beomi/gemma-mling-7b

Gemma-Mling: Multilingual Gemma

Update @ 2024.04.15: First release of Gemma-Mling 7B model

Original Gemma Model Page: Gemma

This model card corresponds to the 7B base version of the Gemma-Mling model, continual pretrained on mainly Korean/English/Chinese/Japanese + 500 multilingual corpus.

Resources and Technical Documentation:

• Original Google's Gemma-7B
• Training Code @ Github: Gemma-EasyLM

Terms of Use: Terms

Citation

@misc {gemma_mling_7b,
    author       = { {Junbum Lee, Taekyoon Choi} },
    title        = { gemma-mling-7b },
    year         = 2024,
    url          = { https://huggingface.co./beomi/gemma-mling-7b },
    publisher    = { Hugging Face }
}

Model Developers: Junbum Lee (Beomi) & Taekyoon Choi (Taekyoon)

Model Information

Usage

Below we share some code snippets on how to get quickly started with running the model. First make sure to pip install -U transformers, then copy the snippet from the section that is relevant for your usecase.

Running the model on a CPU

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("beomi/gemma-mling-7b")
model = AutoModelForCausalLM.from_pretrained("beomi/gemma-mling-7b")

input_text = "머신러닝과 딥러닝의 차이는"
input_ids = tokenizer(input_text, return_tensors="pt")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

Running the model on a single / multi GPU

# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("beomi/gemma-mling-7b")
model = AutoModelForCausalLM.from_pretrained("beomi/gemma-mling-7b", device_map="auto")

input_text = "머신러닝과 딥러닝의 차이는"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

Inputs and outputs

• Input: Text string, such as a question, a prompt, or a document to be summarized.
• Output: Generated Multilingual-language text in response to the input, such as an answer to a question, or a summary of a document.

Implementation Information

Details about the model internals.

Software

Training was done using beomi/Gemma-EasyLM.

Dataset

We trained a mixture of multiple language datasets and trained until 100B. The released model is the best performance model based on our Evaluation below from model checkpoints.

For Korean and English datasets, we utilized sampled llama2ko training dataset which combined 1:1 ratio in each language.

Dataset	Jsonl (GB)	Sampled
range3/cc100-ja	96.39	No
Skywork/SkyPile-150B	100.57	Yes
llama2ko dataset (ko/en)	108.5	Yes
cis-lmu/Glot500	181.24	No
Total	486.7	.

Training Progress

• Report Link: https://api.wandb.ai/links/tgchoi/6lt0ce3s

Evaluation

Model evaluation metrics and results.

Evaluation Scripts

• For Knowledge / KoBest / XCOPA / XWinograd
  • EleutherAI/lm-evaluation-harness v0.4.2

    !git clone https://github.com/EleutherAI/lm-evaluation-harness.git
    !cd lm-evaluation-harness && pip install -r requirements.txt && pip install -e .
    
    !lm_eval --model hf \
      --model_args pretrained=beomi/gemma-mling-7b,dtype="float16" \
      --tasks "haerae,kobest,kmmlu_direct,cmmlu,ceval-valid,mmlu,xwinograd,xcopa \
      --num_fewshot "0,5,5,5,5,5,0,5" \
      --device cuda 
    

• For JP Eval Harness
  • Stability-AI/lm-evaluation-harness (jp-stable branch)

    !git clone -b jp-stable https://github.com/Stability-AI/lm-evaluation-harness.git
    !cd lm-evaluation-harness && pip install -e ".[ja]"
    !pip install 'fugashi[unidic]' && python -m unidic download
    
    !cd lm-evaluation-harness && python main.py \
        --model hf-causal \
        --model_args pretrained=beomi/gemma-mling-7b,torch_dtype='auto'"
        --tasks "jcommonsenseqa-1.1-0.3,jnli-1.3-0.3,marc_ja-1.1-0.3,jsquad-1.1-0.3,jaqket_v2-0.2-0.3,xlsum_ja,mgsm"
        --num_fewshot "3,3,3,2,1,1,5"
    

Benchmark Results

Category	Metric	Shots	Score
Default Metric	ACC
Knowledge (5-shot)	MMLU		61.76
	KMMLU (Exact Match)		42.75
	CMLU		50.93
	JMLU
	C-EVAL		50.07
	HAERAE	0-shot	63.89
KoBest (5-shot)	BoolQ		85.47
	COPA		83.5
	Hellaswag (acc-norm)		63.2
	Sentineg		97.98
	WiC		70.95
XCOPA (5-shot)	IT		72.8
	ID		76.4
	TH		60.2
	TR		65.6
	VI		77.2
	ZH		80.2
JP Eval Harness (Prompt ver 0.3)	JcommonsenseQA	3-shot	85.97
	JNLI	3-shot	39.11
	Marc_ja	3-shot	96.48
	JSquad (Exact Match)	2-shot	70.69
	Jaqket (Exact Match)	1-shot	81.53
	MGSM	5-shot	28.8
XWinograd (0-shot)	EN		89.03
	FR		72.29
	JP		82.69
	PT		73.38
	RU		68.57
	ZH		79.17

Usage and Limitations

These models have certain limitations that users should be aware of.

Intended Usage

Open Large Language Models (LLMs) have a wide range of applications across various industries and domains. The following list of potential uses is not comprehensive. The purpose of this list is to provide contextual information about the possible use-cases that the model creators considered as part of model training and development.

• Content Creation and Communication
  • Text Generation: These models can be used to generate creative text formats such as poems, scripts, code, marketing copy, and email drafts.
• Research and Education
  • Natural Language Processing (NLP) Research: These models can serve as a foundation for researchers to experiment with NLP techniques, develop algorithms, and contribute to the advancement of the field.
  • Language Learning Tools: Support interactive language learning experiences, aiding in grammar correction or providing writing practice.
  • Knowledge Exploration: Assist researchers in exploring large bodies of text by generating summaries or answering questions about specific topics.

Limitations

• Training Data
  • The quality and diversity of the training data significantly influence the model's capabilities. Biases or gaps in the training data can lead to limitations in the model's responses.
  • The scope of the training dataset determines the subject areas the model can handle effectively.
• Context and Task Complexity
  • LLMs are better at tasks that can be framed with clear prompts and instructions. Open-ended or highly complex tasks might be challenging.
  • A model's performance can be influenced by the amount of context provided (longer context generally leads to better outputs, up to a certain point).
• Language Ambiguity and Nuance
  • Natural language is inherently complex. LLMs might struggle to grasp subtle nuances, sarcasm, or figurative language.
• Factual Accuracy
  • LLMs generate responses based on information they learned from their training datasets, but they are not knowledge bases. They may generate incorrect or outdated factual statements.
• Common Sense
  • LLMs rely on statistical patterns in language. They might lack the ability to apply common sense reasoning in certain situations.

Ethical Considerations and Risks

The development of large language models (LLMs) raises several ethical concerns. In creating an open model, we have carefully considered the following:

• Bias and Fairness
  • LLMs trained on large-scale, real-world text data can reflect socio-cultural biases embedded in the training material. These models underwent careful scrutiny, input data pre-processing described and posterior evaluations reported in this card.
• Misinformation and Misuse
  • LLMs can be misused to generate text that is false, misleading, or harmful.
  • Guidelines are provided for responsible use with the model, see the Responsible Generative AI Toolkit.
• Transparency and Accountability:
  • This model card summarizes details on the models' architecture, capabilities, limitations, and evaluation processes.
  • A responsibly developed open model offers the opportunity to share innovation by making LLM technology accessible to developers and researchers across the AI ecosystem.

Risks identified and mitigations:

• Perpetuation of biases: It's encouraged to perform continuous monitoring (using evaluation metrics, human review) and the exploration of de-biasing techniques during model training, fine-tuning, and other use cases.
• Generation of harmful content: Mechanisms and guidelines for content safety are essential. Developers are encouraged to exercise caution and implement appropriate content safety safeguards based on their specific product policies and application use cases.
• Misuse for malicious purposes: Technical limitations and developer and end-user education can help mitigate against malicious applications of LLMs. Educational resources and reporting mechanisms for users to flag misuse are provided. Prohibited uses of Gemma models are outlined in the Gemma Prohibited Use Policy.
• Privacy violations: Models were trained on data filtered for removal of PII (Personally Identifiable Information). Developers are encouraged to adhere to privacy regulations with privacy-preserving techniques.

Acknowledgement

The training is supported by TPU Research Cloud program.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	11.18
IFEval (0-Shot)	20.29
BBH (3-Shot)	17.63
MATH Lvl 5 (4-Shot)	4.15
GPQA (0-shot)	0.00
MuSR (0-shot)	6.85
MMLU-PRO (5-shot)	18.14