85 7 months ago

tools

7 months ago

69fd7024ab53 ยท 4.7GB ยท

qwen2
ยท
7.62B
ยท
Q4_K_M
{{- if .Suffix }}<|fim_prefix|>{{ .Prompt }}<|fim_suffix|>{{ .Suffix }}<|fim_middle|> {{- else if .M

Readme

DeepSeek-7B-1M

DeepSeek-7B-1M is a hybrid model combining Qwen2.5-1.5B-1M and DeepSeek-R1-Distill-Qwen-1.5B, designed for enhanced reasoning, long-context understanding, and structured output generation. This model is optimized for mathematical problem-solving, code generation, and natural language understanding.

๐Ÿš€ Features

  • Combines the strengths of Qwen2.5-1.5B-1M and DeepSeek-R1-Distill-Qwen-1.5B
  • Improved mathematical reasoning and structured output generation
  • Extended context length handling up to 32K tokens
  • Optimized for logical inference, code completion, and technical problem-solving
  • Supports multi-turn conversations with enhanced coherence

๐Ÿ“ฅ Installation

1๏ธโƒฃ Install Ollama

If Ollama is not installed, install it using:

For macOS & Linux:

curl -fsSL https://ollama.com/install.sh | sh

For Windows (WSL required):

wsl --install
curl -fsSL https://ollama.com/install.sh | sh

For more details, check the official Ollama installation guide:
https://ollama.com/download


๐Ÿ”ง Running the Model

2๏ธโƒฃ Pull the Model

After installing Ollama, download DeepSeek-7B-1M:

ollama pull myrepo/deepseek-7b-1m

3๏ธโƒฃ Run the Model

To start generating responses:

ollama run myrepo/deepseek-7b-1m

For interactive chat mode:

ollama chat myrepo/deepseek-7b-1m

๐Ÿ“Œ Customizing the Model

Using a Custom Modelfile

You can adjust model behavior using a Modelfile.

  1. Create a new file named Modelfile and add the following:
FROM myrepo/deepseek-7b-1m

PARAMETER temperature 0.7
PARAMETER top_p 0.9

SYSTEM "You are an AI expert trained for advanced reasoning, coding, and mathematical problem-solving. Provide detailed, structured, and optimized responses."
  1. Build and run your custom model:
ollama create deepseek-7b-custom -f Modelfile
ollama run deepseek-7b-custom

๐ŸŽฏ Performance Optimization

For the best performance: - Run Ollama on a GPU-enabled system - Use quantized versions (e.g., fp16, int4) for efficient inference - Deploy on high-memory cloud instances (32GB RAM or more)


๐Ÿ“„ License

This model is released under the MIT License, ensuring open and accessible AI development.


๐Ÿ”— Resources

  • Hugging Face Repository: (if applicable)
  • Ollama Documentation: https://ollama.com/docs
  • GitHub Repository: (if applicable)

๐Ÿ™Œ Acknowledgments

DeepSeek-7B-1M is built by merging Qwen2.5-1.5B-1M and DeepSeek-R1-Distill-Qwen-1.5B, leveraging their strengths in math, structured reasoning, and technical NLP.