- 模型資訊
- 模型資料
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 fine-tuned model, optimized for dialogue use cases.
Llama 2是一系列預(yù)訓(xùn)練和微調(diào)的生成文本模型的集合，參數(shù)規(guī)模從70億到700億不等。這是7B微調(diào)模型的存儲(chǔ)庫，專為對(duì)話用例優(yōu)化，并適配到ModelScope生態(tài)，可以通過ModelScope library加載。
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.
注意：使用此模型受Meta許可證的約束。為了下載模型權(quán)重和分詞器，請(qǐng)?jiān)L問網(wǎng)站并接受我們的許可證，然后在這里請(qǐng)求訪問。
Meta開發(fā)并公開發(fā)布了Llama 2系列的大型語言模型（LLMs），這是一系列預(yù)訓(xùn)練和微調(diào)的生成文本模型的集合，參數(shù)規(guī)模從70億到700億不等。我們微調(diào)的LLMs，稱為Llama-2-Chat，專為對(duì)話用例優(yōu)化。Llama-2-Chat模型在我們測(cè)試的大多數(shù)基準(zhǔn)上優(yōu)于開源聊天模型，并且在我們對(duì)幫助性和安全性的人類評(píng)估中，與一些流行的閉源模型如ChatGPT和PaLM相當(dāng)。
Model Developers Meta
Variations Llama 2 comes in a range of parameter sizes — 7B, 13B, and 70B — 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.
Llama 2有多種參數(shù)大小——7B、13B和70B——以及預(yù)訓(xùn)練和微調(diào)的版本。
輸入 模型只輸入文本。
輸出 模型只生成文本。
模型架構(gòu) Llama 2是一個(gè)自回歸語言模型，使用優(yōu)化的Transformer架構(gòu)。調(diào)整版本使用監(jiān)督微調(diào)（SFT）和人類反饋的強(qiáng)化學(xué)習(xí)（RLHF）來符合人類對(duì)幫助性和安全性的偏好。
| Training Data | Params | Content Length | GQA | Tokens | LR | |
|---|---|---|---|---|---|---|
| Llama 2 | A new mix of publicly available online data | 7B | 4k | ? | 2.0T | 3.0 x 10-4 |
| Llama 2 | A new mix of publicly available online data | 13B | 4k | ? | 2.0T | 3.0 x 10-4 |
| Llama 2 | A new mix of publicly available online data | 70B | 4k | ? | 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/
模型日期 Llama 2的訓(xùn)練時(shí)間為2023年1月至2023年7月。
狀態(tài) 這是一個(gè)在離線數(shù)據(jù)集上訓(xùn)練的靜態(tài)模型。隨著我們改進(jìn)模型安全性的社區(qū)反饋，調(diào)整模型的未來版本將會(huì)發(fā)布。
許可證 自定義商業(yè)許可證可在以下網(wǎng)址獲?。?a >https://ai.meta.com/resources/models-and-libraries/llama-downloads/
示例代碼
pip install modelscope -U
推理代碼
import torch
from modelscope import Model, snapshot_download
from modelscope.models.nlp.llama2 import Llama2Tokenizer
model_dir = snapshot_download("modelscope/Llama-2-7b-chat-ms", revision='v1.0.2',
ignore_file_pattern=[r'.+\.bin$'])
tokenizer = Llama2Tokenizer.from_pretrained(model_dir)
model = Model.from_pretrained(
model_dir,
torch_dtype=torch.float16,
device_map='auto')
system = 'you are a helpful assistant!'
inputs = {'text': 'Where is the capital of Zhejiang?', 'system': system, 'max_length': 512}
output = model.chat(inputs, tokenizer)
print(output['response'])
inputs = {'text': 'What are the interesting places there?',
'system': system,
'history': output['history'],
'max_length': 512}
output = model.chat(inputs, tokenizer)
print(output['response'])
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.
預(yù)期用例 Llama 2預(yù)期用于商業(yè)和研究用途，語言為英語。調(diào)整模型預(yù)期用于類似助手的聊天，而預(yù)訓(xùn)練模型可以適應(yīng)各種自然語言生成任務(wù)。
為了獲得聊天版本的預(yù)期特性和性能，需要遵循特定的格式，包括INST和<
超出范圍的用途 以任何違反適用法律或法規(guī)（包括貿(mào)易合規(guī)法）的方式使用。使用英語以外的語言。以任何其他方式使用，這被Llama 2的可接受使用政策和許可協(xié)議所禁止。
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.
訓(xùn)練因素 我們使用了定制的訓(xùn)練庫，Meta的研究超級(jí)集群，以及用于預(yù)訓(xùn)練的生產(chǎn)集群。微調(diào)、注釋和評(píng)估也在第三方云計(jì)算上進(jìn)行。
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.
碳足跡 預(yù)訓(xùn)練使用了累計(jì)330萬GPU小時(shí)的計(jì)算，硬件類型為A100-80GB（TDP為350-400W）。估計(jì)的總排放量為539 tCO2eq，其中100%由Meta的可持續(xù)性計(jì)劃抵消。
| 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.
預(yù)訓(xùn)練期間的CO2排放。時(shí)間：訓(xùn)練每個(gè)模型所需的總GPU時(shí)間。功耗：根據(jù)功耗效率調(diào)整的每個(gè)GPU設(shè)備的峰值功率容量。100%的排放直接由Meta的可持續(xù)性計(jì)劃抵消，因?yàn)槲覀児_發(fā)布這些模型，所以不需要其他人承擔(dān)預(yù)訓(xùn)練的成本。
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.
概述 Llama 2在來自公開可用源的2萬億令牌的數(shù)據(jù)上進(jìn)行了預(yù)訓(xùn)練。微調(diào)數(shù)據(jù)包括公開可用的指令數(shù)據(jù)集，以及超過一百萬個(gè)新的人類注釋示例。預(yù)訓(xùn)練和微調(diào)數(shù)據(jù)集都不包括Meta用戶數(shù)據(jù)。
Data Freshness The pretraining data has a cutoff of September 2022, but some tuning data is more recent, up to July 2023.
數(shù)據(jù)新鮮度 預(yù)訓(xùn)練數(shù)據(jù)的截止日期為2022年9月，但一些調(diào)整數(shù)據(jù)更近，最近的數(shù)據(jù)為2023年7月。
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.
在這一部分，我們報(bào)告了Llama 1和Llama 2模型在標(biāo)準(zhǔn)學(xué)術(shù)基準(zhǔn)上的結(jié)果。對(duì)于所有的評(píng)估，我們使用我們的內(nèi)部評(píng)估庫。
| 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.
Llama 2是一種新技術(shù)，使用時(shí)帶有風(fēng)險(xiǎn)。到目前為止進(jìn)行的測(cè)試是用英語進(jìn)行的，并且沒有覆蓋，也無法覆蓋所有的情況。因此，與所有LLMs一樣，Llama 2的潛在輸出無法預(yù)先預(yù)測(cè)，模型可能在某些情況下對(duì)用戶提示產(chǎn)生不準(zhǔn)確、有偏見或其他令人反感的響應(yīng)。因此，在部署Llama 2的任何應(yīng)用之前，開發(fā)者應(yīng)進(jìn)行針對(duì)他們的特定應(yīng)用的安全測(cè)試和調(diào)整。
Please see the Responsible Use Guide available at https://ai.meta.com/llama/responsible-use-guide/
請(qǐng)參閱https://ai.meta.com/llama/responsible-use-guide/上的負(fù)責(zé)任使用指南
Reporting Issues
Please report any software “bug,” or other problems with the models through one of the following means:
- Reporting issues with the model: github.com/facebookresearch/llama
- Reporting problematic content generated by the model: developers.facebook.com/llama_output_feedback
- Reporting bugs and security concerns: facebook.com/whitehat/info