$ ls local-llm/reviews/ # the step-back reads
Local LLMs in review
One review per week (and month): what shipped, what broke, what to act on.
Nov 2–8, 2026 weekly [act] 8 stories
vLLM 0.33.0 ships as an unauthenticated crash resurfaces
The advisory is an incomplete fix of GHSA-25q3-v2hm-8vpf; the release lands 927 commits from 389 contributors, 122 of them new.
Oct 26–Nov 1, 2026 weekly [act] 8 stories
vLLM leaks heap addresses through lowercase regex
Two vLLM security fixes lead the period: the address leak plus a ReDoS in the validation error path sanitizer.
Oct 19–25, 2026 weekly [act] 8 stories
Three vLLM breaking issues expose default-server attack paths
Permissive CORS, forged multimodal SHM handles, and a DeepSeek-OCR image that kills the engine all ship in default config.
Oct 12–18, 2026 weekly [act] 8 stories
vLLM v0.32.0 lands systemone, security fixes behind it
llama.cpp's v0.7.0 shipped a host-side MoE expert cache and speculative decoding, while vLLM closed a ZMQ socket exhaustion hole and tigh...
Oct 5–11, 2026 weekly [act] 8 stories
vLLM 0.31.0 ships, and six security holes close
The release lands alongside a cluster of remote memory-exhaustion, RCE, and SSRF advisories affecting self-hosted vLLM deployments.
Oct 1–31, 2026 monthly [act] 8 stories
vLLM closes nine request-path holes in October
Unauthenticated requests can hang, corrupt, or execute code on default vLLM 0.31.x deployments; 0.31.0 and 0.32.0 also shipped.
Sep 28–Oct 4, 2026 weekly [act] 8 stories
vLLM fixes cache_salt oracle and multimodal IPC desync
Three breaking vLLM items outrank the week's features: a cross-tenant cache oracle, an IAMF heap overflow, and a desyncing multimodal cac...
Sep 21–27, 2026 weekly [act] 8 stories
Five engine-fatal flaws hit vLLM, llama.cpp ships v0.5.0
vLLM's disclosure covers request-handling paths that take down the shared EngineCore, affecting anyone serving structured output or multi...
Sep 14–20, 2026 weekly [act] 8 stories
Sparse attention lands across vLLM, SGLang, llama.cpp
Vulkan and SYCL backends got sparse Flash Attention and memory fixes, while SGLang gated Responses persistence off by default.
Sep 7–13, 2026 weekly [act] 7 stories
vLLM EngineCore can be killed by crafted stop_token_ids
Two vLLM reports this week show out-of-range stop_token_ids with min_tokens terminating EngineCore, including via the Rust HTTP/gRPC path.
Sep 1–30, 2026 monthly [act] 8 stories
vLLM 0.30.0 lands and nineteen engine-fatal flaws close
The month's security arc across vLLM, SGLang, and llama.cpp ranks above every feature ship.
Aug 31–Sep 6, 2026 weekly [act]
Vulkan top_k radix select arrives for long rows
llama.cpp's Vulkan backend adds top_k radix select; vLLM security patch bounds validation-error responses, and AutoRound FP8 support ships.
Aug 24–30, 2026 weekly
vLLM 0.28.0 lands with P2P weight sync, llama.cpp and partners broaden model reach
vLLM 0.28.0 shipped with 584 commits, headlined by a peer-to-peer weight synchronization for RL workloads, while llama.cpp, Ollama, and S...
Aug 1–31, 2026 monthly
Speculative decoding consolidates across the stack as DeepSeek V4 lands
August 2026 saw the local LLM stack converge on speculative decoding and DeepSeek V4, with every major engine shipping optimized paths.
Keep up with Local LLMs in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.
Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.