releasing 1.4.0

[keighrim]

Overview

This release introduces three new fields to the app metadata to improve the orchestration and handling of large models within apps.

Additions

• analyzer_versions: A new app metadata field (complementing the existing analyzer_version) designed to store version strings for multiple models wrapped in a single CLAMS app (see analyzer_version when using HF model family #251 (comment) for discussion).
• est_gpu_mem_min and est_gpu_mem_typ: New app metadata fields that allow developers to specify estimated GPU memory usage (minimum and typical) for better resource management.

Changes

• Updated to the latest mmif-python SDK (1.2.1)
• Profiling storage: Runtime profiling results (CPU/CUDA architecture, running time) are now stored under a new appProfiling field in the output view metadata (Refactor metadata structure to nest profiling and runtime metrics #261).
  • Note: This is an experimental field intended primarily for human inspection; its internal structure may change without notice.
• View timestamps: The timestamp property of views is now updated based on the time when the _annotate() call completes. This ensures all views from a single app execution share a consistent "marker" for easier identification (Implement "run ID" property for views to distinguish app executions #269).
• Running an app that declares an est_gpu_mem_min value in production mode (--production) will now automatically limit the number of Gunicorn workers spawned to prevent GPU out-of-memory (OOM) errors.
• VRAM usage record: Apps running on CUDA and torch now cache VRAM usage statistics in the local disk cache (typically in $XDG_CACHE_HOME) (gunicorn, torch, and cuda #243). Currently, this information is for logging only, and we plan to add a retrieval API in the future.
  • See current private implementation at

    clams-python/clams/app/__init__.py

    Lines 348 to 361 in 671560b

    	def _get_profile_path(self, param_hash: str) -> pathlib.Path:
    	"""
    	Get filesystem path for memory profile file.
    	Profile files are stored in a per-app directory under user's cache.
    	:param param_hash: Hash of parameters from :func:`mmif.utils.cli.describe.generate_param_hash`
    	:return: Path to the profile file
    	"""
    	# Sanitize app identifier for filesystem use
    	app_id = self.metadata.identifier.replace('/', '-').replace(':', '-')
    	cache_base = pathlib.Path(os.environ.get('XDG_CACHE_HOME', pathlib.Path.home() / '.cache'))
    	cache_dir = cache_base / 'clams' / 'memory_profiles' / app_id
    	return cache_dir / f"memory_{param_hash}.json"