[Feature] Support Mixtral 8x7b

[pppppM]

Due to the lack of official dialogue templates from Mixtral, we use InternLM's dialogue templates for its SFT fine-tuning.

# QLoRA (only need a single A100-80G)
xtuner train mixtral_8x7b_qlora_oasst1_internlm_template_e3 --deepspeed deepspeed_zero2


#Full Parameters（with slurm）
srun ${SRUN_ARGS} xtuner train mixtral_8x7b_full_oasst1_internlm_template_e3 --deepspeed deepspeed_zero3 --launcher slurm

[LZHgrla]

Thanks! I left one comment about the requirements/runtime.txt

Others are LGTM!