beomi/Llama-3-Open-Ko-8B

Update @ 2024.05.20: Re-Upload RoPE fixed model

Update @ 2024.05.01: Pre-Release Llama-3-KoEn-8B model & Llama-3-KoEn-8B-Instruct-preview

Update @ 2024.04.24: Release Llama-3-Open-Ko-8B model & Llama-3-Open-Ko-8B-Instruct-preview

Model Details

Llama-3-Open-Ko-8B

Llama-3-Open-Ko-8B model is continued pretrained language model based on Llama-3-8B.

This model is trained fully with publicily available resource, with 60GB+ of deduplicated texts.

With the new Llama-3 tokenizer, the pretraining conducted with 17.7B+ tokens, which slightly more than Korean tokenizer(Llama-2-Ko tokenizer).

The train was done on TPUv5e-256, with the warm support from TRC program by Google.

Note for Llama-3-Open-Ko-8B-Instruct-preview

With applying the idea from Chat Vector paper, I released Instruction model named Llama-3-Open-Ko-8B-Instruct-preview.

Since it is NOT finetuned with any Korean instruction set(indeed preview), but it would be great starting point for creating new Chat/Instruct models.

Meta Llama-3

Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.

Model developers Junbum Lee (Beomi)

Variations Llama-3-Open-Ko comes in one size — 8B.

Input Models input text only.

Output Models generate text and code only.

Model Architecture Llama 3 is an auto-regressive language model that uses an optimized transformer architecture.

	Training Data	Params	Context length	GQA	Token count	Knowledge cutoff
Llama-3-Open-Ko	Same as *Open-Solar-Ko Dataset	8B	8k	Yes	17.7B+	Jun, 2023

*You can find dataset list here: https://huggingface.co/beomi/OPEN-SOLAR-KO-10.7B/tree/main/corpus

Model Release Date 2024.04.24.

Status This is a static model trained on an offline dataset.

License Llama3 License: https://llama.meta.com/llama3/license

Intended Use

Intended Use Cases Llama 3 is intended for commercial and research use in English. Instruction tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.

Out-of-scope Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3 Community License. Use in languages other than English**.

**Note: Developers may fine-tune Llama 3 models for languages beyond English provided they comply with the Llama 3 Community License and the Acceptable Use Policy.

How to use

TBD

Responsibility & Safety

We believe that an open approach to AI leads to better, safer products, faster innovation, and a bigger overall market. We are committed to Responsible AI development and took a series of steps to limit misuse and harm and support the open source community.

Foundation models are widely capable technologies that are built to be used for a diverse range of applications. They are not designed to meet every developer preference on safety levels for all use cases, out-of-the-box, as those by their nature will differ across different applications.

Rather, responsible LLM-application deployment is achieved by implementing a series of safety best practices throughout the development of such applications, from the model pre-training, fine-tuning and the deployment of systems composed of safeguards to tailor the safety needs specifically to the use case and audience.

As part of the Llama 3 release, we updated our Responsible Use Guide to outline the steps and best practices for developers to implement model and system level safety for their application. We also provide a set of resources including Meta Llama Guard 2 and Code Shield safeguards. These tools have proven to drastically reduce residual risks of LLM Systems, while maintaining a high level of helpfulness. We encourage developers to tune and deploy these safeguards according to their needs and we provide a reference implementation to get you started.

Responsible release

In addition to responsible use considerations outlined above, we followed a rigorous process that requires us to take extra measures against misuse and critical risks before we make our release decision.

Misuse

If you access or use Llama 3, you agree to the Acceptable Use Policy. The most recent copy of this policy can be found at https://llama.meta.com/llama3/use-policy/.

Ethical Considerations and Limitations

The core values of Llama 3 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.

But Llama 3 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has been in English, and has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3 models, developers should perform safety testing and tuning tailored to their specific applications of the model. As outlined in the Responsible Use Guide, we recommend incorporating Purple Llama solutions into your workflows and specifically Llama Guard which provides a base model to filter input and output prompts to layer system-level safety on top of model-level safety.

Please see the Responsible Use Guide available at http://llama.meta.com/responsible-use-guide

Benchmark Scores

• vllm (pretrained=beomi/Llama-3-Open-Ko-8B,revision=081e85a,tensor_parallel_size=1,dtype=bfloat16,data_parallel_size=2,gpu_memory_utilization=0.8), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: auto

Tasks	Version	Filter	n-shot	Metric	Value		Stderr
haerae	N/A	none	5	acc	0.6801	±	0.0138
		none	5	acc_norm	0.6801	±	0.0138
- haerae_general_knowledge	1	none	5	acc	0.4375	±	0.0375
		none	5	acc_norm	0.4375	±	0.0375
- haerae_history	1	none	5	acc	0.7340	±	0.0323
		none	5	acc_norm	0.7340	±	0.0323
- haerae_loan_word	1	none	5	acc	0.7870	±	0.0316
		none	5	acc_norm	0.7870	±	0.0316
- haerae_rare_word	1	none	5	acc	0.7012	±	0.0228
		none	5	acc_norm	0.7012	±	0.0228
- haerae_standard_nomenclature	1	none	5	acc	0.7190	±	0.0365
		none	5	acc_norm	0.7190	±	0.0365
kmmlu_direct	N/A	none	5	exact_match	0.4054	±	0.0026
- kmmlu_direct_accounting	2	none	5	exact_match	0.3600	±	0.0482
- kmmlu_direct_agricultural_sciences	2	none	5	exact_match	0.3130	±	0.0147
- kmmlu_direct_aviation_engineering_and_maintenance	2	none	5	exact_match	0.3690	±	0.0153
- kmmlu_direct_biology	2	none	5	exact_match	0.3330	±	0.0149
- kmmlu_direct_chemical_engineering	2	none	5	exact_match	0.4190	±	0.0156
- kmmlu_direct_chemistry	2	none	5	exact_match	0.3833	±	0.0199
- kmmlu_direct_civil_engineering	2	none	5	exact_match	0.3870	±	0.0154
- kmmlu_direct_computer_science	2	none	5	exact_match	0.6340	±	0.0152
- kmmlu_direct_construction	2	none	5	exact_match	0.3340	±	0.0149
- kmmlu_direct_criminal_law	2	none	5	exact_match	0.2850	±	0.0320
- kmmlu_direct_ecology	2	none	5	exact_match	0.4210	±	0.0156
- kmmlu_direct_economics	2	none	5	exact_match	0.4077	±	0.0433
- kmmlu_direct_education	2	none	5	exact_match	0.5000	±	0.0503
- kmmlu_direct_electrical_engineering	2	none	5	exact_match	0.3620	±	0.0152
- kmmlu_direct_electronics_engineering	2	none	5	exact_match	0.4790	±	0.0158
- kmmlu_direct_energy_management	2	none	5	exact_match	0.3110	±	0.0146
- kmmlu_direct_environmental_science	2	none	5	exact_match	0.3210	±	0.0148
- kmmlu_direct_fashion	2	none	5	exact_match	0.4190	±	0.0156
- kmmlu_direct_food_processing	2	none	5	exact_match	0.3600	±	0.0152
- kmmlu_direct_gas_technology_and_engineering	2	none	5	exact_match	0.3320	±	0.0149
- kmmlu_direct_geomatics	2	none	5	exact_match	0.3640	±	0.0152
- kmmlu_direct_health	2	none	5	exact_match	0.5100	±	0.0502
- kmmlu_direct_industrial_engineer	2	none	5	exact_match	0.3970	±	0.0155
- kmmlu_direct_information_technology	2	none	5	exact_match	0.5720	±	0.0157
- kmmlu_direct_interior_architecture_and_design	2	none	5	exact_match	0.4740	±	0.0158
- kmmlu_direct_korean_history	2	none	5	exact_match	0.2700	±	0.0446
- kmmlu_direct_law	2	none	5	exact_match	0.3990	±	0.0155
- kmmlu_direct_machine_design_and_manufacturing	2	none	5	exact_match	0.4080	±	0.0155
- kmmlu_direct_management	2	none	5	exact_match	0.4660	±	0.0158
- kmmlu_direct_maritime_engineering	2	none	5	exact_match	0.4417	±	0.0203
- kmmlu_direct_marketing	2	none	5	exact_match	0.6720	±	0.0149
- kmmlu_direct_materials_engineering	2	none	5	exact_match	0.4130	±	0.0156
- kmmlu_direct_math	2	none	5	exact_match	0.2567	±	0.0253
- kmmlu_direct_mechanical_engineering	2	none	5	exact_match	0.3800	±	0.0154
- kmmlu_direct_nondestructive_testing	2	none	5	exact_match	0.3890	±	0.0154
- kmmlu_direct_patent	2	none	5	exact_match	0.2700	±	0.0446
- kmmlu_direct_political_science_and_sociology	2	none	5	exact_match	0.4433	±	0.0287
- kmmlu_direct_psychology	2	none	5	exact_match	0.3620	±	0.0152
- kmmlu_direct_public_safety	2	none	5	exact_match	0.3200	±	0.0148
- kmmlu_direct_railway_and_automotive_engineering	2	none	5	exact_match	0.3200	±	0.0148
- kmmlu_direct_real_estate	2	none	5	exact_match	0.3650	±	0.0341
- kmmlu_direct_refrigerating_machinery	2	none	5	exact_match	0.3210	±	0.0148
- kmmlu_direct_social_welfare	2	none	5	exact_match	0.4500	±	0.0157
- kmmlu_direct_taxation	2	none	5	exact_match	0.3550	±	0.0339
- kmmlu_direct_telecommunications_and_wireless_technology	2	none	5	exact_match	0.5490	±	0.0157
kobest_boolq	1	none	5	acc	0.7984	±	0.0107
		none	5	f1	0.7961	±	N/A
kobest_copa	1	none	5	acc	0.8150	±	0.0123
		none	5	f1	0.8148	±	N/A
kobest_hellaswag	1	none	5	acc	0.4800	±	0.0224
		none	5	f1	0.4771	±	N/A
		none	5	acc_norm	0.6120	±	0.0218
kobest_sentineg	1	none	5	acc	0.9597	±	0.0099
		none	5	f1	0.9597	±	N/A
haerae	N/A	none	5	acc	0.6801	±	0.0138
		none	5	acc_norm	0.6801	±	0.0138
kmmlu_direct	N/A	none	5	exact_match	0.4054	±	0.0026

Citation instructions

Llama-3-Open-Ko

@article{llama3openko,
  title={Llama-3-Open-Ko},
  author={L, Junbum},
  year={2024},
  url={https://huggingface.co/beomi/Llama-3-Open-Ko-8B}
}

Original Llama-3

@article{llama3modelcard,
  title={Llama 3 Model Card},
  author={AI@Meta},
  year={2024},
  url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
}