Fix vllm models - #49

  olmes \
      --model "allenai/OLMo-2-0425-1B-Instruct" \
      --model-args '{"chat_model": true, "max_length": 4096, "metadata": {}, "trust_remote_code": true, "gpu_memory_utilization": 0.85}' \
      --model-type vllm \
      --task gsm8k::tulu ifeval::tulu \
      --task-args '{ "generation_kwargs": { "max_gen_toks": 4096, "truncate_context": false } }' \
      --output-dir "eval_results"