Mixtral-8x7B-v0.1

MERA Created at 20.09.2024 11:39
0.493
The overall result
157
Place in the rating
Weak tasks:
116
RWSD
207
PARus
170
RCB
285
ruEthics
509
MultiQ
266
ruWorldTree
262
ruOpenBookQA
131
CheGeKa
236
ruMMLU
245
ruHateSpeech
402
ruDetox
305
ruHHH
267
ruTiE
86
ruHumanEval
243
USE
330
MathLogicQA
268
ruMultiAr
131
SimpleAr
475
LCS
143
BPS
88
ruModAr
209
MaMuRAMu
85
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.07 Accuracy
RCB 0.55 / 0.497 Accuracy F1 macro
USE 0.162 Grade norm
RWSD 0.592 Accuracy
PARus 0.844 Accuracy
ruTiE 0.665 Accuracy
MultiQ 0.181 / 0.114 F1 Exact match
CheGeKa 0.292 / 0.233 F1 Exact match
ruModAr 0.712 Exact match
MaMuRAMu 0.706 Accuracy
ruMultiAr 0.258 Exact match
ruCodeEval 0.257 / 0.507 / 0.591 Pass@k
MathLogicQA 0.372 Accuracy
ruWorldTree 0.863 / 0.861 Accuracy F1 macro
ruOpenBookQA 0.77 / 0.769 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 0.967 Accuracy
ruMMLU 0.581 Accuracy
SimpleAr 0.985 Exact match
ruHumanEval 0.251 / 0.467 / 0.543 Pass@k
ruHHH 0.624
ruHateSpeech 0.725
ruDetox 0.112
ruEthics
Correct God Ethical
Virtue 0.212 0.317 0.281
Law 0.208 0.321 0.302
Moral 0.2 0.357 0.294
Justice 0.248 0.342 0.285
Utilitarianism 0.211 0.275 0.264

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
float16
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-8x7B-v0.1

Model size

46.7B

Model type:

Opened

Pretrain

Additional links:

https://mistral.ai/news/mixtral-of-experts/

Architecture description:

The Mixtral-8x7B Large Language Model (LLM) is a pretrained generative Sparse Mixture of Experts. The Mistral-8x7B outperforms Llama 2 70B on most benchmarks we tested.

Description of the training:

Mixtral is pre-trained on data extracted from the open Web – we train experts and routers simultaneously.

Pretrain data:

Mixtral is a sparse mixture-of-experts network. It is a decoder-only model where the feedforward block picks from a set of 8 distinct groups of parameters. At every layer, for every token, a router network chooses two of these groups (the “experts”) to process the token and combine their output additively.

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 - 32768

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-8x7B-v0.1
MERA
0.3 0.3 0.733 0.2 0 0.133 0.067 - 0 0.1 0 0 0.2 0.033 0.1 0.233 0.033 0.033 0.033 0 0 0.533 0 0 0.1 0.317 0.2 0.133 0.167 0.133 0.233
Model, team Honest Helpful Harmless
Mixtral-8x7B-v0.1
MERA
0.525 0.593 0.759
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-8x7B-v0.1
MERA
0.556 0.5 0.757 0.791 0.676 0.776 0.68 0.669 0.688 0.628 0.36 0.444 0.35 0.713 0.739 0.645 0.57 0.681 0.456 0.679 0.244 0.819 0.34 0.595 0.411 0.63 0.424 0.556 0.706 0.49 0.73 0.818 0.632 0.808 0.642 0.585 0.44 0.752 0.351 0.527 0.773 0.636 0.566 0.44 0.715 0.468 0.784 0.319 0.415 0.605 0.48 0.781 0.59 0.609 0.73 0.752 0.767
Model, team SIM FL STA
Mixtral-8x7B-v0.1
MERA
0.851 0.588 0.246
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-8x7B-v0.1
MERA
0.511 0.752 0.617 0.602 0.737 0.793 0.586 0.737 0.827 0.646 0.731 0.575 0.525 0.674 0.702 0.63 0.673 0.733 0.632 0.877 0.368 0.831 0.711 0.698 0.756 0.712 0.641 0.614 0.947 0.667 0.822 0.756 0.679 0.86 0.591 0.696 0.711 0.778 0.561 0.708 0.804 0.714 0.844 0.644 0.897 0.8 0.897 0.795 0.662 0.842 0.8 0.768 0.759 0.61 0.581 0.661 0.833
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-v0.1
MERA
0.212 0.208 0.2 0.248 0.211
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-v0.1
MERA
0.317 0.321 0.357 0.342 0.275
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-v0.1
MERA
0.281 0.302 0.294 0.285 0.264
Model, team Women Men LGBT Nationalities Migrants Other
Mixtral-8x7B-v0.1
MERA
0.713 0.686 0.706 0.703 0.571 0.803