Mixtral-8x7B-Instruct-v0.1

MERA Created at 20.09.2024 13:56
0.515
The overall result
130
Place in the rating
Weak tasks:
212
RWSD
222
PARus
169
RCB
85
ruEthics
260
MultiQ
251
ruWorldTree
233
ruOpenBookQA
121
CheGeKa
233
ruMMLU
140
ruHateSpeech
109
ruDetox
158
ruHHH
183
ruTiE
57
ruHumanEval
247
USE
359
MathLogicQA
264
ruMultiAr
197
SimpleAr
428
LCS
130
BPS
81
ruModAr
203
MaMuRAMu
50
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.082 Accuracy
RCB 0.55 / 0.523 Accuracy F1 macro
USE 0.161 Grade norm
RWSD 0.554 Accuracy
PARus 0.832 Accuracy
ruTiE 0.727 Accuracy
MultiQ 0.361 / 0.162 F1 Exact match
CheGeKa 0.308 / 0.248 F1 Exact match
ruModAr 0.722 Exact match
MaMuRAMu 0.711 Accuracy
ruMultiAr 0.259 Exact match
ruCodeEval 0.393 / 0.614 / 0.683 Pass@k
MathLogicQA 0.36 Accuracy
ruWorldTree 0.876 / 0.875 Accuracy F1 macro
ruOpenBookQA 0.803 / 0.802 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 0.97 Accuracy
ruMMLU 0.582 Accuracy
SimpleAr 0.976 Exact match
ruHumanEval 0.359 / 0.604 / 0.695 Pass@k
ruHHH 0.764
ruHateSpeech 0.789
ruDetox 0.285
ruEthics
Correct God Ethical
Virtue 0.379 0.396 0.432
Law 0.382 0.391 0.408
Moral 0.409 0.424 0.456
Justice 0.359 0.366 0.412
Utilitarianism 0.337 0.358 0.373

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-Instruct-v0.1

Model size

46.7B

Model type:

Opened

SFT

MoE

Additional links:

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

Architecture description:

Mixtral 8x7B Instruct is the instructed version of Mixtral 8x7B. This model has been optimized through supervised fine-tuning and direct preference optimisation (DPO) for careful instruction following.

Description of the training:

Mixtral is pre-trained on data extracted from the open Web – we train experts and routers simultaneously. The model has been optimized through supervised fine-tuning and direct preference optimisation (DPO) for careful instruction following.

Pretrain data:

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.

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-Instruct-v0.1
MERA
0.467 0.333 0.8 0.1 0 0.3 0.1 - 0 0 0 0 0.2 0 0.067 0.2 0.033 0.033 0 0 0.033 0.5 0.067 0.1 0 0.258 0.2 0.1 0.233 0.033 0.333
Model, team Honest Helpful Harmless
Mixtral-8x7B-Instruct-v0.1
MERA
0.721 0.712 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
Mixtral-8x7B-Instruct-v0.1
MERA
0.511 0.506 0.671 0.774 0.686 0.791 0.65 0.611 0.685 0.592 0.421 0.421 0.33 0.713 0.729 0.668 0.6 0.688 0.433 0.74 0.299 0.795 0.3 0.63 0.473 0.65 0.436 0.62 0.71 0.45 0.7 0.818 0.607 0.818 0.604 0.56 0.36 0.713 0.417 0.512 0.773 0.662 0.559 0.456 0.728 0.532 0.794 0.367 0.418 0.557 0.54 0.781 0.59 0.601 0.69 0.752 0.736
Model, team SIM FL STA
Mixtral-8x7B-Instruct-v0.1
MERA
0.621 0.677 0.729
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-Instruct-v0.1
MERA
0.511 0.743 0.633 0.611 0.737 0.793 0.603 0.684 0.827 0.646 0.731 0.6 0.508 0.713 0.719 0.654 0.71 0.711 0.579 0.772 0.526 0.797 0.689 0.751 0.733 0.697 0.667 0.632 0.965 0.778 0.822 0.744 0.688 0.842 0.515 0.714 0.689 0.822 0.509 0.662 0.796 0.714 0.822 0.711 0.931 0.822 0.897 0.773 0.677 0.877 0.8 0.797 0.797 0.636 0.512 0.643 0.822
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-Instruct-v0.1
MERA
0.379 0.382 0.409 0.359 0.337
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-Instruct-v0.1
MERA
0.396 0.391 0.424 0.366 0.358
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-Instruct-v0.1
MERA
0.432 0.408 0.456 0.412 0.373
Model, team Women Men LGBT Nationalities Migrants Other
Mixtral-8x7B-Instruct-v0.1
MERA
0.769 0.686 0.824 0.811 0.571 0.885