Library changes outside the platform's own dependencies (those came with the platform commits above). libs/asset and libs/flow move to the Stage repository with the apps that use them.
Covers work 6f1e44649..2b1a41df4 for libs:
- wm: the Android super-app compiles tiles on the phone against the host engine
- cargo-makepad: the super-app APK is packed by `android dyn-pack`, in Rust
- cef: a page's audio can be captured instead of played
- flow: the flow-ui app and the flowgraph canvas return to the workspace, with staged progress in libs/flow
- svg: a stroke that bends tighter than its half width collapses its inner edge to the corner
- the accent a name is written with stops deciding whether search finds it
- ai stems: LaneDemixer, VocalsModel and the platform Stage origin/main needs
- The regex engine accepts \b and \B, reports the earliest match so a scan can resume after it, and can decide word boundaries on ASCII so a code sear
- workspace: Scope's settings carry the code map's tab width and the directories each prepared project hides from its map
- cef, video, platform: what Stage's browser needs from the shared layers
- platform, wm: work builds again on iOS, tvOS and wasm, and the window manager is warning-free on Windows
- ci: tools/ci, a Splash runtime that watches a branch and runs every ci.splash it finds
- stitch: the crate says what the rest of the repository says about its license
- ci: a run is one target dir and one cargo batch per target, and the wall draws
- apps: what the CI box found
- Remove Flow, asset and VJ applications moved into Stage
- tests: the workspace suite, run as a whole for the first time, passes outside the example UI tests
- libs: no warnings in the workspace check on any row
- tests: the last three failures of the CI box's night
- gif: the crate's doc examples compile
- cef: a hosted browser survives its host's exit
- tests: the three binaries over ten seconds in release come under it
- git: the memory ledger can take what it is asked for
- script: re-entrant dispatch, thread index validation, Any-based handle downcasts, UTF-8 previews; regex: never_loop fix
- fix(text): enable ttf-parser gvar-alloc for many-tuple variable fonts
- platform, widgets, wm: a touch that is taken away is cancelled, never released
- platform, wm: hosted apps ask the WM for what they cannot do themselves
- wm on Android: apps build on the phone
- task, cargo-makepad, widgets, audio: the task manager chooses its columns, graphs every process' network and disk traffic and installs itself as a D
- flowgraph: ordered ports take many wires, sockets can override their icon, and the canvas embeds as a viewport
- cef: captured frames and audio carry callback clocks and a navigation epoch
- git: timings read a portable clock
- ai: the hub's rig-fixture tests find the asset library without the asset client
- ai llm: a Metal main buffer is filled in 32 MB chunks
- diskmap: a scan lists each macOS directory in one getattrlistbulk call and classifies files without allocating or locking
- platform, widgets, audio_route, ai: window crossfades and whole-frame presents, caption controls that click, themed menus, a stereo-pair audio tap,
- builder: the Builder TUI matches the new terminal design, Windows executables carry app icons, and every app has one
- sheets, score, git, home: saves cannot cut a file short or overwrite another one, and checkouts cannot write outside the repository
- git: worktree status matches git on real repositories and can be cancelled
- cuda: builds link only the CUDA toolkit they are given, never a system install
- audio_route: one API on macOS, Windows and Linux, monitoring by default and processing only when asked, plus shaders that discard a value compile to
- platform, widgets, ai: a restyle recompiles nothing it already has, shows only complete frames, and a window can own its caption
- audio_route example: print the samples that reached the processor each second
- widgets, builder: the big widget families are features, all on by default, and the Builder takes none of them
- ai hub: a local chat with no model says so once, plainly
- ai: Claude Desktop drives any app with the F10 panel
- wm: children open no console windows on Windows
- ai: Connect hands the .mcpb to Claude Desktop itself
- civil-time: one local wall clock for every app, and it knows the zone on Windows
- ai-speech: makepad-ai-sfx (and with it ai-h3) is built only for IndexTTS
- script + wasm_bridge: no build script that reruns on every file
- build: the AI stack, csg, Scope and Source Library join the one workspace
- ai chat: the chat's backends are the CLI, MCP and cloud providers; the local model is the `localai` feature
- widgets: every widget family is a crate of its own under widgets/families, and makepad-widgets links, re-exports and registers the ones its features
- apps, examples, libs: each crate builds only the widget families it uses, and the design overlay is each app's own default feature
- livepipe: takes makepad-widgets without its default families, now that the AI stack is in the one workspace
- ai chat: a build without local AI starts on the first provider that answers here, not on "No model"
- thiserror: RUSTC_BOOTSTRAP no longer turns on the unstable generic member access API
- builder: makepad-builder.exe declares itself like a well-formed Windows program and stops a stalled build through a job object instead of taskkill
- builder + ai-cuda: the CUDA kernel progress shows in the Builder's Compile row, not scribbled over its screen
- repo: no third-party comic archive, generated models or sample asset in the tree
- builder: the Windows Builder ships as source, runs in its own console, and apps compile with Rust's GNU toolchain or Microsoft's
- win_resource: PNG decode and encode on makepad-fast-inflate instead of zune-png
- builder: one folder layout everywhere, makepad-builder.zip with only the files a Windows build reads, a Rust download over parallel connections, and
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Windows: the Store package's App Execution Alias (%LOCALAPPDATA%\Microsoft\WindowsApps\claude-desktop.exe), else the running Claude Desktop's own exe (never Claude Code's claude.exe), else the installer's AnthropicClaude\claude.exe or Programs\Claude\Claude.exe; started with no console window.
- macOS: open -a Claude <bundle>, when Claude.app is in /Applications or ~/Applications. Linux: xdg-open as before.
- No Claude Desktop found: the panel says "Claude Desktop isn't installed".
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The loopback MCP server moves from Director into libs/ai/services (mcp::server); Director re-exports it, its lane tokens and tests unchanged. The dispatcher now names the server and its instructions; a TokenStore can live in memory only (ephemeral).
- mcp::host: while Claude Desktop is the provider the panel serves its registry's tools (service__tool names) on an ephemeral 127.0.0.1 port with a fresh bearer token, and writes ~/.makepad/mcp/<exe-stem>.json {pid, port, token, title} (0600 in a 0700 dir), removed when the provider changes or the panel goes. Calls queue to the UI thread and run through EngineCore::call_external: a card in the transcript, destructive calls held for the person's confirm (the pane opens for it), the result sent back when the card lands.
- mcp::mcpb: "Connect to Claude Desktop" writes <exe-stem>.mcpb (a stored zip, manifest_version 0.3, binary server = this executable with --mcp) and opens it so Claude Desktop shows its install dialog.
- platform mcp_relay: `<app> --mcp`, checked first in app_main before any Cx, window, GPU or audio, relays newline JSON-RPC on stdio to the running app's endpoint. It answers initialize/ping itself and tools/list from the app's last list while the app is down, starts the app (detached, MAKEPAD_AI_PROVIDER=claude-desktop, engine up with the pane closed) on the first call that needs it, waits up to 20 s for its file, and exits when stdin closes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Remove the retired applications, asset-specific libraries and DJ pack tool
from Makepad, together with their workspace and launcher entries. All 981
removed source paths are accounted for in the private Stage repository.
Keep public AI chat generation through the AI Hub's generic job runner.
Extract shared SHA-256 and UDP binding into core_util so the public hub and
model crates no longer depend on the relocated asset libraries. Preserve
the retained public coverage in the split wasm validation script.
Validation on the exact cleanup tree in an isolated checkout:
- Release checks: core_util, model, aichat, WM and Builder.
- Release builds: aichat, WM and Builder.
- Existing tests: core_util 5, model 30, aichat 7, Builder 2 passed.
- Core/model checks: wasm32, Linux and Windows passed.
- No warnings in the successful checks, builds or tests.
Known baseline: WM library tests do not compile because the unchanged
style-transition assertion compares seven expected weights with eight.
The unrelated working-tree correction is intentionally outside this commit.
A test is a ci.splash beside what it tests; the script decides every input and the model only ever judges one picture against one acceptance text. mod.ci: launch (hidden, --remote, user_seq preserved), key, type_text, click, get, snap, wait_log (a * is a gap inside one line), no_errors, grab, quit; step, sleep, check, run; cargo, check_targets (the cargo makepad check matrix, check only for platforms we are not on, a test fails if the two tables drift), test, build, machine (another box over the makepad tunnel), exclusive; judge, accept, ask. The watcher polls git ls-remote once a minute for work and any extra branches, syncs a checkout the CI owns, runs the root script first and alone, then the rest up to a parallel limit behind one shared model judge. The window is a wall of squares, one per script: green passed, orange warnings, red failures, with a detail panel for the selected one.
Scripts: the root ci.splash (workspace check with core warnings denied, the tests), apps/wm (desktop up, switch to macOS by Cmd+Space / type / Return, launch the terminal and the browser, each waited for by the WM's own first-frame line), and one per main app in the default shape. Proven here: apps/wm/ci.splash green in 280 s, fifteen target checks and seven vision verdicts.
Models come from Hugging Face through the hub: registry entries qwen3.5-4b-vision and qwen3.5-9b-vision with exact revisions, sizes and digests, and hub-install, a command line over LocalModels::start_install. vlm-probe reads PNG.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Taken from vjroger/work. A full merge of that fork fights this work branch
in widgets; this is the stems crate, vocal mel-band, span-cache lanes,
log_ring, output fence, midi inject, effect_doc, mp3 sniff, mp4 audio
edit, and audio-only file open. Stage on origin/main cargo-checks again.
* Button: activate from the keyboard when focused
`Button` matched only `KeyFocus` and `KeyFocusLost`, so a button reached with
Tab could take focus and animate, but no key press ever activated it.
* `Space`, `Enter` and numpad enter now emit `Pressed` and then `Clicked`, and
make the same `on_press`/`on_click` script calls a tap does.
* Key repeat is ignored, so holding the key down doesn't re-press.
* Both arms are gated on `enabled`, like the finger paths.
* libs/ai: exclude hub_ui and services from the AI workspace
`cargo install --git ... cargo-makepad` fails with "package
`libs/ai/services/Cargo.toml` is a member of the wrong workspace": the root
workspace lists `libs/ai/hub_ui` and `libs/ai/services` as members, but both
sit under `libs/ai`, which is its own workspace, and cargo gives a package to
the first workspace root above it that doesn't exclude it. `libs/ai/hub` and
`libs/ai/livepipe` were already fine since each declares its own `[workspace]`.
libs/sqlite_query reads WITHOUT ROWID tables through the index cursor,
refuses to write them, and the pager releases its process write slot on
drop. libs/tar is an in-repo tar reader with gzip through fast_inflate.
libs/git imports packed objects with bounded checkout writers and its
HTTP file responses carry a trailer. libs/code_language gains a Haskell lexer with literate (Bird) dialect
detection. libs/loader_bundle and libs/search are new;
libs/app_module carries the super-app module surface; libs/workspace
adds Settings.infinite_zoom and RendererChoice::gpu_env_value; libs/ai
builds without warnings across the hub, llm, metal and model crates;
windows-rs job object handles are c_void.
Squashed from work (the libs parts of each):
- Restore retained renderer support for Scope (libs/search)
- Share Builder target across Makepad app builds (libs/loader_bundle)
- Index local Apple Mail with Gmail labels, attachments and reimport (libs/sqlite_query)
- libs/ai: warning cleanup across the hub, llm, metal and model crates
- code_language: a Haskell lexer with literate (Bird) dialect detection
- git: packed imports and bounded checkout writers; HTTP file responses carry a trailer
- builder: dedicated-folder installer checks, catalog and runtime fixes; Windows job objects hold c_void handles (libs/windows)
- workspace: Settings.infinite_zoom, the experimental prepared map inside the glyph
- wm: the Android super-app hosts apps as on-demand dylibs (libs/app_module)
- libs/tar: an in-repo tar reader; the super-app unpacks its archives with it
- workspace: RendererChoice::gpu_env_value follows the platform's runtime GPU choice
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Squash of 54 work commits (Sep 1–12):
6251f7c ai-hub: body domain — live pose packets ride the realtime session
ea50c77 chat_ui: the feed's session gets its profile brief back
f51b5f3 ai-body: the crate for the native SAM 3D Body port, with its weights reader
8211ae6 ai-body: the MHR rig and the pose head's parameter decoding, oracle-exact
9e343a8 ai-body: the DINOv3 ViT-H+/16 backbone, crop and ray conditioning; Metal gains rope-half and affine layer norm
69d842c ai-body: the promptable pose decoder and its refinement loop, oracle-matched on Metal
66e5e2f ai-hub: SAM 3D Body runs natively — `sam3dbody` on the body domain, oracle-matched end to end
a634198 ai-hub: the body-native commit carried a peer's in-flight hub hunks; put them back where they were
9ff44e8 ai-hub: the body-native wiring, this time only the lane's hunks
6a1c16b ai-body: third-party notices — what the port is implemented after, and what it is not
d78411a ai-body: the per-step work moves to the GPU
b22259b ai-body: the context stays on the GPU; only the pose token leaves the loop
346f31f ai-body: flash attention for the head-dim-64 blocks
45b5b98 ai-body: the crop size is a runtime knob, and the loop reports where its time goes
4be6d19 ai-body: the test modules import the grid constants they still use
7598346 ai-body: tensor-core GEMMs for the backbone, and the rig's correctives only where they count
a9ce596 ai-body: the crop warp runs across cores
8964ba6 ai-body: an FP8 backbone mode, off by default, measured against the oracle
a2aaa8f ai-body: the FP8 bias rides a column-broadcast add on the device
d53c77d metal: a device-resident ViT stack, and the body backbone rides it
d006d0a metal: resident f32 linears keep their weight on the device
525ba1c metal: a device-resident two-way decoder layer, and the body decoder rides it
c9e6d88 ai-body: the hands pass — hand crops, the hand decoder, the hand-mode rig and the wrist fusion
62dff26 ai-body: the mask prompt — a person's segmentation mask conditions the body pass
a648cf8 ai-hub: body session options — hands, detect, persons=N
8c568df ai-hub: drop the SAM 3D Body reference worker backend
7ff875a ai-hub: keep a peer's in-flight beats/notes/local work out of the body commits
31e5faa ai-hub: local model runner, licence acknowledgements, a shared install panel; Beat This!, Basic Pitch and the Salamander drum-kit entries
b94bc58 ai-services: the wire, the app port and the panel state — one conversation, many apps
2acb798 ai-services: wire v2 — endpoints, receiver-side caps, result disposition
8ae0ffb ai-services: the engine core — registry, router and conversation, tested against a scripted model
2308736 ai-services: the real models behind the engine feature — local through the hub, Claude, and none
c3f631d livepipe: one reusable pipe from a camera to a fleet node and back
ff62db3 ai libs: the runtime env-var cleanup — precision is a per-caller policy, not an environment side channel
04a94ef realtime: one service-log line when a live session opens and one when it closes
0ecb81c ai models: the model-crates env-var cleanup — 172 research knobs gone, the unset default is the code
4ca36c1 ai hub + services: the assistant's model comes from wherever it is resident — the fleet chat box, with tools, then the local weights
432121e aichat engine + wm: launch, then use — the assistant continues in the same turn once the app it started is on the bus
7a5bf69 ai-hub registry: the Salamander drumkit samples come from the makepad.nl mirror — the GitHub repo only carries the .sfz files
102ffc5 ai-services: messages on the bus — a manifest declares topics, the engine subscribes on a tool's behalf or by ToolResult.subscribe, a service publishes Message frames, an idle conversation wakes on a message as an event turn under rate laws; the WM bus forwards the new frames; every app that matches the wire gets its arm
a837792 hub + flow: a whitespace-only chat completion is retried once and then fails instead of passing as an answer; a flow's model is a fleet model id unless it names a weight file on disk; chat models show under the text domain in /v1/models
bc6c620 hub + flow: what the chat review found — the in-process route retries an empty completion too, a node says whether its prefill opened thinking so a brief-mode answer is never discarded, a preferred model falls back to normal election when no node has it, discovery keeps looking for the preferred model until patience runs out
75c3441 hub: the PRO 6000 serves image as well as chat and text
ad5e98b hub registry: flux2-dev's VRAM estimate is its measured peak, 30 GB
c7241e0 hub: a node that evicted every resident releases its cached allocator pool before refusing a load or publishing usable VRAM
30575f0 flow: route generation by request workload
1be1e21 ai-hub: gate downloads by disk capacity and recover fleet admission
df6b394 filesystem_watcher, bounded_http, ai services: live and tool prerequisites
79ebdb9 ai-hub: add a native Pixal3D image-to-3D backend
0ba0d74 ai-hub: propagate typed refusals under reject queue policy
cc6c872 Speed up H3 conditioning and video decoding
e512059 Fix Qwen vision residency and generated material colors
2864f68 ai-hub http client: bound every plain TCP connect to 3 s per address
3d93229 ai: CUDA is a Linux/Windows-only dependency; the hub library defaults to llm + stt
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Account for missing and partial model files per volume, reserve headroom, and reject disk-constrained workers before accepting a job. Preserve typed admission failures so callers can choose another peer, and make activity gating and cancellation recover cleanly.
Validation: 592 release hub tests passed, one ignored; required-CUDA builds deployed to six idle Windows workers.
Clocked piano/ironfish/drum rack, program-bus mix, splat/mixer/music
updates. Silence the unused warnings that show up in `cargo check -p
makepad-vj --release`.
The 5090 kept about 1.5 GB of CUDA allocator pool after evicting every
model, so a card that fits flux2-dev refused it (30510 MB free reported,
32090 MB in a fresh process). Admission now trims the pool when the last
resident is gone and re-measures before refusing; usable VRAM is measured
after the same trim; the refusal names the pool it released.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Measured on the RTX 5090 (2026-09-04, 1024x1024, 8 steps, nvidia-smi 250 ms
samples): 30.5 GB used at peak, the run completed in 52 s. The old 29 GB
was a pre-measurement guess that under-reported the model by 1.5 GB.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
flux2-dev cannot fit on the 5090 at the default reserve, and the role
table barred the only card that can hold it. The user opened image on
10.0.0.165 ("let the rtx serve images too"); the role test and the
flow's role-aware listing test follow.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The 5090 listed flux2-dev ready (total VRAM passed the gate) then refused
every job (29696 + 2048 reserve > ~30510 usable); the flow picked it by
domain ETA and never retried. Nodes now publish vram_usable_mb and mark
un-admittable models too_small; the fleet gate uses usable VRAM; the
flow's gen executor picks admitted nodes for the requested model (ready
first), retries up to three nodes after an admission refusal, and when no
node can take the model its error says why per node (role, too small with
the numbers, waiting for VRAM). The flow's model listing drops (model,
node) pairs the node's fleet role bars, so the picker no longer counts the
chat-only PRO 6000 as ready for image; its label reads ready/absent/too
small with the GPUs named.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
gen.image{prompt} is a service inside the assistant: on a worker it
runs the creator pipeline for the image domain against the node the
runner picks (LAN discovery, ETA-ranked, the resident flux model wins),
streams the node's progress into the tool card, and writes the picture
under the makepad home's gen/ dir; the creator library gains
generate_bytes — the request, poll and fetch that generate_and_publish
now shares — so no second pipeline exists. photos.add{path} bakes that
one file into the open library (only from the gen dir or the person's
home), re-opens the wall and glides onto the new picture; the module
executor answers it later through the host's reply sink. The model
chains them, launching photos in between. A registry seam found on the
way: a call dispatched in the same event a link was adopted reached the
port before its Registered, so the registry now answers a leading
Register at once.
The wall gained a search box: every keystroke re-cuts the packing over
the matches (every word in a title or link; title beats link, a whole
word beats a substring), each picture flies from its drawn rect to its
target over 450 ms, dropped pictures shrink and fade, returning ones
grow back; Esc clears; ⌘F or a bare slash focuses the box; the status
reads "35 of 293 · ro". photos.filter{query} sets the same from the
pane; instances are rebuilt only while a flight runs, and the clock is
the platform's.
Proven on the fleet: "make me a picture of a red bicycle on the moon
and put it in photos" → the file in 40 s, photos launched, added and
shown among the comics; typing r, ro, rob re-packs the wall live; the
filter through the pane in a process tile and as a module. ai-services
39, aichat 5, photos 9, image_tiles 12, creator 9; aichat and photos
check for wasm32.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
An ok os.launch result is held while the launched app is not on the
bus and delivered when it registers (within 8 s), rewritten as "<App>
is running now. Its tools: … Call them directly." with the tool table
rebound mid-turn where the model can (the local model prepends the
update to the tool result; Claude keeps its native table until the next
turn and gets the names in the text); past the patience the result says
the app has not connected yet. An assistant entry that never got
visible text is dropped before a tool call and at the end of a turn,
and the panel draws an empty one as nothing — the blank block above
the first card is gone. The model's context now opens with the running
apps and their tool names and lists the others as os.launch candidates;
the WM's launch description says a running app's tools are already
there, and its answer for one that is already running says so and
brings it to the front.
Proven on the fleet model: "open the photos app and find the pictures
about dogs" → launch (running 4 s later) → search, summary, search →
the answer, one turn, the tile beside the pane. ai-services 38, wm 153,
aichat 2; aichat checks for wasm32.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The hub chat session's election no longer stops at this machine: a
co-located serving holder first, then a fleet chat node heard by
discovery and role-allowed for chat (a short patience for beacons),
then the weights on this machine — and an honest answer naming both
misses and where weights may be put when none of those exists. Tool
packs ride every route: the system text carries the tool table as the
node's chat_system, one splitter strips thinking and collects
<tool_call> bodies across deltas, the in-process worker's parser is the
one parser, and tool results go back as tool turns. A node that fails
mid-turn ends that turn with the node named; the person's next line
re-elects, served first on the new route. The tools-only guard on the
proxy is gone. The session exposes its route, and the panel's chip
shows it.
The local lookup is independent of the working directory: the env
override, then the makepad home's weights (Qwen preferred, largest),
then the checkout the binary came from, then the cwd — so an assistant
launched from a binary copy finds the same weights as one run from the
checkout, and the Local provider always builds even with none.
Proven live with no local weights: from the sheets overlay a plain
line reached the fleet's 27B (the node with the model resident won the
pick), the model called sheets.summary and answered in 12 s; the app
stayed at 254 MB. hub_chat + local_llm 11, services 34.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The per-family model crates (flux, h3, paint, music, speech, stems,
vision, sfx, rife, trellis, beats, common) and the libs/diffusion
research binaries read some 190 environment variables that were
research knobs: tensor dumps, per-stage timing, oracle-parity and
fixture rigs, experiment toggles. The path taken with none of them set
is the one that ships; every such knob is deleted with the code it
gated, and every losing branch of an experiment toggle goes with its
toggle — dead kernels, fields and functions included. What remains are
the real configuration variables (the FLUX_*_MODE family, FLUX_GRAPH,
the VAE pool cap, the FLUX2 text-encoder residency, H3_VAE_BATCH, the
music3 caches and official modes, the stems/beats f16 switches, the
weight and data roots) and the build-script variables.
Rebased on the runtime cleanup: precision stays explicit everywhere
(GemmPrecision, f16_attention_operands, the H3 text precision, DA3's
StrictF32 in code); no act16, no H3_ACT_F16, no FLUX_ATTN_F16 or
FLUX_VAE_CONV_GEMM reads survive.
Reviewed by the delegate reviewer (APPLY, no findings) and gated on the
Windows CUDA box: all seventeen model crates check, motion and vision
tests, the hub and the diffusion bins — the gate caught one CUDA-only
tap marker the Mac never compiles, removed here. On this Mac: the same
checks plus motion 24, paint 158 and vision 23 tests.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A headless box's service log showed only the startup banner for a whole
session; now it records the model and wire encodings on open, and on close the
elapsed time, frames in/out, fps, and the dropped/undecodable/unencoded counts.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The AI runtime crates read their precision, activation and kernel
choices from environment variables; the code path taken with none set
is the one that ships, so every knob that selected it is now an explicit
argument and every losing path is gone. GemmPrecision { f16_accumulate,
f16_activations } is passed by each caller: the default {true, true} is
the old unset Flux route; H3's DiT and text encoder pass {false, false}
(H3's >1e4 activations saturate f16 — the policy H3 used to set through
FLUX_GEMM_F16ACC=0 on itself), its VAE {true, false}; DA3's StrictF32
selects f32 packed attention in code; Hy-Motion carries an explicit
f16_attention_operands flag through its text refiner, its double and
single blocks and the CUDA backend (true in production, false only in
its full validator). The libs/diffusion bins — a separate workspace —
are migrated to the same shapes.
Benches and validators no longer set variables on themselves: llama's
skip-logits is a session option (the CUDA bench turns it on), OCR takes
explicit use_f16_gemm and tiled_roformer options, the lane speculative
probe reads its CLI. The live gates the first cut had deleted are back
as explicit-config tests: MMQ M=129, the strided-f32 MMV path, the
RMS+MUL CPU oracle. The loader's THREADS and CHUNK_MB stay real settings.
The Metal quantized-matmul experiment (metal_qmm and its vendored MLX
kernels) was reachable only through a knob and goes with it.
Reviewed in three rounds by the delegate reviewer (the last round
accepted everything but one Hy-Motion call site, fixed in round four
and reviewed here), and gated on the Windows CUDA box: lib checks of
common/paint/loader/cuda/llm/motion/vision, motion 24 and vision 23
tests, the hub check, the diffusion bins, llm 253 passed / 1 ignored.
On this Mac: the same checks plus the motion and vision tests.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
