Qwen2.5 14B Instruct

НГУ Created at 16.10.2024 14:20
0.528
The overall result
118
Place in the rating
Weak tasks:
137
RWSD
123
PARus
235
RCB
41
ruEthics
103
MultiQ
129
ruWorldTree
147
ruOpenBookQA
201
CheGeKa
119
ruMMLU
71
ruHateSpeech
55
ruDetox
98
ruHHH
98
ruTiE
208
ruHumanEval
104
USE
98
MathLogicQA
104
ruMultiAr
246
SimpleAr
155
LCS
136
BPS
129
ruModAr
126
MaMuRAMu
218
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.138 Accuracy
RCB 0.534 / 0.502 Accuracy F1 macro
USE 0.297 Grade norm
RWSD 0.581 Accuracy
PARus 0.89 Accuracy
ruTiE 0.814 Accuracy
MultiQ 0.519 / 0.381 F1 Exact match
CheGeKa 0.159 / 0.111 F1 Exact match
ruModAr 0.631 Exact match
MaMuRAMu 0.765 Accuracy
ruMultiAr 0.35 Exact match
ruCodeEval 0.021 / 0.038 / 0.043 Pass@k
MathLogicQA 0.58 Accuracy
ruWorldTree 0.956 / 0.956 Accuracy F1 macro
ruOpenBookQA 0.87 / 0.697 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 0.969 Accuracy
ruMMLU 0.682 Accuracy
SimpleAr 0.961 Exact match
ruHumanEval 0.049 / 0.082 / 0.091 Pass@k
ruHHH 0.837
ruHateSpeech 0.834
ruDetox 0.331
ruEthics
Correct God Ethical
Virtue 0.436 0.448 0.453
Law 0.427 0.422 0.438
Moral 0.469 0.478 0.473
Justice 0.371 0.394 0.398
Utilitarianism 0.364 0.395 0.428

Information about the submission

Mera version
v.1.2.0
Torch Version
2.3.1
The version of the codebase
430295f
CUDA version
11.8
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.40.0
The number of GPUs and their type
1 x NVIDIA A100 80GB PCIe
Architecture
hf

Team:

НГУ

Name of the ML model:

Qwen2.5 14B Instruct

Model size

14.8B

Model type:

Opened

SFT

Architecture description:

Qwen2.5 14B Instruct is a language model including decoder of 14B size. It is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, etc. Additionally, tokenizer is improved for adaptation to multiple natural languages and codes. Qwen2.5 brings some improvements upon Qwen2, including larger context length: full 131,072 tokens and generation 8192 tokens.

Description of the training:

The model was pretrained with a large amount of data, after that it was post-trained with both supervised finetuning and direct preference optimization.

Pretrain data:

The model was pretrained with a large amount of data of English, Chinese and over 27 additional languages including Russian. In terms of the context length, the model was pretrained on data with different context length which was augmented from 4,096 tokens to 32,768 tokens during the concluding phase of pre-training.

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, chegeka, rudetox, rumultiar, use, multiq, rumodar, ruhumaneval, rucodeeval - 32768

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 tools %} \n {{- '<|im_start|>system\n' }} \n {%- if messages[0]['role'] == 'system' %} \n {{- messages[0]['content'] }} \n {%- else %} \n {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }} \n {%- endif %} \n {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }} \n {%- for tool in tools %} \n {{- "\n" }} \n {{- tool | tojson }} \n {%- endfor %} \n {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }} \n{%- else %} \n {%- if messages[0]['role'] == 'system' %} \n {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }} \n {%- else %} \n {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }} \n {%- endif %} \n{%- endif %} \n{%- for message in messages %} \n {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %} \n {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }} \n {%- elif message.role == "assistant" %} \n {{- '<|im_start|>' + message.role }} \n {%- if message.content %} \n {{- '\n' + message.content }} \n {%- endif %} \n {%- for tool_call in message.tool_calls %} \n {%- if tool_call.function is defined %} \n {%- set tool_call = tool_call.function %} \n {%- endif %} \n {{- '\n<tool_call>\n{"name": "' }} \n {{- tool_call.name }} \n {{- '", "arguments": ' }} \n {{- tool_call.arguments | tojson }} \n {{- '}\n</tool_call>' }} \n {%- endfor %} \n {{- '<|im_end|>\n' }} \n {%- elif message.role == "tool" %} \n {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %} \n {{- '<|im_start|>user' }} \n {%- endif %} \n {{- '\n<tool_response>\n' }} \n {{- message.content }} \n {{- '\n</tool_response>' }} \n {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %} \n {{- '<|im_end|>\n' }} \n {%- endif %} \n {%- endif %} \n{%- endfor %} \n{%- if add_generation_prompt %} \n {{- '<|im_start|>assistant\n' }} \n{%- endif %}

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
Qwen2.5 14B Instruct
НГУ
0.5 0.5 0.767 0.133 0.133 0.267 0.067 - 0 0.2 0.067 0.033 0.233 0.067 0.2 0.383 0.033 0.033 0.033 0.033 0.167 0.767 0.467 0.233 0.233 0.633 0.167 0.367 0.467 0.233 0.4
Model, team Honest Helpful Harmless
Qwen2.5 14B Instruct
НГУ
0.836 0.814 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
Qwen2.5 14B Instruct
НГУ
0.615 0.53 0.849 0.829 0.739 0.841 0.816 0.723 0.775 0.691 0.535 0.571 0.49 0.713 0.784 0.688 0.71 0.819 0.522 0.763 0.441 0.836 0.5 0.665 0.491 0.75 0.481 0.602 0.763 0.49 0.72 0.785 0.699 0.818 0.709 0.782 0.51 0.826 0.609 0.635 0.818 0.746 0.662 0.724 0.846 0.681 0.863 0.526 0.535 0.684 0.66 0.831 0.785 0.819 0.86 0.824 0.86
Model, team SIM FL STA
Qwen2.5 14B Instruct
НГУ
0.71 0.676 0.712
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
Qwen2.5 14B Instruct
НГУ
0.556 0.832 0.683 0.685 0.816 0.759 0.655 0.702 0.827 0.723 0.833 0.725 0.517 0.705 0.725 0.642 0.71 0.8 0.649 0.807 0.772 0.864 0.867 0.817 0.822 0.803 0.808 0.702 0.895 0.8 0.844 0.833 0.741 0.895 0.727 0.768 0.889 0.778 0.702 0.662 0.841 0.794 0.8 0.956 0.897 0.889 0.828 0.909 0.846 0.895 0.8 0.783 0.772 0.675 0.605 0.667 0.833
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5 14B Instruct
НГУ
0.436 0.427 0.469 0.371 0.364
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5 14B Instruct
НГУ
0.448 0.422 0.478 0.394 0.395
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5 14B Instruct
НГУ
0.453 0.438 0.473 0.398 0.428
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5 14B Instruct
НГУ
0.852 0.829 0.765 0.838 0.714 0.836