[gilby]

What this changes

The decode forward over every layer for 1-16 rows of one or several slots in one launch: Engram before its layers, the DSpark taps and the head, with a whole checkpoint mapped zero-copy where its shards allow.

Part 12 of 18 of the stack for #1676; stacks on dsv41-metal-pr/11-decode-layer (part 11, #1715), on main with parts 1 to 5 (#1705 to #1709) merged.

Receipt

• Environment, with the TensorFold commit, MLX or PyTorch version, machine and GPU, checkpoint and revision: TensorFold d4a47dd on main ad98b60 (native Zig 0.17.0; no MLX or PyTorch); Apple M3 Ultra (60-core GPU), 96 GB, macOS, for the kernel and device tests; checkpoint deepseek-ai/DeepSeek-V4.1-Flash-1017.
• Exactness, from tools/bench_concurrent.py --alone --serial or token_sha against "draft": false: n/a for served output; the tiny model's forward: R rows in one launch == one at a time, 8 decode steps at 4 rows a launch == one row a launch.
• Decode speed before and after, from tools/bench_openai.py: measured on the last PR of the stack.
• Prompt speed before and after: the cold first request's ttft_s from tools/bench_cache.py --require-cold on a long prompt, or "n/a" with the reason when the change cannot reach prompt processing: measured on the last PR of the stack.
• Tests run, with pass, fail and skip counts: zig build test tf-kernel-contract -Dcpu=apple_m1: exit 0, 625/625 steps succeeded; kernel contract, tf-kernel-contract --only dsv41 (Metal): SUMMARY 89 kernels: 89 pass gates 1-4, 0 fail, 0 known findings; 186 controls: 186 red as planted, 0 missed; 0 limit problems; gates 118944 ms, limits 160 ms; zig build test-dsv41-metal -Dcpu=apple_m1 (the device suite, its own step): exit 0, 7/7 steps succeeded; 16/16 tests passed; python3 tools/zig/lean_check.py: 5117 files, 424 problems, the same count as main (5027 files, 424). Zig's summary counts only the test binaries it ran in that invocation, so the step count is the comparable figure.
• Not run, and why: CUDA: no NVIDIA device here; Metal on M1, M2, M4 and M5: one M3 Ultra only; the release-layer checks (test-dsv41-metal-real, test-dsv41-metal-vision-real) were run at the stack's top and are in dsv41 metal: image prompts #1722's receipt.

Checklist

• Every token still goes through the lane rounds. No serial path, and nothing that needs drafts off.
• Drafted output equals "draft": false, a resumed prompt equals a fresh one, and concurrent equals solo.
• No precision traded for speed. If the bits change, the description says which sums change and why.
• A new or changed GPU kernel is registered with the kernel contract (docs/kernel-contract.md); each backend's SUMMARY line is in the receipt.
• Prompt processing is no slower than the last release.
• The description says which platforms I ran, Metal M1 to M5 and CUDA, and which I could not.
• New tests fail before the change, pass after it, and skip cleanly without their dependency.
• Comments and docstrings are one line. No measurements or history in the source.
• No personal data, machine names, internal hosts or local paths. No AI attribution lines.

[ashhart]

Thanks @gilby. We've read parts 12 to 18 as they stand. They apply on main as of 29 Oct (22f3d289a) with nothing new for lean, and the four things DeepSeek's lanes opt out of (trees by default, lanes of trees, copies, and gpu_tokens) stay off through the moved lane half and the Metal target. Since they still carry the pre-31 Oct versions of the earlier parts (Kimi's kernel, the device suite in zig build test, the CUDA serve_engine.zig rewrite), we'll take each rebased part in order after #1705-#1707 and parts 4-11, as you planned. One thing to coordinate: #1722's image-token fields in model_text.zig and prompt.zig are the same as #1698's (@jayleaton), so one form should win. @jayleaton, could you two agree on it there?

[ashhart]

Thanks @gilby. This part landed on main with #1719's merge (cfcbecb), which carries #1712-#1719 as your own commits in order; closing it as included. The same Metal contract, full test and DeepSeek device suite cleared the whole stack.