feat(phi4): run Phi-4-mini-instruct Q8_0 GGUF on AIE4 via ryzenai-corelib - #1

Open

chizamd wants to merge 38 commits into

feature/phi4-gguf-aie4-medusa

Commits on Sep 11, 2026

docs: design Phi-4 GGUF AIE4 integration
chizamd
committed
df83bf6
docs: plan Phi-4 GGUF AIE4 integration
chizamd
committed
4948126
build: add optional dynamic corelib 0.3.0 runtime
chizamd
committed
d1b0a05
test: strengthen corelib integration guards
chizamd
committed
1f67adb
feat: add validated Phi-4 Q8_0 GGUF reader
chizamd
committed
f2c1dba
fix: harden Phi-4 GGUF validation
chizamd
committed
5e6423b
test: assert exact Phi-4 mismatch diagnostics
chizamd
committed
804b856
feat: add corelib-backed Phi-4 AIE4 engine
chizamd
committed
0c3c39f
fix: harden Phi-4 corelib engine state
chizamd
committed
ee66551
feat: route Phi-4 GGUF models through AIE4
chizamd
committed
5d50361
fix: harden Phi-4 request lifecycle
chizamd
committed
8f831fe
fix: preserve legacy Phi-4 tokenizer semantics
chizamd
committed
17d7525
feat: pull Phi-4 GGUF and tokenizer from pinned sources
chizamd
committed
1319e07
fix: preserve Phi-4 downloader compatibility semantics
chizamd
committed
32f7698
test: validate Phi-4 GGUF AIE4 integration
chizamd
committed
dfdd1f4
test: strengthen Phi-4 AIE4 integration coverage
chizamd
committed
1eb82e4
test: exercise production Phi-4 integration seams
chizamd
committed
14f6161
fix: release idle NPU queue immediately
chizamd
committed
a6ce095
fixup! build: add optional dynamic corelib 0.3.0 runtime
chizamd
committed
867df75
fixup! feat: add validated Phi-4 Q8_0 GGUF reader
chizamd
committed
b0a13ec
fixup! feat: add corelib-backed Phi-4 AIE4 engine
chizamd
committed
a2f8550
fixup! feat: route Phi-4 GGUF models through AIE4
chizamd
committed
ee78f6f
fixup! feat: add corelib-backed Phi-4 AIE4 engine
chizamd
committed
f7eb0c1
fixup! build: add optional dynamic corelib 0.3.0 runtime
chizamd
committed
cd2223a
fixup! feat: route Phi-4 GGUF models through AIE4
chizamd
committed
7347f5b
fixup! feat: add corelib-backed Phi-4 AIE4 engine
chizamd
committed
edd5cad
docs: drop the superpowers plan and design pages from the branch

chizamd
and
claude
committed
465f49b

Commits on Sep 12, 2026

fixup! feat: add corelib-backed Phi-4 AIE4 engine
chizamd
committed
c6827a4
fixup! feat: route Phi-4 GGUF models through AIE4

chizamd
and
claude
committed
87b9ce8
test: add the real-AIE4 acceptance runner

chizamd
and
claude
committed
e6fc279
test: stop the acceptance runner mangling its own curl arguments

chizamd
and
claude
committed
8772108
docs: document developer setup and hardware results

chizamd
and
claude
committed
0032816
feat: report model load time, and account for it on the AIE4 path

chizamd
and
claude
committed
1df3566
fix: report load time on the server path too, not only the CLI

chizamd
and
claude
committed
53b625b
docs: account for the 45 s startup, measured rather than assumed

chizamd
and
claude
committed
a55edfd
perf: stop re-hashing the model at startup, and give the packer a thread hint

chizamd
and
claude
committed
95c5c82
docs: startup is 5 s, not 45 s

chizamd
and
claude
committed
07ee52c
Merge branch 'feature/phi4-gguf-aie4' into feature/phi4-gguf-aie4-medusa
chizamd
committed
fd753ff