deps(backend)(deps): bump llama-cpp-4 from 0.2.54 to 0.2.60 in /backend

[dependabot]

Bumps llama-cpp-4 from 0.2.54 to 0.2.60.

Release notes

Sourced from llama-cpp-4's releases.

v0.2.60

What's Changed

• Rust bindings for the upstream pre-norm embeddings API change in ggml-org/llama.cpp#23198:
  • llama-cpp-sys-4/wrapper.h now pulls in llama-ext.h, exposing llama_set_embeddings_pre_norm(ctx, value, masked) and the pre-norm getters to bindgen.
  • New safe methods on LlamaContext: set_embeddings_pre_norm, get_embeddings_pre_norm, get_embeddings_pre_norm_ith.
• MTP example (examples/mtp): prompt prefill no longer forces logits=true on every token — only the final position needs it. MtpSession::new already drives the target context into unmasked pre-norm mode, so the per-token logits workaround is gone.

Full Changelog: eugenehp/llama-cpp-rs@v0.2.59...v0.2.60

v0.2.59

What's Changed

• Updated llama-cpp (bot) by @​eugenehp in eugenehp/llama-cpp-rs#218

Full Changelog: eugenehp/llama-cpp-rs@v0.2.56...v0.2.59

v0.2.56

Highlights

• llama.cpp bumped to 64b38b561 (May 2026) — includes upstream MTP support ([PR #22673](ggml-org/llama.cpp#22673)).
• In-tree MTP patch removed — replaced by upstream. Breaking changes: the mtp Cargo feature, LlamaContext::set_mtp, and LlamaModelParams::with_override_arch are gone.
• New Rust API for MTP — LlamaContextType { Default, Mtp }, with_ctx_type / with_n_rs_seq, and llama_cpp_4::mtp::MtpSession (the speculative-decoding draft loop, wrapped via a small C++ shim in llama-cpp-sys-4/mtp_shim/).
• New examples/mtp/ — drives the full draft loop end-to-end. Verified on Qwen3.6-27B-IQ2_M / Apple M4 Pro / Metal at 94.1% draft acceptance.

Changed

• llama.cpp submodule → 64b38b561 (master, 2026-05-16).
• llama-cpp-sys-4/patches/0002-mtp.patch deleted. The mtp Cargo feature removed from both crates.

Added

• LlamaContextType { Default, Mtp } enum and LlamaContextParams::with_ctx_type / ctx_type wrapping upstream's llama_context_type.
• LlamaContextParams::with_n_rs_seq / n_rs_seq and LlamaContext::n_rs_seq now always available (no feature gate).
• llama_cpp_4::mtp::MtpSession — Rust-callable MTP speculative-decoding draft loop. Wraps upstream's common_speculative_* MTP path via a small C++ shim with stable C linkage.
• examples/mtp/ — with `--predict N` drives the full draft loop via MtpSession. Without `--predict`, just configures contexts (smoke test).

Removed (breaking)

• mtp Cargo feature on both crates.
• LlamaContext::set_mtp — upstream removed the llama_set_mtp C API; MTP is now configured via ctx_type on the context, not by post-hoc attachment.
• LlamaModelParams::with_override_arch / override_arch — upstream removed the field; MTP head architecture is autodetected from GGUF metadata.
• llama_cpp_sys_4::llama_context_seq_rm — patched alias is gone; use llama_get_memory + llama_memory_seq_rm (which clear_kv_cache_seq already does internally).

Migration

• Drop `features = ["mtp"]` from your `Cargo.toml`.
• Replace `set_mtp(Some(&draft_ctx))` with constructing the draft context from `LlamaContextParams::default().with_ctx_type(LlamaContextType::Mtp)`.
• Replace `with_override_arch(...)` calls with nothing — autodetected.
• `scripts/bench-mtp.sh` now passes `--spec-type draft-mtp` (was `mtp`).

... (truncated)

Commits

• 5bcde45 0.2.60
• 6d064aa updated api based on ggml-org/llama.cpp#23198
• 59fe9c5 0.2.59
• 759bbbe 0.2.57
• 726fac4 Merge pull request #218 from eugenehp/update-llama-cpp-2026-05-18
• 5a244c9 updated llama.cpp
• 08f128b 0.2.56
• a5cb307 Merge pull request #211 from eugenehp/dependabot/cargo/cc-1.2.62
• b3e3d61 Merge pull request #210 from eugenehp/dependabot/cargo/tokio-1.52.3
• 3a76b13 build(deps): Bump cc from 1.2.61 to 1.2.62
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)