feat: integrate Jetson-PI (public repo) as Pi0/LLM/MLLM provider#148

Open

heiheiha798 wants to merge 34 commits into

flashrt-project:mainfrom

heiheiha798:feat/jetson-pi-provider

Commits on Jul 15, 2026

Add provider-owned model runtime v2 stages
heiheiha798
committed
Add initial llama.cpp Pi0 provider boundary
heiheiha798
committed
Add llama.cpp Pi0 engine factory open path
heiheiha798
committed
cpp: add Jetson-PI llama.cpp/mtmd link-check provider gate

heiheiha798
and
claude
committed
cpp: real Jetson-PI Pi0 engine + end-to-end smoke test (Phase 1)

heiheiha798
and
claude
committed
cpp: Phase 1 wrap-up — multi-tick + dim-mismatch tests, engine contract docs

heiheiha798
and
claude
committed
python: load_model(framework="jetson_pi") drives the Jetson-PI Pi0 provider (Phase 2)

heiheiha798
and
claude
committed
cpp+python: generic GGUF LLM provider (Phase 3)

heiheiha798
and
claude
committed
cpp+python: multimodal LLM provider (Phase 4)

heiheiha798
and
claude
committed
docs: Phase 5 finer-Pi0-stages evaluation — DEFER (no-go under current conditions)

heiheiha798
and
claude
committed
docs: Phase 6 backend-vtable evaluation — GO-MINIMAL-CONTRACT

heiheiha798
and
claude
committed
cpp+python: Phase 6 memory-domain contract (minimal, append-only)

heiheiha798
and
claude
committed
cpp: expose FLASHRT_PI0_BACKEND in Pi0 e2e test for real CUDA run

heiheiha798
and
claude
committed
python: verify Jetson-PI Pi0 backend="cuda" end-to-end on GPU2

heiheiha798
and
claude
committed
docs: clarify pi0_base-only weights + mark mllm citations as historical

heiheiha798
and
claude
committed
python: verify Jetson-PI LLM + MLLM backend="cuda" end-to-end on GPU2

heiheiha798
and
claude
committed
cpp: wire a real Jetson-PI token consumer into the Phase 6 contract

heiheiha798
and
claude
committed
cpp: numerical-parity tests — FlashRT provider vs direct jetson_pi_* (§14)

heiheiha798
and
claude
committed
docs: Phase 7 eval — LLM repeatable decode (TODO-2) DEFER (no-go under current conditions)

heiheiha798
and
claude
committed
docs: verify Jetson-PI Vulkan backend (Pi0 + LLM) on RTX 4090 (TODO-3)

heiheiha798
and
claude
committed
build: add production Jetson-PI package integration
heiheiha798
committed
feat: expose repeatable LLM and MLLM stages
heiheiha798
committed
feat: expose Pi0 context and action stages
heiheiha798
committed
feat: add host SWAP and DLPack action views
heiheiha798
committed
docs: complete accelerator backend integration
heiheiha798
committed
feat: fingerprint Jetson-PI checkpoint files
heiheiha798
committed
test: complete Jetson-PI validation gates
heiheiha798
committed
test: validate staged decode interruption limits
heiheiha798
committed
test: enforce read-only DLPack action exports
heiheiha798
committed
feat: complete Jetson-PI migration validation
heiheiha798
committed
docs: correct Vulkan teardown validation
heiheiha798
committed
docs: record merge-branch migration validation
heiheiha798
committed
docs: 切换 Jetson-PI 后端指向公开 repo + 修正 cpp 测试 guard 注释
 heiheiha798
committed

Commits on Jul 16, 2026

docs(jetson_pi): 在 usage 文档开头加 Jetson-PI 论文与定位说明
 heiheiha798
committed