releasing 1.7.0

[keighrim]

Overview

This release introduces a specialized base class ClamsPromptableApp and its HuggingFace specialization ClamsHFPromptableApp for CLAMS apps that wrap LLM / VLM / ALM / LMM-based analyzers, along with other HF-related helper code in various places. Additionally, it adds appTags field for app categorization in app metadata, and new timeframe related local utility recipe for clams develop --recipe utl-tf scaffold.

Additions

• ClamsPromptableApp, a new specialized base class (subclassing ClamsApp) for prompt-driven, instruction-tuned LLM apps with pre-equipped with commonly used runtime parameter set (prompt, systemPrompt, promptMode, maxNewTokens, temperature, topP, topK, parallelPrompts). Plus, ClamsHFPromptableApp (subclassing ClamsPromptableApp) for the common case of "promptable app + local HuggingFace transformers model". This subclass should significanly simplify wrapping HF models into a CLAMS app. See https://clams.ai/clams-python/app-baseclasses.html for the developer guide. (helper class/modules for wrapping promptable LLMs #291)
• clams.backends.hf (new module) with two general HF loaders: load_hf_model (from_pretrained flow) and load_hf_pipeline (transformers.pipeline flow). Wraps the device auto-detection, revision pinning, dtype handling, and kwargs pass-through that every HF-backed CLAMS app does identically; used internally by ClamsHFPromptableApp and callable directly from any non-promptable HF app. Requires the [hf] extra (pip install clams-python[hf]) to keep torch and transformers out of the base install.
• clams develop --recipe utl-tf scaffold drops a utils/timeframe.py into the new app that factors out the canonical TimeFrame iteration / vdh.extract_images_by_mode_with_sources / TimePoint-on-fallback / task-tuple pattern. (7bc91d9)
• appTags field on AppMetadata, a sibling to added appTags field to view metadata mmif#253. Allows categorization of CLAMS apps for app-directory search / filtering. (appTags field and helper #290)

Changes

• Updated to the latest mmif-python SDK (1.5.0).
• Tightened the semantics of the AppMetadata.analyzer_versions field (introduced in 1.4.0) for CHFPApp-based development. When an app subclasses ClamsHFPromptableApp, the dict is now the source of truth for the family of supported HF model IDs and their pinned commit hashes: the SDK auto-derives a model runtime parameter from the dict's keys, and the HF base class refines model=<id> into model=<id>@<revision> on output so the resolved revision lands in view.metadata.appConfiguration automatically.
• clams develop scaffold templates (app.py.template, metadata.py.template) updated with commented-out blocks for both promptable variants.

[codecov]

Codecov Report

❌ Patch coverage is 62.04082% with 93 lines in your changes missing coverage. Please review.
✅ Project coverage is 63.85%. Comparing base (85ae527) to head (f391efd).
⚠️ Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
clams/backends/hf.py	5.45%	52 Missing ⚠️
clams/app/__init__.py	79.77%	36 Missing ⚠️
clams/develop/__init__.py	0.00%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #292      +/-   ##
==========================================
- Coverage   64.48%   63.85%   -0.64%     
==========================================
  Files           8        9       +1     
  Lines         887     1126     +239     
==========================================
+ Hits          572      719     +147     
- Misses        315      407      +92     

Flag	Coverage Δ
unittests	63.85% <62.04%> (-0.64%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.