mirror of
https://github.com/Mesh-LLM/mesh-llm.git
synced 2026-08-08 22:23:19 -04:00
4bd1453fad
* feat(skippy): expose authoritative generation lifecycle * refactor(mesh): remove injected local model launcher * fix(skippy): make serving hooks independent * feat(mesh): load deadline-safe native serving plugins (#1155) * feat(openai): propagate stable agent session identity * feat(mesh): load deadline-safe native serving plugins * docs: document native serving integrations * fix(skippy): preserve native proposal query contract * fix(skippy): compile native proposal plugin tests * fix(plugin): use public proposal query constructor * fix(skippy): make lifecycle delivery nonblocking * refactor(skippy): add lifecycle ingress adapter * fix: address generation lifecycle review --------- Co-authored-by: Nick DiZazzo <nick.dizazzo@gmail.com>
119 lines
3.9 KiB
TOML
119 lines
3.9 KiB
TOML
[workspace]
|
|
members = [
|
|
"crates/mesh-llm",
|
|
"crates/mesh-llm-cli",
|
|
"crates/mesh-llm-commands",
|
|
"crates/mesh-llm-config",
|
|
"crates/mesh-llm-events",
|
|
"crates/mesh-llm-build-info",
|
|
"crates/mesh-llm-gpu-bench",
|
|
"crates/mesh-llm-host-runtime",
|
|
"crates/mesh-llm-hardware-profile",
|
|
"crates/mesh-llm-identity",
|
|
"crates/mesh-llm-native-runtime",
|
|
"crates/mesh-llm-protocol",
|
|
"crates/mesh-llm-release-footer",
|
|
"crates/mesh-llm-routing",
|
|
"crates/mesh-llm-runtime-install",
|
|
"crates/mesh-llm-sdk",
|
|
"crates/mesh-llm-guardrails",
|
|
"crates/mesh-llm-system",
|
|
"crates/mesh-llm-tui",
|
|
"crates/mesh-llm-types",
|
|
"crates/mesh-llm-console-server",
|
|
"crates/mesh-llm-embedded-runtime",
|
|
"crates/mesh-llm-ui",
|
|
"crates/mesh-llm-plugin",
|
|
"crates/mesh-llm-skills",
|
|
"crates/mesh-llm-plugin-manager",
|
|
"crates/mesh-native-serving-plugin-api",
|
|
"crates/mesh-native-serving-plugin-host",
|
|
"crates/mesh-client",
|
|
"crates/mesh-llm-api-client",
|
|
"crates/mesh-llm-api-server",
|
|
"crates/mesh-llm-node",
|
|
"crates/mesh-llm-ffi",
|
|
"crates/mesh-llm-nodejs",
|
|
"crates/mesh-llm-test-harness",
|
|
"crates/model-ref",
|
|
"crates/model-artifact",
|
|
"crates/model-hf",
|
|
"crates/model-resolver",
|
|
"crates/skippy-protocol",
|
|
"crates/skippy-coordinator",
|
|
"crates/skippy-topology",
|
|
"crates/skippy-cache",
|
|
"crates/skippy-metrics",
|
|
"crates/openai-frontend",
|
|
"crates/skippy-ffi",
|
|
"crates/skippy-runtime",
|
|
"crates/skippy-server",
|
|
"crates/metrics-server",
|
|
"crates/skippy-model-package",
|
|
"crates/skippy-quantize",
|
|
"crates/model-package",
|
|
"crates/skippy-correctness",
|
|
"crates/llama-spec-bench",
|
|
"crates/skippy-bench",
|
|
"crates/skippy-prompt",
|
|
"crates/mesh-mixture-of-agents",
|
|
"crates/llama-quant-ffi",
|
|
"tools/xtask",
|
|
]
|
|
default-members = [
|
|
"crates/mesh-llm",
|
|
"crates/mesh-llm-plugin",
|
|
]
|
|
resolver = "2"
|
|
|
|
[workspace.package]
|
|
edition = "2024"
|
|
license = "MIT OR Apache-2.0"
|
|
version = "0.72.1"
|
|
|
|
[workspace.dependencies]
|
|
ahash = "0.8.12"
|
|
anyhow = "1"
|
|
blake3 = "1"
|
|
clap = { version = "4", features = ["derive"] }
|
|
mesh-llm-build-info = { path = "crates/mesh-llm-build-info", version = "0.72.1" }
|
|
mesh-llm-release-footer = { path = "crates/mesh-llm-release-footer", version = "0.72.1" }
|
|
mesh-llm-skills = { path = "crates/mesh-llm-skills", version = "0.72.1" }
|
|
serde = { version = "1", features = ["derive"] }
|
|
serde_json = "1"
|
|
sha2 = "0.10"
|
|
strum = { version = "0.28", features = ["derive"] }
|
|
|
|
[workspace.lints.clippy]
|
|
cognitive_complexity = "warn"
|
|
too_many_lines = "warn"
|
|
|
|
# Do not remove this section: direct GGUF debug startup spends most of its time
|
|
# hashing the source model via sha2/sha2-asm before native skippy model open,
|
|
# so keep those crates near release speed in dev builds.
|
|
[profile.dev.package.sha2]
|
|
opt-level = 3
|
|
|
|
[profile.dev.package.sha2-asm]
|
|
opt-level = 3
|
|
|
|
# Tightened release profile for shipped binaries.
|
|
#
|
|
# - `lto = "thin"`: parallel link-time optimization across the crate graph.
|
|
# Roughly 80% of fat-LTO's binary-size and perf benefit at ~30% of the
|
|
# link cost. Fat LTO would shave a few more MB but pushes release link
|
|
# from ~30 s to several minutes on this codebase.
|
|
# - `codegen-units = 1`: pairs with thin LTO so the optimizer can inline
|
|
# across the whole crate without re-link boundaries. Trades parallelism
|
|
# in codegen for output quality; cargo still parallelises the dep graph.
|
|
# - `strip = "debuginfo"`: drops DWARF from the Mach-O / ELF output. We
|
|
# keep function symbols so backtraces from production nodes remain
|
|
# readable (`nm`, `addr2line` on the symbol table still works). DWARF
|
|
# on macOS arm64 alone is several MB of dead weight in the binary.
|
|
# - panic stays at the default ("unwind"). Plugin / MCP error recovery
|
|
# relies on unwinding; the AGENTS.md notes explicitly forbid switching
|
|
# to `abort`.
|
|
[profile.release]
|
|
lto = "thin"
|
|
codegen-units = 1
|
|
strip = "debuginfo"
|