Release 5.0.4: LangChain4j integration, fine-tuning API, and Maven reactor#291
Conversation
Bump the reactor (llama-parent + llama + llama-langchain4j) from 5.0.4-SNAPSHOT to 5.0.4, point the README release dependency examples at 5.0.4 (snapshot examples stay 5.0.4-SNAPSHOT), and write the CHANGELOG [5.0.4] section (LangChain4j adapters, LlamaTrainer fine-tuning API, ModelMeta introspection, Maven reactor restructure, llama.cpp b9842->b9859). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QEgh6vtcPWAgQ6scK8pJk7
The "change the root pom.xml <version> only" note was aspirational and
never exercised (the only prior bump predated the llama/ reactor split).
In reality each child hardcodes the parent version in its <parent><version>
pointer and there is no ${revision}/CI-friendly setup, so a version change
must touch all three poms in lockstep or the reactor build fails with
"Could not find artifact net.ladenthin:llama-parent:pom:X.Y.Z". Document the
three edit sites, recommend `mvn versions:set`, and note that ${revision} +
flatten-maven-plugin would be needed for true root-only bumps.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QEgh6vtcPWAgQ6scK8pJk7
|
Review of PR #291 - Release 5.0.4 Version bumps: All three POMs correctly synchronized to 5.0.4 (root + two children). README: All 10 version references updated (5.0.3 → 5.0.4). Snapshot example uses 5.0.4-SNAPSHOT. CHANGELOG: Well-structured with Added/Changed sections. Date correct (2026-07-02). Tracks llama.cpp b9842→b9859. CLAUDE.md: Second commit fixes aspiration to reality—now documents that all three POMs must be bumped in lockstep, provides mvn versions:set remedy, explains the failure mode clearly. Commits: Both follow Conventional Commits. Clear messages. Co-Authored properly. Scope: Surgical diff (POMs, README, CHANGELOG, CLAUDE.md only). Status: No issues found. Approved for merge. |
|



Summary
ChatModel,StreamingChatModel,EmbeddingModel,ScoringModel), experimental fine-tuning API (LlamaTrainer), and model introspection (ModelMeta).llama/module under new aggregator parent POM (llama-parent), alongside newllama-langchain4jmodule. Both inherit a single version and ship in lockstep. Published coordinates (net.ladenthin:llama) unchanged.llama/,llama-langchain4j/) and documentation (README, CHANGELOG).Test plan
Related issues / PRs
Closes the 5.0.4 release.
Checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.mdhttps://claude.ai/code/session_01QEgh6vtcPWAgQ6scK8pJk7