kfkas's picture
Update README.md
3870247
metadata
language:
  - en
  - ko
pipeline_tag: text-generation
inference: false
tags:
  - facebook
  - meta
  - pytorch
  - llama
  - llama-2
  - kollama
  - llama-2-ko
  - llama-2-ko-chat
  - legal-llama
  - law-llama
  - legal-gpt
  - law-gpt

Llama-2-Ko-7b-Chat은 kfkas/Llama-2-ko-7b-Chatλ₯Ό ν† λŒ€λ‘œ λ§Œλ“€μ–΄μ‘ŒμŠ΅λ‹ˆλ‹€. ν•™μŠ΅ λ°μ΄ν„°λŠ” 자체 법λ₯  질의 응닡 데이터λ₯Ό 톡해 ν•™μŠ΅ν•˜μ˜€μŠ΅λ‹ˆλ‹€.

Model Details

Backbone Model kfkas/Llama-2-ko-7b-Chat

Model Developers Boostcamp AI Tech Team YoonSeul

data leagl data

prompts kullm

Input Models input text only.

Output Models generate text only.


Model 질문 : μ•„λΉ κ°€ λŒμ•„κ°€μ‹œκ³  λ‚˜μ„œ ν•œ μ •μŠΉμΈκ³Ό 상속 포기 νŒκ²°λ¬Έμ„ λ°›μ•˜λŠ”λ°μš”. 그러고 λ‚˜μ„œ 근둜 κ±΄μ„€κ³΅μ œνšŒμ—μ„œ μ „ν™”κ°€ μ™”λŠ”λ°μš”. ν‡΄μ§κΈˆ κ³΅μ œκ°€ μžˆλ‹€κ³  μ‹ μ²­ν•˜μ—¬μ„œ 찾아가라고 ν•˜μ˜€λ”λΌκ³ μš”. 근데 이게 μƒμ†μΈμ˜ κ³ μœ μž¬μ‚°μ΄ λ‹€ μ•„λ‹ˆλ‹€ ν•˜μ—¬ 법원에 μž¬μ‚°μœΌλ‘œ μ‹ κ³ ν•˜μ—¬μ•Ό ν•œλ‹€κ³  μ•ˆ 해도 λœλ‹€κ³  ν•©λ‹ˆλ‹€. μ •ν™•ν•˜κ²Œ μ•Œκ³  μ‹ΆμŠ΅λ‹ˆλ‹€. μƒμ†μΈμ˜ κ³ μœ μž¬μ‚°μœΌλ‘œ 법원에 μž¬μ‚° μ‹ κ³ ν•˜μ—¬μ•Ό ν•˜λ‚˜μš”? μ•„λ‹ˆλ©΄ μ•ˆ 해도 λ˜λ‚˜μš”?
Llama-2-ko-7b-Chat [μ•ˆλ…•ν•˜μ„Έμš”! 상속인 고유 μž¬μ‚°μ΄ μ•„λ‹ˆλ”λΌλ„ κ·Όλ‘œκ±΄μ„€κ³΅μ œνšŒλ‘œλΆ€ν„° ν‡΄μ§κΈˆμ„ 지급받기 μœ„ν•΄ 법원에 μž¬μ‚°μ„ μ‹ κ³ ν•΄μ•Ό ν•  수 μžˆμŠ΅λ‹ˆλ‹€. ν•˜μ§€λ§Œ ꡬ체적인 상황은 상속인이 μ†ν•œ κ΅­κ°€μ˜ 법λ₯ μ— 따라 λ‹¬λΌμ§ˆ 수 μžˆμœΌλ―€λ‘œ, ν•΄λ‹Ή 법λ₯  μ „λ¬Έκ°€μ—κ²Œ λ¬Έμ˜ν•˜κ±°λ‚˜ ν™•μΈν•˜λŠ” 것이 μ’‹μŠ΅λ‹ˆλ‹€. 상속인이 상속 μž¬μ‚°μ„ μ‹ κ³ ν•˜μ§€ μ•Šκ³ λ„ ν‡΄μ§κΈˆμ„ 받을 수 μžˆλŠ” κ²½μš°λ„ μžˆμ§€λ§Œ, κ²½μš°μ— 따라 좔가적인 μ ˆμ°¨κ°€ ν•„μš”ν•  μˆ˜λ„ μžˆμŠ΅λ‹ˆλ‹€. 상속인은 상속 μž¬μ‚°μ˜ λ²”μœ„κ°€ λΆˆλΆ„λͺ…ν•˜κ±°λ‚˜ 상속 μž¬μ‚°μ΄ μ „ν˜€ 없을 수 μžˆλŠ” λ“± λ‹€μ–‘ν•œ 상황에 직면할 수 있기 λ•Œλ¬Έμ—, κ΄€λ ¨ 법λ₯  μ „λ¬Έκ°€μ˜ 쑰언을 κ΅¬ν•˜λŠ” 것이 μ€‘μš”ν•©λ‹ˆλ‹€.]
kfkas/Legal-Llama-2-ko-7b-Chat [μƒμ†μΈμ˜ 고유 μž¬μ‚°μ€ μƒμ†μΈμ—κ²Œ κ·€μ†λ˜λŠ” 상속 μž¬μ‚°μž…λ‹ˆλ‹€. 상속인이 상속 μž¬μ‚°μ— λŒ€ν•œ κΆŒλ¦¬μ™€ 의무λ₯Ό 행사할 수 μžˆμŠ΅λ‹ˆλ‹€. 상속 μž¬μ‚°μ—λŠ” 상속인 고유 μž¬μ‚°κ³Ό 일반 상속 μž¬μ‚°μ΄ ν¬ν•¨λ©λ‹ˆλ‹€. 상속인은 상속 μž¬μ‚°μ˜ λ²”μœ„ λ‚΄μ—μ„œ 상속 μž¬μ‚°μ„ κ΄€λ¦¬ν•˜κ³  μ²˜λΆ„ν•  수 있으며, 상속 μž¬μ‚° 쀑 일뢀λ₯Ό λ‹€λ₯Έ μƒμ†μΈμ—κ²Œ 증여할 μˆ˜λ„ μžˆμŠ΅λ‹ˆλ‹€.κ·ΈλŸ¬λ‚˜ 상속 μž¬μ‚° μ€‘μ—μ„œ 상속인이 고유 μž¬μ‚°μœΌλ‘œ μ·¨λ“ν•œ μž¬μ‚°μ΄ μžˆλŠ” 경우, 상속인은 κ·Έ μž¬μ‚°μ„ 상속 μž¬μ‚°μ—μ„œ μ œμ™Έμ†μ„Έκ°€ κ³Όμ„Έλ©λ‹ˆλ‹€.λ”°λΌμ„œ 상속 μž¬μ‚° 쀑에 상속인이 μ†Œμœ ν•œ 고유 μž¬μ‚°μ„ μ œμ™Έν•˜κ³  상속 μž¬μ‚°μ„ κ³„μ‚°ν•˜μ—¬ 상속세 μ‹ κ³ λ₯Ό ν•΄μ•Ό ν•˜λ©°, 상속세 μ‹ κ³  μ‹œμ—λŠ” 상속 μž¬μ‚° λͺ©λ‘μ„ μž‘μ„±ν•˜μ—¬ μ œμΆœν•˜μ—¬μ•Ό ν•©λ‹ˆλ‹€. λ˜ν•œ, μƒμ†μ„ΈλŠ” 상속 κ°œμ‹œμΌ(사망일)λ‘œλΆ€ν„° 6κ°œμ›” 이내에 μ‹ κ³ ν•΄μ•Ό ν•˜λ©°, μ‹ κ³ ν•˜μ§€ μ•Šμ„ 경우 κ°€μ‚°μ„Έκ°€ 뢀과될 수 μžˆμœΌλ―€λ‘œ μ£Όμ˜ν•΄μ•Ό ν•©λ‹ˆλ‹€.(μ°Έκ³ : 상속세법 제18μ‘°)]

Model 질문 : 재판이 λλ‚œ 후에 손해배상 청ꡬλ₯Ό ν•  수 μžˆλŠ”μ§€ κΆκΈˆν•©λ‹ˆλ‹€. 뢀동산 κ΄€λ ¨ 사기λ₯Ό λ‹Ήν•΄μ„œ μžμ„Έν•œ 상담을 λ°›κ³  μ‹Άμ–΄μš”
Llama-2-ko-7b-Chat [예, 재판이 λλ‚œ 후에도 민사 μ†Œμ†‘μ„ μ œκΈ°ν•  수 μžˆμŠ΅λ‹ˆλ‹€. 뢀동산 사기 λ˜λŠ” 기타 민사 λΆ„μŸκ³Ό κ΄€λ ¨ν•˜μ—¬ λ³€ν˜Έμ‚¬μ˜ 도움을 ꡬ할 수 μžˆλŠ” λͺ‡ 가지 μ˜΅μ…˜μ΄ μžˆμŠ΅λ‹ˆλ‹€:1. 민사 μ†Œμ†‘: 뢀동산 사기와 같은 민사 λΆ„μŸμ΄ μžˆλŠ” 경우 λ³€ν˜Έμ‚¬λ₯Ό κ³ μš©ν•˜μ—¬ λ―Όμ‚¬μ†Œμ†‘μ„ μ œκΈ°ν•˜μ—¬ 손해 배상을 μ²­κ΅¬ν•˜λŠ” 것이 κ°€μž₯ μΌλ°˜μ μž…λ‹ˆλ‹€.2. 개인 μ†Œμž₯: 민사 μ‚¬κ±΄μ—μ„œ 민사 μ†Œμž₯을 μž‘μ„±ν•˜μ—¬ 법원에 μ œμΆœν•˜λŠ” 것도 ν•œ 가지 λ°©λ²•μž…λ‹ˆλ‹€.3. μ€‘μž¬: μ€‘μž¬ 기관을 톡해 λΆ„μŸμ„ ν•΄κ²°ν•  μˆ˜λ„ μžˆμŠ΅λ‹ˆλ‹€.4. ν˜•μ‚¬ κ³ μ†Œ: 뢀동산 μ‚¬κΈ°λ‘œ ν˜•μ‚¬ κ³ μ†Œκ°€ κ°€λŠ₯ν•œμ§€ 확인해야 ν•©λ‹ˆλ‹€. μ΄λŸ¬ν•œ μ˜΅μ…˜ 쀑 μ–΄λ–€ 것을 선택할지 κ²°μ •ν•˜κΈ° 전에 λ³€ν˜Έμ‚¬μ™€ μƒλ‹΄ν•˜μ—¬ 쑰언을 κ΅¬ν•˜λŠ” 것이 μ’‹μŠ΅λ‹ˆλ‹€.]
Legal-Llama-2-ko-7b-Chat [예, 재판이 λλ‚œ 후에도 손해배상 청ꡬ가 κ°€λŠ₯ν•©λ‹ˆλ‹€. λ―Όμ‚¬μ†Œμ†‘λ²• 제766μ‘°(μ†ν•΄λ°°μƒμ²­κ΅¬κΆŒμ˜ μ†Œλ©Έμ‹œνš¨)에 λ”°λ₯΄λ©΄, λΆˆλ²•ν–‰μœ„λ‘œ μΈν•œ μ†ν•΄λ°°μƒμ˜ μ²­κ΅¬κΆŒμ€ ν”Όν•΄μžλ‚˜ κ·Έ λ²•μ •λŒ€λ¦¬μΈμ΄ κ·Έ 손해 및 κ°€ν•΄μžλ₯Ό μ•ˆ λ‚ λ‘œλΆ€ν„° 3λ…„κ°„ 이λ₯Ό ν–‰μ‚¬ν•˜μ§€ μ•„λ‹ˆν•˜λ©΄ μ‹œνš¨λ‘œ μΈν•˜μ—¬ μ†Œλ©Έν•œλ‹€κ³  κ·œμ •ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€.λ”°λΌμ„œ 뢀동산 μ‚¬κΈ°λ‘œ 인해 μž…μ€ 손해λ₯Ό μž…μ¦ν•  수 μžˆλŠ” 증거와 증인을 ν™•λ³΄ν•˜κ³ , 법원에 손해배상을 청ꡬ할 수 μžˆμŠ΅λ‹ˆλ‹€. λ˜ν•œ, 뢀동산 사기λ₯Ό λ‹Ήν•œ 사싀을 경찰에 μ‹ κ³ ν•˜μ—¬ μˆ˜μ‚¬λ₯Ό μ˜λ’°ν•˜λŠ” 것도 쒋은 λ°©λ²•μž…λ‹ˆλ‹€.]

Inference

def gen(x, model, tokenizer, device):
    prompt = (
        f"μ•„λž˜λŠ” μž‘μ—…μ„ μ„€λͺ…ν•˜λŠ” λͺ…λ Ήμ–΄μž…λ‹ˆλ‹€. μš”μ²­μ„ 적절히 μ™„λ£Œν•˜λŠ” 응닡을 μž‘μ„±ν•˜μ„Έμš”.\n\n### λͺ…λ Ήμ–΄:\n{x}\n\n### 응닡:"
    )
    len_prompt = len(prompt)
    gened = model.generate(
        **tokenizer(prompt, return_tensors="pt", return_token_type_ids=False).to(
            device
        ),
        max_new_tokens=1024,
        early_stopping=True,
        do_sample=True,
        top_k=20,
        top_p=0.92,
        no_repeat_ngram_size=3,
        eos_token_id=2,
        repetition_penalty=1.2,
        num_beams=3
    )
    return tokenizer.decode(gened[0])[len_prompt:]

def LLM_infer(input):
    device = (
        torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
    )
    model_id = "kfkas/Legal-Llama-2-ko-7b-Chat"
    model = AutoModelForCausalLM.from_pretrained(
        model_id, device_map={"": 0},torch_dtype=torch.float16, low_cpu_mem_usage=True
    )
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model.eval()
    model.config.use_cache = (True)
    tokenizer.pad_token = tokenizer.eos_token
    output = gen(input, model=model, tokenizer=tokenizer, device=device)

    return output


if __name__ == "__main__":
    text = LLM_infer("μŒμ£Όμš΄μ „μ„ ν•˜λ©΄ μ–΄λ–»κ²Œ 처벌 λ°›μ•„?")
    print(text)

Note for oobabooga/text-generation-webui

Remove ValueError at load_tokenizer function(line 109 or near), in modules/models.py.

diff --git a/modules/models.py b/modules/models.py
index 232d5fa..de5b7a0 100644
--- a/modules/models.py
+++ b/modules/models.py
@@ -106,7 +106,7 @@ def load_tokenizer(model_name, model):
                 trust_remote_code=shared.args.trust_remote_code,
                 use_fast=False
             )
-        except ValueError:
+        except:
             tokenizer = AutoTokenizer.from_pretrained(
                 path_to_model,
                 trust_remote_code=shared.args.trust_remote_code,

Since Llama-2-Ko uses FastTokenizer provided by HF tokenizers NOT sentencepiece package, it is required to use use_fast=True option when initialize tokenizer.

Apple Sillicon does not support BF16 computing, use CPU instead. (BF16 is supported when using NVIDIA GPU)


Below is the original model card of the Llama-2 model.

Llama 2

Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 7B pretrained model, converted for the Hugging Face Transformers format. Links to other models can be found in the index at the bottom.

Model Details

Note: Use of this model is governed by the Meta license. In order to download the model weights and tokenizer, please visit the website and accept our License before requesting access here.

Meta developed and publicly released the Llama 2 family of large language models (LLMs), a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. Our fine-tuned LLMs, called Llama-2-Chat, are optimized for dialogue use cases. Llama-2-Chat models outperform open-source chat models on most benchmarks we tested, and in our human evaluations for helpfulness and safety, are on par with some popular closed-source models like ChatGPT and PaLM.

Model Developers Meta

Variations Llama 2 comes in a range of parameter sizes β€” 7B, 13B, and 70B β€” as well as pretrained and fine-tuned variations.

Input Models input text only.

Output Models generate text only.

Model Architecture Llama 2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align to human preferences for helpfulness and safety.

Training Data Params Content Length GQA Tokens LR
Llama 2 A new mix of publicly available online data 7B 4k βœ— 2.0T 3.0 x 10-4
Llama 2 A new mix of publicly available online data 13B 4k βœ— 2.0T 3.0 x 10-4
Llama 2 A new mix of publicly available online data 70B 4k βœ” 2.0T 1.5 x 10-4

Llama 2 family of models. Token counts refer to pretraining data only. All models are trained with a global batch-size of 4M tokens. Bigger models - 70B -- use Grouped-Query Attention (GQA) for improved inference scalability.

Model Dates Llama 2 was trained between January 2023 and July 2023.

Status This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we improve model safety with community feedback.

License A custom commercial license is available at: https://ai.meta.com/resources/models-and-libraries/llama-downloads/

Research Paper "Llama-2: Open Foundation and Fine-tuned Chat Models"

Intended Use

Intended Use Cases Llama 2 is intended for commercial and research use in English. Tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.

To get the expected features and performance for the chat versions, a specific formatting needs to be followed, including the INST and <<SYS>> tags, BOS and EOS tokens, and the whitespaces and breaklines in between (we recommend calling strip() on inputs to avoid double-spaces). See our reference code in github for details: chat_completion.

Out-of-scope Uses Use in any manner that violates applicable laws or regulations (including trade compliance laws).Use in languages other than English. Use in any other way that is prohibited by the Acceptable Use Policy and Licensing Agreement for Llama 2.

Hardware and Software

Training Factors We used custom training libraries, Meta's Research Super Cluster, and production clusters for pretraining. Fine-tuning, annotation, and evaluation were also performed on third-party cloud compute.

Carbon Footprint Pretraining utilized a cumulative 3.3M GPU hours of computation on hardware of type A100-80GB (TDP of 350-400W). Estimated total emissions were 539 tCO2eq, 100% of which were offset by Meta’s sustainability program.

Time (GPU hours) Power Consumption (W) Carbon Emitted(tCO2eq)
Llama 2 7B 184320 400 31.22
Llama 2 13B 368640 400 62.44
Llama 2 70B 1720320 400 291.42
Total 3311616 539.00

CO2 emissions during pretraining. Time: total GPU time required for training each model. Power Consumption: peak power capacity per GPU device for the GPUs used adjusted for power usage efficiency. 100% of the emissions are directly offset by Meta's sustainability program, and because we are openly releasing these models, the pretraining costs do not need to be incurred by others.

Training Data

Overview Llama 2 was pretrained on 2 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over one million new human-annotated examples. Neither the pretraining nor the fine-tuning datasets include Meta user data.

Data Freshness The pretraining data has a cutoff of September 2022, but some tuning data is more recent, up to July 2023.

Evaluation Results

In this section, we report the results for the Llama 1 and Llama 2 models on standard academic benchmarks.For all the evaluations, we use our internal evaluations library.

Model Size Code Commonsense Reasoning World Knowledge Reading Comprehension Math MMLU BBH AGI Eval
Llama 1 7B 14.1 60.8 46.2 58.5 6.95 35.1 30.3 23.9
Llama 1 13B 18.9 66.1 52.6 62.3 10.9 46.9 37.0 33.9
Llama 1 33B 26.0 70.0 58.4 67.6 21.4 57.8 39.8 41.7
Llama 1 65B 30.7 70.7 60.5 68.6 30.8 63.4 43.5 47.6
Llama 2 7B 16.8 63.9 48.9 61.3 14.6 45.3 32.6 29.3
Llama 2 13B 24.5 66.9 55.4 65.8 28.7 54.8 39.4 39.1
Llama 2 70B 37.5 71.9 63.6 69.4 35.2 68.9 51.2 54.2

Overall performance on grouped academic benchmarks. Code: We report the average pass@1 scores of our models on HumanEval and MBPP. Commonsense Reasoning: We report the average of PIQA, SIQA, HellaSwag, WinoGrande, ARC easy and challenge, OpenBookQA, and CommonsenseQA. We report 7-shot results for CommonSenseQA and 0-shot results for all other benchmarks. World Knowledge: We evaluate the 5-shot performance on NaturalQuestions and TriviaQA and report the average. Reading Comprehension: For reading comprehension, we report the 0-shot average on SQuAD, QuAC, and BoolQ. MATH: We report the average of the GSM8K (8 shot) and MATH (4 shot) benchmarks at top 1.

TruthfulQA Toxigen
Llama 1 7B 27.42 23.00
Llama 1 13B 41.74 23.08
Llama 1 33B 44.19 22.57
Llama 1 65B 48.71 21.77
Llama 2 7B 33.29 21.25
Llama 2 13B 41.86 26.10
Llama 2 70B 50.18 24.60

Evaluation of pretrained LLMs on automatic safety benchmarks. For TruthfulQA, we present the percentage of generations that are both truthful and informative (the higher the better). For ToxiGen, we present the percentage of toxic generations (the smaller the better).

TruthfulQA Toxigen
Llama-2-Chat 7B 57.04 0.00
Llama-2-Chat 13B 62.18 0.00
Llama-2-Chat 70B 64.14 0.01

Evaluation of fine-tuned LLMs on different safety datasets. Same metric definitions as above.

Ethical Considerations and Limitations

Llama 2 is a new technology that carries risks with 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 2’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 2, developers should perform safety testing and tuning tailored to their specific applications of the model.

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

Reporting Issues

Please report any software β€œbug,” or other problems with the models through one of the following means:

Llama Model Index

Model Llama2 Llama2-hf Llama2-chat Llama2-chat-hf
7B Link Link Link Link
13B Link Link Link Link
70B Link Link Link Link