Tamil Llama 7b Instruct V0.2

🚀 泰米尔语LLaMA 7B指令模型v0.2

泰米尔语LLaMA 7B指令模型的首次发布是推进泰米尔语大语言模型发展的重要一步。该模型可立即用于推理，也可进一步微调以满足您特定的自然语言处理任务需求。

若您想深入了解该模型的开发过程和能力，请阅读研究论文和介绍性博客文章（待完成），其中概述了我们的研究历程以及该模型的潜在影响。

🚀 快速开始

本模型可立即用于推理，也可进一步微调以满足您特定的自然语言处理任务需求。您可以通过以下方式深入了解模型：

• 阅读研究论文，了解模型的开发过程和能力。
• 查看介绍性博客文章（待完成），了解我们的研究历程以及该模型的潜在影响。

✨ 主要特性

• 双语支持：支持英语和泰米尔语，适用于双语自然语言处理任务。
• 可微调：可进一步微调以满足特定的自然语言处理任务需求。
• 高性能：在多个基准测试中表现出色，如ARC Challenge、TruthfulQA等。

📦 安装指南

文档未提供具体安装步骤，暂不展示。

💻 使用示例

基础用法

from transformers import LlamaForCausalLM, AutoTokenizer, pipeline

model = LlamaForCausalLM.from_pretrained(
    "abhinand/tamil-llama-instruct-v0.2",
    #load_in_8bit=True, # Set this depending on the GPU you have
    torch_dtype=torch.bfloat16,
    device_map={"": 0}, # Set this depending on the number of GPUs you have
    local_files_only=False # Optional
)
model.eval()

tokenizer = AutoTokenizer.from_pretrained("abhinand/tamil-llama-instruct-v0.2")

inf_pipeline = pipeline("conversational", model=model, tokenizer=tokenizer)


def format_instruction(system_prompt, question, return_dict=False):
	if system_prompt is None:
		messages = [
			{'content': question, 'role': 'user'},
		]
	else:
		messages = [
			{'content': system_prompt, 'role': 'system'},
			{'content': question, 'role': 'user'},
		]

	if return_dict:
		return messages

	prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

	return prompt


# Set the generation configuration according to your needs
temperature = 0.6
repetition_penalty = 1.1
max_new_tokens = 256

SYSTEM_PROMPT = "You are an AI assistant who follows instructions extremely well. Do your best your best to help."
INPUT = "Can you explain the significance of Tamil festival Pongal?"

instruction = format_instruction(
    system_prompt=SYSTEM_PROMPT,
    question=INPUT,
    return_dict=True,
)

output = inf_pipeline(
    instruction,
    temperature=temperature,
    max_new_tokens=max_new_tokens,
    repetition_penalty=repetition_penalty
)
print(output)

示例输出

Conversation id: d57cdf33-01ff-4328-8efe-5c4fefdd6e77
system: You are an AI assistant who follows instructions extremely well. Do your best your best to help.
user: Can you explain the significance of Tamil festival Pongal?
assistant: Pongal is a significant harvest festival celebrated in Tamil Nadu and other parts of southern India. It marks the end of the rainy season and beginning of the agricultural year. The festival primarily revolves around giving gratitude to nature, particularly the Sun God Surya for his bountiful gifts like agriculture and health. People offer prayers to cattle, which play a significant role in agriculture, as well as their families for their continued support during the harvest season. The festival is marked by various colorful events, including preparing traditional Pongal dishes like rice cooked with milk, sugarcane, and banana, followed by exchanging gifts and celebrating among family members and friends. It also serves as a time for unity and strengthens the bond between people in their communities.

📚 详细文档

模型描述

泰米尔语LLaMA模型在原始LLaMA - 2的基础上进行了增强和定制，拥有约16,000个标记的广泛泰米尔语词汇。

提示模板：ChatML

<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

基准测试结果

使用LLM - Autoeval在runpod的RTX 3090上进行基准测试。

⚠️ 重要提示

请注意，在Open LLM排行榜分数与使用LM Eval Harness在相同配置下本地运行获得的分数之间存在差异。此处提到的结果基于我们自己的基准测试。若要复现这些结果，您可以使用LLM - Autoeval或在本地使用lm - evaluation - harness，并按照Open LLM排行榜“关于”页面中描述的配置进行操作。

相关模型

使用注意事项

⚠️ 重要提示

请注意，这些模型未经过去毒化/审查处理。因此，虽然它们具有出色的语言能力，但有可能生成被认为有害或冒犯性的内容。我们敦促用户谨慎使用，并密切监督模型的输出，特别是在公共或敏感应用场景中。

开发者介绍

认识一下这款创新模型背后的开发者，并关注他们在该领域的贡献：

• Abhinand Balachandran

引用

如果您在研究中使用了该模型或任何与泰米尔语LLaMA相关的工作，请引用：

@misc{balachandran2023tamilllama,
      title={Tamil-Llama: A New Tamil Language Model Based on Llama 2}, 
      author={Abhinand Balachandran},
      year={2023},
      eprint={2311.05845},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

我们希望该模型能成为您自然语言处理工具包中的宝贵工具，并期待看到它在泰米尔语理解和生成方面带来的进步。