Last updated: 2026-07-19

CUDA Runtime

Dbg Cuda

All API Sections

CUDA forward-pass debug harness for the qwen35 hybrid-SSM model. Two modes:

zig build cuda-dbg -- <token> [model.gguf] Per-layer residual-norm dump at pos 0 (used to pinpoint the attention gate bug: diff vs a reference implementation eval-callback `l_out-N` reference).

zig build cuda-dbg -- gen <id,id,...> <ngen> [model.gguf] Autoregressive greedy generation from a prompt token-id list. Prefills the ids (exercising pos>0 RoPE + multi-entry attention + SSM state carry), then greedily emits ngen tokens. Diff GEN_IDS vs `/tmp/gen` (the reference implementation greedy) to validate the full decode path beyond pos 0.

Read-only w.r.t. the engine — uses only public ForwardCuda methods.

1 exports 0 methods src/dbg_cuda.zig

1 exports shown

function

main

# 
pub fn main() !void

No public doc comment yet.

src/dbg_cuda.zig:408