prompt generator 0.1.0

Updates:
vLLM backend:

• added support for inferencing models on multiple GPUs;
• added extra parameters for controlling the process of generation;
• added support for speculative decoding (https://docs.vllm.ai/en/latest/models/spec_decode.html, experimental vLLM feature);
• added several optimizations;

Other:

• improved instruction prompt;
• adjusted object categories from where objects are being sampled;
• bug fixes

prompt generator 0.0.0

Updates:

• Reworked structure of the project. Now it is easy to add new backends if needed;
• Code is cleaner;
• Better instruction prompt;
• More categories that are used for prompt generation;