Phi-3-medium-4k-instruct

MERA Created at 20.09.2024 12:06
0.509
The overall result
139
Place in the rating
Weak tasks:
257
RWSD
110
PARus
296
RCB
316
ruEthics
294
MultiQ
119
ruWorldTree
150
ruOpenBookQA
207
CheGeKa
137
ruMMLU
119
ruHateSpeech
150
ruDetox
135
ruHHH
169
ruTiE
188
ruHumanEval
200
USE
133
MathLogicQA
223
ruMultiAr
389
SimpleAr
257
LCS
102
BPS
112
ruModAr
181
MaMuRAMu
124
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.112 Accuracy
RCB 0.516 / 0.449 Accuracy F1 macro
USE 0.179 Grade norm
RWSD 0.538 Accuracy
PARus 0.898 Accuracy
ruTiE 0.74 Accuracy
MultiQ 0.346 / 0.207 F1 Exact match
CheGeKa 0.148 / 0.115 F1 Exact match
ruModAr 0.668 Exact match
MaMuRAMu 0.719 Accuracy
ruMultiAr 0.279 Exact match
ruCodeEval 0.163 / 0.334 / 0.384 Pass@k
MathLogicQA 0.495 Accuracy
ruWorldTree 0.96 / 0.96 Accuracy F1 macro
ruOpenBookQA 0.868 / 0.867 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 0.979 Accuracy
ruMMLU 0.663 Accuracy
SimpleAr 0.857 Exact match
ruHumanEval 0.074 / 0.184 / 0.256 Pass@k
ruHHH 0.798
ruHateSpeech 0.8
ruDetox 0.233
ruEthics
Correct God Ethical
Virtue 0.209 0.214 0.257
Law 0.198 0.2 0.244
Moral 0.231 0.232 0.279
Justice 0.179 0.185 0.233
Utilitarianism 0.187 0.193 0.222

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-4k-instruct

Model size

14.0B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/abs/2404.14219

Architecture description:

Phi-3-Medium-4k-Instruct has 14B parameters and is a dense decoder-only Transformer model, with context length of 4K 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 - 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-medium-4k-instruct
MERA
0.5 0.233 0.767 0.2 0.033 0.233 0.033 - 0.067 0.133 0 0.067 0.2 0 0.1 0.233 0 0.033 0 0.033 0.167 0.5 0.1 0.033 0.167 0.183 0.1 0.167 0.5 0.3 0.233
Model, team Honest Helpful Harmless
Phi-3-medium-4k-instruct
MERA
0.787 0.746 0.862
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-4k-instruct
MERA
0.578 0.464 0.789 0.816 0.729 0.796 0.786 0.704 0.725 0.673 0.5 0.579 0.35 0.741 0.71 0.682 0.76 0.771 0.5 0.809 0.541 0.754 0.49 0.705 0.563 0.74 0.504 0.648 0.771 0.51 0.74 0.851 0.724 0.818 0.694 0.701 0.45 0.858 0.536 0.645 0.823 0.651 0.538 0.562 0.837 0.63 0.804 0.359 0.535 0.649 0.54 0.819 0.759 0.845 0.73 0.788 0.839
Model, team SIM FL STA
Phi-3-medium-4k-instruct
MERA
0.476 0.7 0.757
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-4k-instruct
MERA
0.556 0.792 0.667 0.611 0.75 0.759 0.621 0.719 0.731 0.6 0.756 0.675 0.533 0.705 0.608 0.605 0.729 0.756 0.702 0.754 0.754 0.797 0.756 0.769 0.756 0.803 0.679 0.614 0.895 0.778 0.822 0.731 0.688 0.825 0.561 0.732 0.6 0.778 0.561 0.708 0.796 0.81 0.778 0.778 0.879 0.867 0.862 0.705 0.723 0.842 0.822 0.71 0.785 0.727 0.651 0.637 0.789
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-medium-4k-instruct
MERA
0.209 0.198 0.231 0.179 0.187
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-medium-4k-instruct
MERA
0.214 0.2 0.232 0.185 0.193
Model, team Virtue Law Moral Justice Utilitarianism
Phi-3-medium-4k-instruct
MERA
0.257 0.244 0.279 0.233 0.222
Model, team Women Men LGBT Nationalities Migrants Other
Phi-3-medium-4k-instruct
MERA
0.87 0.771 0.706 0.757 0.714 0.754