Added the ability to Modify the Context Length

[psych0v0yager]

Fixes issue #159

You can now specify how much context you want the model to have.

For example Mixtral 8x7b AWQ:

python -m sglang.launch_server --model-path /path/to/bagel_mixtral_AWQ --port 30000 --tp 2

Rank 1: max_total_num_token=135505, max_prefill_num_token=32768, context_len=32768, model_mode=[]
Rank 0: max_total_num_token=135505, max_prefill_num_token=32768, context_len=32768, model_mode=[]

With the adjustment

python -m sglang.launch_server --model-path /path/to/bagel_mixtral_AWQ --port 30000 --tp 2 --context-length 8192

Rank 0: max_total_num_token=135505, max_prefill_num_token=22584, context_len=8192, model_mode=[]
Rank 1: max_total_num_token=135505, max_prefill_num_token=22584, context_len=8192, model_mode=[]

[comaniac]

LGTM