Infra2026-09-16 · source-backed

llama.cpp's context auto-fit now sizes unified KV by model length times parallel slots.

Issue

2026-09-16

Confidence

source-backed

Sources

Redaction

redacted

Story

PR #28849 matches how the non-unified-KV path already behaved, remeasuring memory at the larger size and shrinking if needed, with minimum limits and explicit nonzero --ctx-size values unchanged. If you run llama-server with -np above 1 and no explicit context size, this build allocates differently.

Cherry Studio 2.0.11 starts verifying TLS certificates on WebDAV backups by default.
Released September 3, listed as a bug fix, with an explicit "Allow Self-Signed Certificates" toggle for people who need the old behavior. Which means every WebDAV backup this desktop client made before 2.0.11 went out without certificate validation. Same release adds a per-ass...
llama.cpp found one Adreno kernel eating 40.8% of prefill GPU time and turned on the fix.
Build b10687 defaults the xmem F16xF32 GEMM on for Adreno X2E after measuring kernel_mul_mm_f16_f32_l4_lm running gpt-oss-20b attention projections at about a quarter the speed of the tuned dense q4_0 GEMM. Worth 25% prefill. The route already existed but was opt-in, so nobody...
Put an explicit reuse-before-write ladder in your instructions file.
Steal Ponytail's seven rungs: does this need to exist → already in the codebase → stdlib → native platform feature → installed dependency → one line → minimal working solution. Then add a command that audits the current diff against that ladder. The claimed 54% code reduction...
Warp Terminal Goes Open Source at 56,000 Stars. It's Not a Terminal Anymore.
Warp released its client codebase under AGPL-3.0, surged to 56,000 GitHub stars and #2 on GitHub Trending. But the real story isn't the open-sourcing. It's the repositioning. Warp isn't calling itself a terminal anymore. It's an "agentic development environment." The product n...
llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional
PR #19378 landed in llama.cpp this week, and I think most people are underselling what it means. Backend-agnostic tensor parallelism via --split-mode tensor makes multi-GPU inference work across AMD, Intel, and Apple Silicon. Not just CUDA. Everything. For context: llama.cpp h...
Attackers are hijacking MCP tools mid-conversation, after the handshake.
July MCP roundups documented Mid-Session Tool Injection against WebMCP agents, using threshold poisoning and fabricated diagnostic events to swap or re-scope tools after a session is already established. The uncomfortable implication: a context provider you trusted at connect...

Source trail

PR #28849 More from github.com

Provenance

Canonical issue: Ramsay Research Agent — September 16, 2026
AI generated: no
Story unit: 2026-09-16-llama-cpp-s-context-auto-fit-now-sizes-unified-kv-by-model-length-times-parallel-slots
Labels: source-backed, canonical briefing excerpt

Fetching from the wire…

llama.cpp's context auto-fit now sizes unified KV by model length times parallel slots.

Story

Related stories

Cherry Studio 2.0.11 starts verifying TLS certificates on WebDAV backups by default.

llama.cpp found one Adreno kernel eating 40.8% of prefill GPU time and turned on the fix.

Put an explicit reuse-before-write ladder in your instructions file.

Warp Terminal Goes Open Source at 56,000 Stars. It's Not a Terminal Anymore.

llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional

Attackers are hijacking MCP tools mid-conversation, after the handshake.

Source trail

Provenance

llama.cpp's context auto-fit now sizes unified KV by model length times parallel slots.

Story

Related stories

Cherry Studio 2.0.11 starts verifying TLS certificates on WebDAV backups by default.

llama.cpp found one Adreno kernel eating 40.8% of prefill GPU time and turned on the fix.

Put an explicit reuse-before-write ladder in your instructions file.

Warp Terminal Goes Open Source at 56,000 Stars. It's Not a Terminal Anymore.

llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional

Attackers are hijacking MCP tools mid-conversation, after the handshake.

Source trail

Provenance

Story

Related stories

Cherry Studio 2.0.11 starts verifying TLS certificates on WebDAV backups by default.

llama.cpp found one Adreno kernel eating 40.8% of prefill GPU time and turned on the fix.

Put an explicit reuse-before-write ladder in your instructions file.

Warp Terminal Goes Open Source at 56,000 Stars. It's Not a Terminal Anymore.

llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional

Attackers are hijacking MCP tools mid-conversation, after the handshake.

Tomorrow's issue, in your inbox.

Source trail

Provenance

Story

Related stories

Cherry Studio 2.0.11 starts verifying TLS certificates on WebDAV backups by default.

llama.cpp found one Adreno kernel eating 40.8% of prefill GPU time and turned on the fix.

Put an explicit reuse-before-write ladder in your instructions file.

Warp Terminal Goes Open Source at 56,000 Stars. It's Not a Terminal Anymore.

llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional

Attackers are hijacking MCP tools mid-conversation, after the handshake.

Tomorrow's issue, in your inbox.

Source trail

Provenance