Phi-3-mini-128k-instruct

MERA Created at 22.09.2024 21:11
0.396
The overall result
333
Place in the rating
Weak tasks:
282
RWSD
386
PARus
401
RCB
348
ruEthics
312
MultiQ
459
ruWorldTree
435
ruOpenBookQA
497
CheGeKa
437
ruMMLU
279
ruHateSpeech
477
ruDetox
338
ruHHH
389
ruTiE
50
ruHumanEval
325
USE
312
MathLogicQA
270
ruMultiAr
402
SimpleAr
485
LCS
397
BPS
269
ruModAr
458
MaMuRAMu
61
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.068 Accuracy
RCB 0.479 / 0.403 Accuracy F1 macro
USE 0.114 Grade norm
RWSD 0.531 Accuracy
PARus 0.656 Accuracy
ruTiE 0.558 Accuracy
MultiQ 0.334 / 0.209 F1 Exact match
CheGeKa 0.014 / 0 F1 Exact match
ruModAr 0.473 Exact match
MaMuRAMu 0.444 Accuracy
ruMultiAr 0.256 Exact match
ruCodeEval 0.363 / 0.606 / 0.695 Pass@k
MathLogicQA 0.382 Accuracy
ruWorldTree 0.615 / 0.615 Accuracy F1 macro
ruOpenBookQA 0.565 / 0.565 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 0.789 Accuracy
ruMMLU 0.408 Accuracy
SimpleAr 0.841 Exact match
ruHumanEval 0.384 / 0.624 / 0.689 Pass@k
ruHHH 0.601
ruHateSpeech 0.687
ruDetox 0.078
ruEthics
Correct God Ethical
Virtue 0.206 0.198 0.206
Law 0.177 0.187 0.202
Moral 0.207 0.214 0.186
Justice 0.147 0.15 0.163
Utilitarianism 0.161 0.189 0.182

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-128k-instruct

Model size

3.8B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/pdf/2404.14219

Architecture description:

Phi-3 Mini-128K-Instruct has 3.8B parameters and is a dense decoder-only Transformer model.

Description of the training:

After initial training, the model underwent a post-training process that involved supervised fine-tuning and direct preference optimization to enhance its ability to follow instructions and adhere to safety measures.

Pretrain data:

Trained using the Phi-3 datasets (between May and June 2024). This dataset (4.9T tokens) includes a wide variety of sources, totaling 4.9 trillion tokens, and is a combination of 1. Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code; 2. Newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.); 3. High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.

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 - 131072

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-128k-instruct
MERA
0.267 0.167 0.4 0.167 0 0.067 0 - 0 0.067 0.033 0 0.1 0 0.133 0.267 0.067 0.033 0 0 0.033 0.267 0 0.033 0.1 0.075 0.133 0.267 0.367 0.167 0.167
Model, team Honest Helpful Harmless
Phi-3-mini-128k-instruct
MERA
0.541 0.576 0.69
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-128k-instruct
MERA
0.393 0.398 0.467 0.607 0.474 0.557 0.485 0.463 0.377 0.444 0.281 0.444 0.32 0.509 0.401 0.462 0.61 0.361 0.222 0.389 0.259 0.292 0.33 0.364 0.339 0.4 0.332 0.444 0.596 0.33 0.62 0.653 0.454 0.556 0.396 0.449 0.36 0.423 0.371 0.419 0.455 0.301 0.469 0.469 0.419 0.463 0.417 0.356 0.365 0.39 0.42 0.489 0.423 0.441 0.61 0.467 0.383
Model, team SIM FL STA
Phi-3-mini-128k-instruct
MERA
0.41 0.406 0.604
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-128k-instruct
MERA
0.4 0.366 0.333 0.463 0.342 0.517 0.345 0.368 0.308 0.385 0.564 0.442 0.342 0.411 0.374 0.346 0.589 0.311 0.333 0.561 0.333 0.237 0.489 0.42 0.689 0.394 0.487 0.281 0.684 0.378 0.8 0.397 0.438 0.386 0.364 0.5 0.533 0.6 0.456 0.369 0.42 0.429 0.622 0.6 0.793 0.733 0.259 0.659 0.538 0.614 0.711 0.333 0.557 0.494 0.465 0.269 0.544
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-mini-128k-instruct
MERA
0.206 0.177 0.207 0.147 0.161
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-mini-128k-instruct
MERA
0.198 0.187 0.214 0.15 0.189
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-mini-128k-instruct
MERA
0.206 0.202 0.186 0.163 0.182
Model, team Women Men LGBT Nationalities Migrants Other
Phi-3-mini-128k-instruct
MERA
0.713 0.543 0.706 0.757 0.429 0.705