Mixtral-8x7B-Instruct-v0.1

MERA Created at 06.10.2024 21:03
0.437
The overall result
258
Place in the rating
Weak tasks:
347
RWSD
260
PARus
94
RCB
203
ruEthics
238
MultiQ
254
ruWorldTree
266
ruOpenBookQA
134
CheGeKa
247
ruMMLU
243
ruHateSpeech
144
ruDetox
244
ruHHH
153
ruTiE
297
USE
339
MathLogicQA
291
ruMultiAr
204
SimpleAr
67
LCS
149
BPS
310
ruModAr
219
MaMuRAMu
+17
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.166 Accuracy
RCB 0.568 / 0.513 Accuracy F1 macro
USE 0.125 Grade norm
RWSD 0.512 Accuracy
PARus 0.796 Accuracy
ruTiE 0.745 Accuracy
MultiQ 0.376 / 0.19 F1 Exact match
CheGeKa 0.212 / 0.137 F1 Exact match
ruModAr 0.444 Exact match
MaMuRAMu 0.696 Accuracy
ruMultiAr 0.245 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.365 Accuracy
ruWorldTree 0.867 / 0.696 Accuracy F1 macro
ruOpenBookQA 0.75 / 0.602 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 0.966 Accuracy
ruMMLU 0.561 Accuracy
SimpleAr 0.974 Exact match
ruHumanEval 0 / 0 / 0 Pass@k
ruHHH 0.674
ruHateSpeech 0.721
ruDetox 0.232
ruEthics
Correct God Ethical
Virtue 0.303 0.348 0.398
Law 0.316 0.339 0.374
Moral 0.343 0.361 0.417
Justice 0.303 0.297 0.346
Utilitarianism 0.262 0.306 0.334

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
9b26db97
CUDA version
12.1
Precision of the model weights
float16
Seed
1234
Batch
1
Transformers version
4.44.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Mixtral-8x7B-Instruct-v0.1

Model size

46.7B

Model type:

Opened

SFT

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, ruhumaneval, rucodeeval, rutie - 32768 \nrutie - 20000

System prompt:
Реши задачу по инструкции ниже. Не давай никаких объяснений и пояснений к своему ответу. Не пиши ничего лишнего. Пиши только то, что указано в инструкции. Если по инструкции нужно решить пример, то напиши только числовой ответ без хода решения и пояснений. Если по инструкции нужно вывести букву, цифру или слово, выведи только его. Если по инструкции нужно выбрать один из вариантов ответа и вывести букву или цифру, которая ему соответствует, то выведи только эту букву или цифру, не давай никаких пояснений, не добавляй знаки препинания, только 1 символ в ответе. Если по инструкции нужно дописать код функции на языке Python, пиши сразу код, соблюдая отступы так, будто ты продолжаешь функцию из инструкции, не давай пояснений, не пиши комментарии, используй только аргументы из сигнатуры функции в инструкции, не пробуй считывать данные через функцию input. Не извиняйся, не строй диалог. Выдавай только ответ и ничего больше.

Description of the template:
{%- if messages[0]['role'] == 'system' %} \n {%- set system_message = messages[0]['content'] %} \n {%- set loop_messages = messages[1:] %} \n{%- else %} \n {%- set loop_messages = messages %} \n{%- endif %} \n \n{{- bos_token }} \n{%- for message in loop_messages %} \n {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) %} \n {{- raise_exception('After the optional system message, conversation roles must alternate user/assistant/user/assistant/...') }} \n {%- endif %} \n {%- if message['role'] == 'user' %} \n {%- if loop.first and system_message is defined %} \n {{- ' [INST] ' + system_message + '\n\n' + message['content'] + ' [/INST]' }} \n {%- else %} \n {{- ' [INST] ' + message['content'] + ' [/INST]' }} \n {%- endif %} \n {%- elif message['role'] == 'assistant' %} \n {{- ' ' + message['content'] + eos_token}} \n {%- else %} \n {{- raise_exception('Only user and assistant roles are supported, with the exception of an initial optional system message!') }} \n {%- endif %} \n{%- endfor %}

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
0.4 0.167 0.733 0.167 0 0.267 0.067 - 0 0 0.033 0.1 0.067 0.133 0.067 0.2 0.067 0 0 0 0.033 0.3 0.133 0.033 0.033 0.183 0 0.067 0.067 0.033 0.167
Model, team Honest Helpful Harmless
Mixtral-8x7B-Instruct-v0.1
MERA
0.639 0.763 0.621
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
0.489 0.494 0.664 0.748 0.637 0.766 0.68 0.605 0.673 0.556 0.43 0.381 0.29 0.694 0.716 0.633 0.63 0.653 0.4 0.71 0.299 0.772 0.29 0.642 0.411 0.66 0.4 0.574 0.71 0.46 0.68 0.777 0.54 0.768 0.589 0.551 0.32 0.719 0.344 0.478 0.737 0.566 0.524 0.488 0.738 0.463 0.794 0.33 0.376 0.551 0.53 0.793 0.521 0.584 0.69 0.733 0.736
Model, team SIM FL STA
Mixtral-8x7B-Instruct-v0.1
MERA
0.537 0.629 0.718
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
0.489 0.733 0.533 0.602 0.763 0.759 0.552 0.702 0.846 0.585 0.731 0.6 0.558 0.667 0.708 0.58 0.673 0.711 0.491 0.825 0.421 0.78 0.622 0.728 0.578 0.742 0.628 0.702 0.93 0.733 0.822 0.744 0.688 0.842 0.682 0.75 0.644 0.8 0.439 0.6 0.8 0.714 0.778 0.689 0.914 0.822 0.862 0.75 0.708 0.842 0.8 0.725 0.772 0.584 0.605 0.637 0.778
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-Instruct-v0.1
MERA
0.303 0.316 0.343 0.303 0.262
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-Instruct-v0.1
MERA
0.348 0.339 0.361 0.297 0.306
Model, team Virtue Law Moral Justice Utilitarianism
Mixtral-8x7B-Instruct-v0.1
MERA
0.398 0.374 0.417 0.346 0.334
Model, team Women Men LGBT Nationalities Migrants Other
Mixtral-8x7B-Instruct-v0.1
MERA
0.722 0.571 0.765 0.757 0.714 0.77