gvisor.dev/gvisor@v0.0.0-20240520182842-f9d4d51c7e0f/images/gpu/ollama/bench/Dockerfile.x86_64 (about) 1 # https://hub.docker.com/r/ollama/ollama 2 FROM ollama/ollama:0.1.26 3 4 ENV PATH=$PATH:/usr/local/nvidia/bin:/bin/nvidia/bin 5 ENV OLLAMA_ORIGINS=* 6 ENV OLLAMA_HOST=0.0.0.0:11434 7 8 COPY pull.sh /tmp 9 10 # Pre-install models useful for benchmarking. 11 # These are huge (total ~120 GiB), but necessary to benchmark 12 # models of various sizes. They are in their own image file to 13 # keep the test-only image lighter by comparison. 14 RUN /tmp/pull.sh codellama:7b-instruct 15 RUN /tmp/pull.sh codellama:34b-instruct 16 RUN /tmp/pull.sh llama2-chinese:7b-chat 17 RUN /tmp/pull.sh llama2:13b-chat 18 RUN /tmp/pull.sh llama2:70b-chat 19 RUN /tmp/pull.sh mistral:7b-instruct 20 RUN /tmp/pull.sh mixtral:instruct 21 RUN /tmp/pull.sh gemma:2b-instruct 22 RUN /tmp/pull.sh gemma:7b-instruct 23 RUN /tmp/pull.sh llava:7b-v1.6 24 RUN /tmp/pull.sh llava:34b-v1.6