Mixtral-8x22B-Instruct-v0.1

MERA Created at 20.09.2024 14:21
0.594
The overall result
68
Place in the rating
Weak tasks:
57
RWSD
90
PARus
145
RCB
154
ruEthics
210
MultiQ
121
ruWorldTree
132
ruOpenBookQA
40
CheGeKa
127
ruMMLU
102
ruHateSpeech
60
ruDetox
97
ruHHH
96
ruTiE
41
ruHumanEval
116
USE
147
MathLogicQA
130
ruMultiAr
104
SimpleAr
142
LCS
53
BPS
50
ruModAr
115
MaMuRAMu
38
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.14 Accuracy
RCB 0.555 / 0.512 Accuracy F1 macro
USE 0.283 Grade norm
RWSD 0.654 Accuracy
PARus 0.904 Accuracy
ruTiE 0.815 Accuracy
MultiQ 0.407 / 0.227 F1 Exact match
CheGeKa 0.42 / 0.344 F1 Exact match
ruModAr 0.832 Exact match
MaMuRAMu 0.778 Accuracy
ruMultiAr 0.332 Exact match
ruCodeEval 0.445 / 0.689 / 0.768 Pass@k
MathLogicQA 0.476 Accuracy
ruWorldTree 0.958 / 0.958 Accuracy F1 macro
ruOpenBookQA 0.878 / 0.877 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 0.992 Accuracy
ruMMLU 0.668 Accuracy
SimpleAr 0.989 Exact match
ruHumanEval 0.417 / 0.683 / 0.744 Pass@k
ruHHH 0.837
ruHateSpeech 0.808
ruDetox 0.329
ruEthics
Correct God Ethical
Virtue 0.349 0.422 0.512
Law 0.377 0.445 0.553
Moral 0.4 0.47 0.576
Justice 0.315 0.374 0.478
Utilitarianism 0.296 0.367 0.464

Information about the submission

Mera version
v.1.2.0
Torch Version
2.3.1
The version of the codebase
6ec745e0
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.1
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Mixtral-8x22B-Instruct-v0.1

Model size

140.6B

Model type:

Opened

SFT

MoE

Additional links:

https://mistral.ai/news/mixtral-8x22b/

Architecture description:

The Mixtral-8x22B-Instruct-v0.1 Large Language Model (LLM) is an instruct fine-tuned version of the https://huggingface.co/mistralai/Mixtral-8x22B-v0.1

Description of the training:

This model has been optimised through supervised fine-tuning and direct preference optimisation (DPO) for careful instruction following.

Pretrain data:

The Mixtral-8x22B Large Language Model (LLM) is a pretrained generative Sparse Mixture of Experts.

License:

Apache 2.0

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 - 65536

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
Mixtral-8x22B-Instruct-v0.1
MERA
0.567 0.4 0.833 0.133 0.067 0.467 0.067 - 0 0.033 0.067 0.033 0.3 0.033 0.2 0.367 0.067 0 0.033 0.067 0.1 0.667 0.3 0.267 0.067 0.575 0.1 0.167 0.633 0.4 0.533
Model, team Honest Helpful Harmless
Mixtral-8x22B-Instruct-v0.1
MERA
0.836 0.763 0.914
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
Mixtral-8x22B-Instruct-v0.1
MERA
0.607 0.506 0.783 0.833 0.788 0.851 0.806 0.73 0.787 0.695 0.561 0.595 0.49 0.741 0.803 0.751 0.69 0.833 0.511 0.74 0.369 0.813 0.54 0.636 0.616 0.77 0.508 0.667 0.8 0.5 0.69 0.769 0.699 0.838 0.687 0.671 0.33 0.826 0.47 0.64 0.843 0.721 0.669 0.565 0.841 0.602 0.843 0.415 0.507 0.675 0.61 0.835 0.7 0.777 0.8 0.818 0.86
Model, team SIM FL STA
Mixtral-8x22B-Instruct-v0.1
MERA
0.697 0.703 0.71
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
Mixtral-8x22B-Instruct-v0.1
MERA
0.489 0.871 0.7 0.648 0.816 0.741 0.638 0.772 0.904 0.692 0.821 0.708 0.55 0.798 0.813 0.704 0.748 0.778 0.719 0.877 0.649 0.78 0.844 0.828 0.889 0.818 0.731 0.702 0.947 0.822 0.889 0.872 0.804 0.86 0.621 0.768 0.778 0.822 0.544 0.723 0.873 0.857 0.822 0.756 0.897 0.844 0.914 0.841 0.754 0.912 0.8 0.826 0.785 0.636 0.605 0.778 0.889
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x22B-Instruct-v0.1
MERA
0.349 0.377 0.4 0.315 0.296
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x22B-Instruct-v0.1
MERA
0.422 0.445 0.47 0.374 0.367
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x22B-Instruct-v0.1
MERA
0.512 0.553 0.576 0.478 0.464
Model, team Women Men LGBT Nationalities Migrants Other
Mixtral-8x22B-Instruct-v0.1
MERA
0.815 0.743 0.882 0.811 0.429 0.852