Mistral-7B-Instruct-v0.3

MERA Created at 20.09.2024 11:18
0.434
The overall result
277
Place in the rating
Weak tasks:
68
RWSD
320
PARus
301
RCB
297
ruEthics
330
MultiQ
332
ruWorldTree
331
ruOpenBookQA
197
CheGeKa
348
ruMMLU
242
ruHateSpeech
216
ruDetox
287
ruHHH
308
ruTiE
82
ruHumanEval
363
USE
421
MathLogicQA
368
ruMultiAr
307
SimpleAr
477
LCS
351
BPS
304
ruModAr
321
MaMuRAMu
73
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.07 Accuracy
RCB 0.516 / 0.467 Accuracy F1 macro
USE 0.093 Grade norm
RWSD 0.638 Accuracy
PARus 0.732 Accuracy
ruTiE 0.63 Accuracy
MultiQ 0.322 / 0.143 F1 Exact match
CheGeKa 0.166 / 0.113 F1 Exact match
ruModAr 0.455 Exact match
MaMuRAMu 0.582 Accuracy
ruMultiAr 0.19 Exact match
ruCodeEval 0.292 / 0.5 / 0.573 Pass@k
MathLogicQA 0.332 Accuracy
ruWorldTree 0.777 / 0.776 Accuracy F1 macro
ruOpenBookQA 0.685 / 0.686 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 0.861 Accuracy
ruMMLU 0.471 Accuracy
SimpleAr 0.932 Exact match
ruHumanEval 0.266 / 0.441 / 0.524 Pass@k
ruHHH 0.64
ruHateSpeech 0.728
ruDetox 0.2
ruEthics
Correct God Ethical
Virtue 0.237 0.256 0.312
Law 0.248 0.251 0.309
Moral 0.268 0.262 0.328
Justice 0.196 0.216 0.275
Utilitarianism 0.212 0.219 0.281

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
1 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Mistral-7B-Instruct-v0.3

Model size

7.2B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/abs/2310.06825

Architecture description:

An instruct fine-tuned version of the Mistral-7B-v0.3, which inherits the Mistral-7B-v0.1 architecture with the following differences: No Sliding-Window Attention, Context length expanded to 32k tokens, Vocabulary extended to 32768.

Description of the training:

Fine-tuned on instruction datasets publicly available on the Hugging Face repository. No proprietary data or training tricks were utilized.

Pretrain data:

Open Web data (further details are not disclosed by the authors).

License:

apache-2.0

Inference parameters

Generation Parameters:
chegeka - 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; \nsimplear - do_sample=false;until=["\n"];

The size of the context:
bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, rutie, ruhumaneval, rucodeeval, simplear - 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
Mistral-7B-Instruct-v0.3
MERA
0.1 0.167 0.633 0.167 0 0.133 0 - 0 0 0 0.033 0.133 0 0.067 0.217 0 0 0 0.033 0 0.2 0.1 0.067 0.033 0.017 0.067 0.033 0.4 0.067 0.233
Model, team Honest Helpful Harmless
Mistral-7B-Instruct-v0.3
MERA
0.639 0.695 0.586
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
Mistral-7B-Instruct-v0.3
MERA
0.474 0.38 0.5 0.701 0.52 0.697 0.602 0.54 0.515 0.538 0.316 0.357 0.27 0.574 0.605 0.558 0.57 0.479 0.278 0.504 0.229 0.673 0.26 0.491 0.375 0.53 0.366 0.463 0.6 0.46 0.53 0.686 0.472 0.677 0.506 0.397 0.3 0.574 0.325 0.438 0.621 0.445 0.386 0.387 0.597 0.384 0.637 0.285 0.379 0.439 0.43 0.667 0.415 0.424 0.62 0.679 0.554
Model, team SIM FL STA
Mistral-7B-Instruct-v0.3
MERA
0.699 0.561 0.573
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
Mistral-7B-Instruct-v0.3
MERA
0.422 0.564 0.417 0.546 0.513 0.672 0.517 0.579 0.635 0.554 0.577 0.558 0.358 0.597 0.567 0.531 0.579 0.533 0.316 0.684 0.193 0.695 0.6 0.592 0.711 0.591 0.615 0.614 0.86 0.756 0.689 0.667 0.563 0.702 0.455 0.536 0.511 0.733 0.404 0.538 0.71 0.492 0.8 0.422 0.81 0.711 0.724 0.477 0.585 0.719 0.622 0.594 0.734 0.519 0.419 0.497 0.689
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Mistral-7B-Instruct-v0.3
MERA
0.237 0.248 0.268 0.196 0.212
Model, team Virtue Law Moral Justice Utilitarianism
Mistral-7B-Instruct-v0.3
MERA
0.256 0.251 0.262 0.216 0.219
Model, team Virtue Law Moral Justice Utilitarianism
Mistral-7B-Instruct-v0.3
MERA
0.312 0.309 0.328 0.275 0.281
Model, team Women Men LGBT Nationalities Migrants Other
Mistral-7B-Instruct-v0.3
MERA
0.75 0.714 0.588 0.73 0.429 0.77