Structured observability for the iroh/SWIM layer: Aggregator, typed Event/Snapshot
types, Sink (NoopSink default), ProbeScheduler, process stats, and host/iroh/swim
introspection, plus the swactor-diag-collector, -postproc, and -iroh-relay binaries
that assemble and render per-run bundles. Generalizes the pipeline-parallel-inference
example to N stages and adds the topology-planner spec.
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Extend the single-GPU example into a two-node pipeline-parallel run that splits llama3.2:1b across two rented vast.ai GPUs and closes the autoregressive decode loop over iroh.
- topology: add linear-chain helpers where each stage derives its neighbours locally from `STAGE`/`NUM_STAGES`, registering `pp-entry`/`pp-exit`/`pp-stage-{i}` SWIM names
- messages: add `StageActivation` (bf16 hidden-state hand-off carrying position/seq_len/is_prefill) and `NextToken` (sampled-token feedback with a `done` flag) that close the autoregressive loop between stage 0 and stage 1
- stage_actor: add `Stage0Actor` (tokenize -> embed_and_forward -> prefill activation; decode_step on each NextToken) and `Stage1Actor` (forward_and_sample -> NextToken back; emit InferenceResponse on EOS/max_tokens)
- vastai: fork the client and add `create_pipeline_instances` (rents one instance per stage, threading `STAGE`/`NUM_STAGES`, best-effort destroys on partial failure) and `destroy_all_instances`
- pp_tinygrad_worker.py: per-stage worker slicing `model.blk[start:end]` in stub and real (GGUF) modes, plus new `pp_gpu_node`/`pp_smoke_run` binaries and ROADMAP/SPEC/TEST_SPEC docs
- reuse: build on the single-GPU example's iroh transport and process bridge unchanged; add actor/codec/topology/integration test suites
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Add a complete single-GPU distributed-inference example that rents a vast.ai GPU, boots a worker container, and runs a prompt end-to-end over iroh/SWIM.
- examples/single-gpu-inference: add the `single_gpu_inference` orchestrator binary that starts a local iroh node, waits for the remote gpu-node to register the `"inference"` SWIM name, then sends an `InferenceRequest` and prints the response
- examples/single-gpu-inference: add the `gpu_node` binary that joins the cluster via `SEED_ADDR`, spawns an `InferenceActor` over `tinygrad_worker.py`, and registers the `"inference"` bridge
- inference_actor: bridge swactor messaging to a Python child process via stdin/stdout JSON, with `ProcessBridge`/`RequestBridge` adapters that satisfy the single-`Incoming` actor constraint
- iroh_transport: add `IrohActorTransport` that sends `WireEnvelope`s over iroh QUIC uni-streams (connection-cached against early close), plus wire encode/decode and an inbound drain helper
- vastai: add a vast.ai REST client (`find_offer` with reliability/cuda/geo filters excluding CN, `create_instance`, `wait_for_running`, `destroy_instance`) parameterised by a mockable `base_url`
- worker/docs/tests: ship `tinygrad_worker.py`/`echo_worker.py` (newline-JSON, `--stub`/`--model` defaulting to llama3.2:1b), a Dockerfile, Makefile, SPEC, and actor/codec/cluster/integration/vastai test suites
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Remove co-dependencies for different modules found in `crates` and migrate the development history to a new repository. The docs were stale, and largely not getting used, so simply deleted for now. When code stabilizes more, they will become useful again.
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Crates continued to grow in number, but most are still quite small and feature specific. This refactor consolidates them.
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Add a Python getting-started Jupyter notebook and reorganize the Python examples under examples/python/.
- examples/python/getting_started.ipynb: add notebook demonstrating the single-threaded tick loop (`spawn`/`send`/`tick`/`inbox`/`try_recv`) and the multi-threaded path via `RuntimeConfig` + `rt.run()`/`handle.shutdown()`
- examples/: relocate `hello_async.py` and `hello_single_thread.py` under `examples/python/`
- pyproject.toml: add a `dev` dependency group containing jupyter and ipykernel
- uv.lock: regenerate the lockfile for the new dev dependencies (2452 lines)
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Show basic information from the runtime, such as number of actors, worker threads, etc.
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Refactoring to logically separate component modules in order to make it easier to develop tests, metrics, tracing, etc.
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Did not like the requirement of a tokio runtime and async/await. Rewrote
to use lock free queues (`crossbeam_queue::ArrayQueue`) as the basic
primitive to enable a runtime consisting of many pseudo-processes that
can pass messages to each other.
Reviewed-on: http://zachery.lol/code/code/zacheryasc/swactor/pulls/1
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Introduce a typed `ActorRequestSender` connection handle for actors and harden the `Error` type.
- src/lib.rs: add `ActorRequestSender<A>` wrapping `mpsc::Sender<ActorRequest<A>>` with async `send`, `Clone`, and `From` impls; expose it via new `Handle::get_connection()` so callers hold a lightweight standalone connection to an actor
- src/lib.rs: route `Handle::send` through the new sender and store `tx` as an `ActorRequestSender`; drop the `Unpin` supertrait bound from the `Actor` trait
- src/error.rs: turn the `Error` type alias into a newtype struct, gate `convert_err` as `pub(crate)`, and add `From<T: AsRef<str>>` plus `ToString` impls
- examples/hello.rs: switch `Greeter::spawn` to method-call syntax (`Greeter.spawn(&rt)`) to match the updated API
- README.md: rename the project heading from "about" to "swactor"
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
Skeletal actor framework. Somewhat unweildy, needs a message box, a better runtime, and different channels. However, hello world example works
Signed-off-by: Zachery Aaron Shores-Chmielewski <zacheryasc@gmail.com>
