Phi-3-small-8k-instruct

MERA Created at 22.09.2024 21:13
0.465
The overall result
205
Place in the rating
Weak tasks:
323
RWSD
247
PARus
274
RCB
165
ruEthics
238
MultiQ
168
ruWorldTree
198
ruOpenBookQA
291
CheGeKa
213
ruMMLU
84
ruHateSpeech
232
ruDetox
184
ruHHH
190
ruTiE
203
ruHumanEval
226
USE
238
MathLogicQA
260
ruMultiAr
378
SimpleAr
158
LCS
137
BPS
300
ruModAr
261
MaMuRAMu
182
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.136 Accuracy
RCB 0.525 / 0.482 Accuracy F1 macro
USE 0.168 Grade norm
RWSD 0.519 Accuracy
PARus 0.812 Accuracy
ruTiE 0.723 Accuracy
MultiQ 0.386 / 0.206 F1 Exact match
CheGeKa 0.098 / 0.065 F1 Exact match
ruModAr 0.458 Exact match
MaMuRAMu 0.661 Accuracy
ruMultiAr 0.261 Exact match
ruCodeEval 0.056 / 0.176 / 0.238 Pass@k
MathLogicQA 0.423 Accuracy
ruWorldTree 0.941 / 0.941 Accuracy F1 macro
ruOpenBookQA 0.838 / 0.838 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 0.969 Accuracy
ruMMLU 0.605 Accuracy
SimpleAr 0.877 Exact match
ruHumanEval 0.053 / 0.152 / 0.22 Pass@k
ruHHH 0.742
ruHateSpeech 0.823
ruDetox 0.191
ruEthics
Correct God Ethical
Virtue 0.388 0.463 0.537
Law 0.43 0.448 0.562
Moral 0.415 0.462 0.566
Justice 0.343 0.403 0.512
Utilitarianism 0.292 0.359 0.431

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-small-8k-instruct

Model size

7.4B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/pdf/2404.14219

Architecture description:

Phi-3 Small-8K-Instruct has 7B parameters and is a dense decoder-only Transformer model with alternating dense and block sparse attentions.

Description of the training:

The model underwent a post-training process that incorporates both supervised fine-tuning and direct preference optimization for the instruction following and safety measures.

Pretrain data:

The training data includes a wide variety of sources, totaling 4.8 trillion tokens (including 10% multilingual), 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 - 8192

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-small-8k-instruct
MERA
0.367 0.333 0.667 0.167 0 0.233 0.033 - 0.033 0 0 0.033 0.133 0 0.133 0.267 0.067 0 0 0.033 0 0.4 0.1 0.033 0.1 0.292 0.067 0.067 0.333 0.267 0.4
Model, team Honest Helpful Harmless
Phi-3-small-8k-instruct
MERA
0.656 0.729 0.845
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-small-8k-instruct
MERA
0.578 0.458 0.691 0.795 0.683 0.761 0.767 0.637 0.63 0.632 0.43 0.524 0.43 0.713 0.659 0.645 0.61 0.708 0.433 0.725 0.451 0.684 0.39 0.595 0.518 0.63 0.455 0.565 0.678 0.48 0.71 0.81 0.712 0.808 0.619 0.679 0.38 0.748 0.45 0.547 0.798 0.599 0.538 0.546 0.772 0.537 0.701 0.396 0.443 0.598 0.51 0.759 0.687 0.71 0.71 0.745 0.736
Model, team SIM FL STA
Phi-3-small-8k-instruct
MERA
0.438 0.66 0.768
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-small-8k-instruct
MERA
0.444 0.782 0.583 0.583 0.684 0.672 0.552 0.614 0.538 0.662 0.641 0.567 0.5 0.667 0.561 0.593 0.701 0.622 0.544 0.754 0.614 0.763 0.689 0.751 0.733 0.712 0.692 0.561 0.93 0.667 0.822 0.731 0.589 0.789 0.621 0.696 0.689 0.778 0.509 0.692 0.718 0.73 0.667 0.733 0.879 0.867 0.793 0.795 0.677 0.737 0.756 0.58 0.747 0.584 0.581 0.456 0.667
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-small-8k-instruct
MERA
0.388 0.43 0.415 0.343 0.292
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-small-8k-instruct
MERA
0.463 0.448 0.462 0.403 0.359
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-small-8k-instruct
MERA
0.537 0.562 0.566 0.512 0.431
Model, team Women Men LGBT Nationalities Migrants Other
Phi-3-small-8k-instruct
MERA
0.833 0.743 0.941 0.784 0.714 0.852