[Bug] vLLM server fails when requests with different temperatures are sent

[cglagovichTT]

Anything you want to discuss about vllm.

You can repro by starting up the server example and sending requests with different temperatures. Failure should look like

    async for i, res in result_generator:
  File "/home/cglagovich/vllm/vllm/utils.py", line 506, in merge_async_iterators
    item = await d
  File "/home/cglagovich/vllm/vllm/engine/multiprocessing/client.py", line 598, in _process_request
    raise request_output
  File "/home/cglagovich/vllm/vllm/engine/multiprocessing/client.py", line 598, in _process_request
    raise request_output
AssertionError: Currently only supporting same temperature for all sequences in batch```