llmff v0.1.1

Second public early-testing release of llmff, an FFmpeg-shaped CLI and Rust library for typed LLM inference pipelines.

Supported Install

cargo install --git https://github.com/syndicalt/llmff --tag v0.1.1 llmff

Supported Platforms

• Linux and macOS with a working Rust/Cargo toolchain.
• Windows is not blocked by design, but this release has not been verified there.

Included Since v0.1.0

• llmff inspect -g dry-run validation for compact inline graphs.
• Inline retrieve(documents=...,top_k=...) stages for local lexical retrieval.
• Inline cache(path=...,key=...) stages for reproducible cache use in CLI pipelines.

Existing Capabilities

• YAML manifest pipeline runner.
• Compact inline graph runner for linear pipelines.
• Built-in stages: load, cache, system, template, retrieve, infer, validate_json, repair, route, tool, and write.
• Mock backends for deterministic tests and examples.
• OpenAI-compatible and Ollama chat backend adapters.
• CLI backend registration with optional API key environment lookup.
• Sequential execution by default and --parallel execution for independent ready stages.
• JSONL traces plus llmff trace summaries.
• GitHub install smoke gate.

Known Limitations

• This is a pipeline runner, not a native inference kernel, model converter, or serving platform.
• No binary artifacts are published for this release.
• Backend calls are non-streaming.
• Inline graphs support linear pipelines only; branching pipelines should use YAML manifests.
• Retrieval is local lexical matching, not vector search.
• Embeddings, reranking, multimodal inputs, native model loading, and plugin loading are not included yet.

Verification

Release verification should pass:

scripts/smoke-install.sh --git https://github.com/syndicalt/llmff --tag v0.1.1