

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
6ec745e0
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Phi-3-mini-4k-instruct

Model size

3.8B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/abs/2404.14219

Architecture description:

The model belongs to the Phi-3 family and has context length of 4K tokens.

Description of the training:

Pre-training is performed in two disjoint and sequential phases; phase-1 comprises mostly of web sources aimed at teaching the model general knowledge and language understanding. Phase-2 merges even more heavily filtered webdata (a subset used in Phase-1) with some synthetic data that teach the model logical reasoning and various niche skills. Post-training of phi-3 went through two stages, including supervised finetuning (SFT) and direct preference optimization (DPO).

Pretrain data:

The training data consists of heavily filtered publicly available web data (according to the “educational level”) from various open internet sources, as well as synthetic LLM-generated data.

License:

MIT

Inference parameters

Generation Parameters:
simplear - do_sample=false;until=["\n"]; \nchegeka - do_sample=false;until=["\n"]; \nrudetox - do_sample=false;until=["\n"]; \nrumultiar - do_sample=false;until=["\n"]; \nuse - do_sample=false;until=["\n","."]; \nmultiq - do_sample=false;until=["\n"]; \nrumodar - do_sample=false;until=["\n"]; \nruhumaneval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6; \nrucodeeval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6;

The size of the context:
simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, rutie, ruhumaneval, rucodeeval - 4096

Ratings by subcategory

Metric: Grade Norm
Model, team 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 8_0 8_1 8_2 8_3 8_4
Phi-3-mini-4k-instruct
MERA
0.133 0.167 0.5 0.167 0 0.067 0 - 0.033 0.067 0.033 0 0.067 0 0.167 0.267 0.033 0.033 0 0.033 0 0.067 0 0.033 0.033 0.033 0.1 0.2 0.267 0.2 0.1
Model, team Honest Helpful Harmless
Phi-3-mini-4k-instruct
MERA
0.508 0.542 0.655
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Management Philosophy Prehistory Human aging Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Biology (college) Physics (college) Human Sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine learning Medical genetics Professional law PR Security studies Chemistry (школьная) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual_physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) European history Government and politics
Phi-3-mini-4k-instruct
MERA
0.333 0.416 0.474 0.607 0.477 0.582 0.515 0.46 0.404 0.417 0.263 0.429 0.34 0.491 0.392 0.454 0.54 0.375 0.256 0.397 0.278 0.251 0.34 0.37 0.33 0.43 0.336 0.435 0.58 0.37 0.57 0.669 0.436 0.566 0.423 0.397 0.39 0.435 0.411 0.414 0.465 0.32 0.434 0.454 0.434 0.477 0.382 0.326 0.358 0.382 0.41 0.489 0.405 0.445 0.61 0.491 0.415
Model, team SIM FL STA
Phi-3-mini-4k-instruct
MERA
0.501 0.441 0.549
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Managment Philosophy Pre-History Gerontology Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Bilology (college) Physics (college) Human sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine Learning Genetics Professional law PR Security Chemistry (college) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical Engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional Accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) Europe History Government and politics
Phi-3-mini-4k-instruct
MERA
0.356 0.327 0.217 0.472 0.355 0.534 0.414 0.368 0.25 0.477 0.603 0.45 0.317 0.38 0.345 0.309 0.617 0.333 0.421 0.491 0.281 0.271 0.378 0.402 0.689 0.394 0.436 0.228 0.702 0.422 0.756 0.359 0.464 0.456 0.348 0.5 0.489 0.489 0.456 0.4 0.388 0.397 0.711 0.533 0.862 0.778 0.448 0.682 0.477 0.667 0.644 0.319 0.582 0.506 0.488 0.251 0.589
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-mini-4k-instruct
MERA
0.179 0.171 0.19 0.13 0.162
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-mini-4k-instruct
MERA
0.184 0.187 0.199 0.14 0.178
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-mini-4k-instruct
MERA
0.193 0.183 0.195 0.131 0.151
Model, team Women Men LGBT Nationalities Migrants Other
Phi-3-mini-4k-instruct
MERA
0.648 0.571 0.529 0.649 0.429 0.705