Phi-3-medium-128k-instruct

MERA Created at 20.09.2024 13:01
0.509
The overall result
142
Place in the rating
Weak tasks:
159
RWSD
115
PARus
193
RCB
261
ruEthics
208
MultiQ
131
ruWorldTree
172
ruOpenBookQA
209
CheGeKa
150
ruMMLU
93
ruHateSpeech
138
ruDetox
164
ruHHH
212
ruTiE
166
ruHumanEval
168
USE
184
MathLogicQA
208
ruMultiAr
360
SimpleAr
376
LCS
239
BPS
138
ruModAr
183
MaMuRAMu
115
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.09 Accuracy
RCB 0.546 / 0.458 Accuracy F1 macro
USE 0.206 Grade norm
RWSD 0.573 Accuracy
PARus 0.894 Accuracy
ruTiE 0.705 Accuracy
MultiQ 0.409 / 0.217 F1 Exact match
CheGeKa 0.148 / 0.111 F1 Exact match
ruModAr 0.605 Exact match
MaMuRAMu 0.719 Accuracy
ruMultiAr 0.287 Exact match
ruCodeEval 0.177 / 0.376 / 0.463 Pass@k
MathLogicQA 0.462 Accuracy
ruWorldTree 0.956 / 0.956 Accuracy F1 macro
ruOpenBookQA 0.853 / 0.853 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 0.943 Accuracy
ruMMLU 0.654 Accuracy
SimpleAr 0.897 Exact match
ruHumanEval 0.097 / 0.23 / 0.305 Pass@k
ruHHH 0.758
ruHateSpeech 0.815
ruDetox 0.247
ruEthics
Correct God Ethical
Virtue 0.3 0.257 0.322
Law 0.278 0.253 0.296
Moral 0.324 0.276 0.336
Justice 0.245 0.214 0.273
Utilitarianism 0.272 0.241 0.268

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
2 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Phi-3-medium-128k-instruct

Model size

14.0B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/abs/2404.14219

Architecture description:

Phi-3-Medium-128k-Instruct has 14B parameters and is a dense decoder-only Transformer model, with context length of 128K tokens.

Description of the training:

The model is fine-tuned with Supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) to ensure alignment with human preferences and safety guidlines. Trained between February and April 2024.

Pretrain data:

The training data (with cutoff date October 2023) 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 - 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-medium-128k-instruct
MERA
0.5 0.233 0.8 0.1 0.1 0.267 0.067 - 0 0.067 0.1 0.033 0.233 0 0.1 0.35 0.033 0 0.033 0.033 0.067 0.533 0.1 0.1 0.067 0.292 0.067 0.133 0.533 0.4 0.433
Model, team Honest Helpful Harmless
Phi-3-medium-128k-instruct
MERA
0.705 0.78 0.793
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-medium-128k-instruct
MERA
0.6 0.44 0.809 0.821 0.752 0.786 0.777 0.682 0.691 0.646 0.5 0.548 0.31 0.75 0.702 0.699 0.74 0.771 0.478 0.779 0.502 0.772 0.46 0.665 0.536 0.75 0.495 0.63 0.784 0.53 0.73 0.835 0.767 0.838 0.694 0.671 0.47 0.835 0.47 0.635 0.793 0.643 0.586 0.546 0.828 0.634 0.794 0.389 0.518 0.646 0.53 0.81 0.756 0.819 0.77 0.745 0.839
Model, team SIM FL STA
Phi-3-medium-128k-instruct
MERA
0.587 0.693 0.675
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-medium-128k-instruct
MERA
0.644 0.752 0.683 0.593 0.789 0.724 0.638 0.719 0.731 0.631 0.705 0.65 0.55 0.636 0.602 0.605 0.729 0.778 0.649 0.789 0.772 0.78 0.711 0.751 0.778 0.848 0.718 0.614 0.895 0.756 0.778 0.782 0.696 0.807 0.606 0.696 0.6 0.778 0.544 0.692 0.808 0.825 0.8 0.756 0.879 0.889 0.862 0.795 0.738 0.842 0.844 0.768 0.759 0.74 0.698 0.608 0.778
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-medium-128k-instruct
MERA
0.3 0.278 0.324 0.245 0.272
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-medium-128k-instruct
MERA
0.257 0.253 0.276 0.214 0.241
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-medium-128k-instruct
MERA
0.322 0.296 0.336 0.273 0.268
Model, team Women Men LGBT Nationalities Migrants Other
Phi-3-medium-128k-instruct
MERA
0.843 0.714 0.824 0.757 0.857 0.852