What are the main differences between different versions of the DeepSeek model (such as 1.5B, 7B, 8B, 14B, 32B, 70B, 671B)?

The larger the number of parameters, the more complex the model is usually, and the stronger its performance, but the computational resource requirements and training costs are also higher.
The following are the main differences between the versions:

1.5B

Parameter quantity: 1.5 billion

Features: Lightweight, suitable for scenarios with limited resources or low performance requirements, fast inference speed, but average performance for complex tasks.

7B

Parameter quantity: 7 billion

Features: Medium scale, suitable for most general tasks, good performance, moderate resource requirements.

8B

Parameter quantity: 8 billion

Features: Similar to 7B, with slightly improved performance, suitable for tasks that require slightly stronger abilities.

14B

Parameter quantity: 14 billion

Features: Further improved performance, suitable for more complex tasks, high resource requirements.

32B

Parameter quantity: 32 billion

Features: High performance, suitable for complex tasks, high resource requirements, high training and inference costs.

70B

Parameter quantity: 70 billion

Features: Approaching top-level performance, suitable for complex tasks with high demands, and with extremely high resource requirements.

671B

Parameter quantity: 671 billion

Features: Top performance, suitable for the most complex tasks, extremely high resource requirements, huge training and inference costs.