Preload tiktoken encoding cache in Docker images

[halfcrazy]

Set TIKTOKEN_CACHE_DIR and warm cl100k_base during image build so runtime token counting does not need to download encoding data in offline environments.

fix #228 #259

[gemini-code-assist]

Code Review

This pull request configures a tiktoken cache directory and preloads the cl100k_base encoding in Dockerfile, Dockerfile-litellm, and Dockerfile-ollama-local. The reviewer suggested setting directory permissions to 777 to prevent permission issues for non-root users and preloading the o200k_base encoding to support newer models like GPT-4o in offline environments.

[gemini-code-assist]

To ensure complete offline support and compatibility with non-root container execution, two improvements should be made here:

1. Permissions: If the container runs as a non-root user (a security best practice), they may encounter a PermissionError when accessing or writing to the cache directory (e.g., if they need to cache other encodings at runtime). Setting the directory permissions to 777 ensures compatibility with any runtime user.
2. GPT-4o Support (o200k_base): Modern OpenAI models like gpt-4o and gpt-4o-mini use the o200k_base encoding. Preloading both cl100k_base and o200k_base ensures that offline environments do not attempt to download encoding data at runtime for these newer models.

RUN mkdir -p "$TIKTOKEN_CACHE_DIR" && \
    chmod -R 777 "$TIKTOKEN_CACHE_DIR" && \
    python -c "import tiktoken; tiktoken.get_encoding('cl100k_base'); tiktoken.get_encoding('o200k_base')"

[gemini-code-assist]

To ensure complete offline support and compatibility with non-root container execution, two improvements should be made here:

1. Permissions: If the container runs as a non-root user (a security best practice), they may encounter a PermissionError when accessing or writing to the cache directory (e.g., if they need to cache other encodings at runtime). Setting the directory permissions to 777 ensures compatibility with any runtime user.
2. GPT-4o Support (o200k_base): Modern OpenAI models like gpt-4o and gpt-4o-mini use the o200k_base encoding. Preloading both cl100k_base and o200k_base ensures that offline environments do not attempt to download encoding data at runtime for these newer models.

RUN mkdir -p "$TIKTOKEN_CACHE_DIR" && \
    chmod -R 777 "$TIKTOKEN_CACHE_DIR" && \
    python -c "import tiktoken; tiktoken.get_encoding('cl100k_base'); tiktoken.get_encoding('o200k_base')"

[gemini-code-assist]

To ensure complete offline support and compatibility with non-root container execution, two improvements should be made here:

1. Permissions: If the container runs as a non-root user (a security best practice), they may encounter a PermissionError when accessing or writing to the cache directory (e.g., if they need to cache other encodings at runtime). Setting the directory permissions to 777 ensures compatibility with any runtime user.
2. GPT-4o Support (o200k_base): Modern OpenAI models like gpt-4o and gpt-4o-mini use the o200k_base encoding. Preloading both cl100k_base and o200k_base ensures that offline environments do not attempt to download encoding data at runtime for these newer models.

RUN mkdir -p "$TIKTOKEN_CACHE_DIR" && \
    chmod -R 777 "$TIKTOKEN_CACHE_DIR" && \
    python -c "import tiktoken; tiktoken.get_encoding('cl100k_base'); tiktoken.get_encoding('o200k_base')"