Skip to content

fix(spu): kernel restore + Jina V4 trained-context cap (PR-1.H) #297

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
toddwbucy merged 5 commits into from
May 8, 2026
Merged

fix(spu): kernel restore + Jina V4 trained-context cap (PR-1.H) #297

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
76c47ce
fix(spu): restore kernels/transformer.cu (PR-0.5.E regression)
r3d91ll May 8, 2026
46d65d6
fix(spu): cap embedder max_seq_len to Jina V4 trained context
r3d91ll May 8, 2026
59660d2
docs(infra): in-process embedder smoke test evidence (task #117)
r3d91ll May 8, 2026
9941b33
fix(spu): align oversized-input error to capped max_seq_length (PR #2…
r3d91ll May 8, 2026
71a64b7
docs(spu): clarify max_seq_len() returns the JINA_V4_TRAINED_CONTEXT-…
r3d91ll May 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 1 addition & 23 deletions crates/weaver-spu/build.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,29 +18,7 @@ use std::path::PathBuf;

fn main() -> Result<(), Box<dyn std::error::Error>> {
if std::env::var("CARGO_FEATURE_CUDA").is_ok() {
// The `cuda` feature gates the legacy cudarc-based decoder path
// (slated for retirement in Phase 3 cleanup per Cargo.toml). The
// hand-written CUDA kernels for that path live at
// `kernels/transformer.cu`; they were lost in the PR-0.5.E
// shell-crate consolidation. Skip the compile gracefully when
// the file is absent so `--features inference` builds aren't
// broken on main pending a separate kernel-restore PR.
let manifest_dir = std::env::var("CARGO_MANIFEST_DIR")
.expect("CARGO_MANIFEST_DIR not set");
let kernel = PathBuf::from(&manifest_dir).join("kernels/transformer.cu");
if kernel.exists() {
compile_cuda_kernels();
} else {
println!(
"cargo:warning=weaver-spu: skipping CUDA kernel compile — \
kernels/transformer.cu absent (pre-existing regression \
from PR-0.5.E; cudarc decoder path will fail at link \
if its symbols are referenced)"
);
// Still emit the rerun-if-changed so a future kernel restore
// triggers a rebuild.
println!("cargo:rerun-if-changed=kernels/transformer.cu");
}
compile_cuda_kernels();
}
// Persephone proto is always compiled — `encoder::grpc_client_legacy`
// is the production embedder backend during the migration window
Expand Down
Loading