chore: return model size after pulled

[louis-menlo]

Describe Your Changes

• Clients like Jan view the model's size as metadata for local models. This helps users determine compatibility with their computer specs.
• This PR saves the model's size as metadata (in bytes).

model.yaml

API Response body

API Playground

Changes made

The provided git diff introduces changes primarily focused on tracking and handling a new attribute called size within the ModelConfig struct. Here are the key changes summarized:

1. ModelConfig Struct Update:

   • A new field uint64_t size is added to the ModelConfig struct to store the size of the model.
   • FromJson method is updated to parse the size attribute from JSON if present.
   • ToJson method is updated to include the size attribute in the JSON output.
   • Print method now prints the size attribute along with other attributes.

2. YamlHandler Update:

   • In UpdateModelConfig, the YAML node is updated to include the size of the model if it is not NaN.
   • Likely missed the context, but casting size of type uint64_t to double for NaN check is unusual since uint64_t can't be NaN.

3. ParseGguf Function Update:

   • The function signature of ParseGguf is updated to include an optional std::uint64_t size parameter.
   • The size is set on model_config with a default of 0 if no size is provided.

4. Model Service Update:

   • In two functions (HandleDownloadUrlAsync and HandleUrl), after a download task is finished, the total downloaded size (model_size) is calculated by summing up bytes from all items in finishedTask.items.
   • ParseGguf is called with the computed model_size to ensure the size is recorded in the ModelConfig.

These changes enhance the system to manage and persist the size of model files across configurations and YAML updates.

[github-actions]

Preview URL: https://f638fa2b.cortex-docs.pages.dev