Repository navigation
Delete dead generate paths + one shared sampling/stop-token fn (verb→engine map follow-up) #4266
Closed as not planned
Description
Follow-up to the verb→engine map (/mnt/nvme-raid0/cop-inbox/evidence/a48-verb-engine-map.md), per the cop ruling. Scope:
- DELETE-DEAD, with callers proven absent first:
C/gguf_generate_result.rs:203chat_generate_session_02.rs:536 generate_ggufS/gguf/inference/fails.rs- VERIFY rows:
generate_cuda*,GpuModelgenerate,AprV2Model::generate
- One shared sampling + stop-token fn that replaces the duplicated argmax, top-k and stop-check copies. It must not touch
qwen35_session.rsorforward_qwen35.rs; those files are ac's Session trait work. - Bugs: wgpu generate has no prefill: only the last prompt token reaches the KV cache (gguf_gpu_generate.rs:296, :765) #4264 (wgpu has no prefill) and apr serve APR-CPU ignores stop tokens and request top_p: hardcoded stop_tokens=[] and top_p=0.9 (serve/handlers.rs:1107) #4265 (serve APR-CPU stop/top_p).
Each change gets a RED→GREEN test. Owner: a48.
Activity
refactor(chat,serve): DELETE-DUP slice 1 — three dense paths the one …
refactor(chat,serve): DELETE-DUP — three callerless dense paths (0.69…
refactor(serve): DELETE-DUP — two test-only generate loops and the te…
release(0.69.3): rc.2 batch — one-engine ports (X, D2, M1/M2b, V2b) (#…
fix(run): wgpu decode prefills the prompt; delete dead run paths; one…
fix(serve): delete the dead AprV2Model::generate loop (#4266 part 1)
refactor(sampling): the last private greedy argmax copies call sampli…
fix(run): wgpu decode prefills the prompt; delete dead run paths; one…
refactor(serve): DELETE-DUP — two test-only generate loops and the te…
comparator: none+ reason #3774noahgift commentedon Oct 14, 2026
P0 to P1 (2026-10-14 P0 triage): P0 now means it blocks the next release or carries the operator's own words on it; this issue does neither, so it stays on the 0.71.0 milestone at P1.
noahgift commentedon Oct 14, 2026
merged into #5313 (master ticket for epic #3999; triage rule 7b, 2026-10-14). It is a checklist row there; reopen this issue when work on it starts or its PR opens.
comparator: none+ reason #3774