metadata

license: other
license_name: deepseek
license_link: LICENSE
datasets:
  - unalignment/spicy-3.1

[🏠Homepage] | [🤖 Chat with DeepSeek LLM] | [Discord] | [Wechat(微信)]

Fine-tune of Deepseek 67B

Fine-tuned with jondurbin's unalignment/spicy-3.1 for 1 epoch.

1. Introduction of Deepseek LLM

Introducing DeepSeek LLM, 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. In order to foster research, we have made DeepSeek LLM 7B/67B Base and DeepSeek LLM 7B/67B Chat open source for the research community.

2. Model Summary

deepseek-llm-67b-base is a 67B parameter model with Grouped-Query Attention trained on 2 trillion tokens from scratch.

Home Page: DeepSeek
Repository: deepseek-ai/deepseek-LLM
Chat With DeepSeek LLM: DeepSeek-LLM

3. How to Use

Here give some examples of how to use our model.

Text Completion

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig

model_name = "deepseek-ai/deepseek-llm-67b-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, where the query, keys, values, and output are all vectors. The output is"
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)

4. License

This code repository is licensed under the MIT License. The use of DeepSeek LLM models is subject to the Model License. DeepSeek LLM supports commercial use.

See the LICENSE-MODEL for more details.

5. Contact

If you have any questions, please raise an issue or contact us at [email protected].