openeuler/deepseek repository overview

⁠Quick reference

• The official DeepSeek-LLM docker image.

• Maintained by: openEuler CloudNative SIG⁠.

• Where to get help: openEuler CloudNative SIG⁠, openEuler⁠.

⁠DeepSeek-LLM | openEuler

Current DeepSeek-LLM docker images are built on the openEuler⁠. This repository is free to use and exempted from per-user rate limits.

DeepSeek LLM is an advanced language model comprising 67 billion parameters. It has been trained from scratch on a vast dataset of 2 trillion tokens in both English and Chinese. DeepSeek LLM 7B/67B Base and DeepSeek LLM 7B/67B Chat are open source for the research community.

Learn more on DeepSeek-LLM⁠.

⁠Supported tags and respective Dockerfile links

The tag of each DeepSeek-LLM docker image is consist of the version of DeepSeek-LLM and the version of basic image. The details are as follows

⁠Usage

Here, users can select the corresponding {Tag} by their requirements.

• Pull the openeuler/deepseek image from docker

  docker pull openeuler/deepseek:{Tag}
  

  Copy

• Run and test deepseek container

  docker run -it --rm openeuler/deepseek:{Tag}
  

  Copy

  From here, users can test DeepSeek-LLM as follows:

  Text Completion:

  import torch
  from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
  
  model_name = "deepseek-ai/deepseek-llm-7b-base"
  tokenizer = AutoTokenizer.from_pretrained(model_name)
  model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
  model.generation_config = GenerationConfig.from_pretrained(model_name)
  model.generation_config.pad_token_id = model.generation_config.eos_token_id
  
  text = "An attention function can be described as mapping a query and a set of key-value pairs to an output"
  inputs = tokenizer(text, return_tensors="pt")
  outputs = model.generate(**inputs.to(model.device), max_new_tokens=100)
  
  result = tokenizer.decode(outputs[0], skip_special_tokens=True)
  print(result)
  

  Copy

  Chat Completion:

  import torch
  from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
  
  model_name = "deepseek-ai/deepseek-llm-7b-chat"
  tokenizer = AutoTokenizer.from_pretrained(model_name)
  model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
  model.generation_config = GenerationConfig.from_pretrained(model_name)
  model.generation_config.pad_token_id = model.generation_config.eos_token_id
  
  messages = [{"role": "user", "content": "Who are you?"}]
  input_tensor = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
  outputs = model.generate(input_tensor.to(model.device), max_new_tokens=100)
  
  result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
  print(result)
  

  Copy

⁠Question and answering

If you have any questions or want to use some special features, please submit an issue or a pull request on openeuler-docker-images⁠.