RepoJournal
Local LLMs Local LLMs
59 wires and counting

$ follow Local LLMs

Keep up with Local LLMs in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

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.

$ status

wire 2026-07-27
stories 125
RSS Archive Reviews Topics

© 2026 RepoJournal Home Showcase Explore How it works Privacy

$ the-wire · showcase

LLAMA.CPP SHIPS MINIMAX-M3 SPARSE ATTENTION, VLLM UNLOCKS HETEROGENEOUS PREFILL-DECODE

By RepoJournal · Filed · About Local LLMs · Composed from the cited sources · methodology

timkhronos Pascal github-actions[bot] edwinlim0919 yushengsu-thu mickqian dhiltgen
14 people shipped this
timkhronos @timkhronos 3 cited Pascal @ServeurpersoCom 2 cited github-actions[bot] @github-actions[bot] 1 cited edwinlim0919 @edwinlim0919 1 cited yushengsu-thu @yushengsu-thu 1 cited mickqian @mickqian 1 cited dhiltgen @dhiltgen AndreasKaratzas @AndreasKaratzas walterbm @walterbm zxd1997066 @zxd1997066 njhill @njhill sshleifer @sshleifer decajoin @decajoin xdtbynd @xdtbynd

The inference stack gets a new model architecture and a fundamental shift in how prefill and decode can run on different hardware configurations.

Llama.cpp landed full MiniMax-M3 support [2] [4] with vision tower integration, adding sparse attention capability to handle 60-layer, 128-expert MoE models where "a lightweight per-GQA-group indexer scores the visible causal context, max-pools the scores into 128-token key blocks, and selects the top-16 blocks" [6]. The model joins an expanding list of specialized architectures getting native dispatch. On the vLLM side, MoRIIO heterogeneous prefill-decode routing [7] is now live, enabling prefill and decode phases to run different parallelism strategies (TP vs DP+DP) on the same cluster, yielding measurably higher throughput and lower TTFT at scale on MI300X. SGLang shipped LoRA support under breakable CUDA graphs [8], restoring prefill graph acceleration to LoRA workloads, and added EmbeddingGemma [9] with bidirectional attention and mean pooling. Meanwhile, llama.cpp fixed critical routing bugs for model names with slashes [5], resolved iOS conversation import by detecting format from file contents [3], and Ollama pushed an mlx update [1]. This period saw 60 commits and 60 PRs across 4 core repos, with 5 releases shipping.

Action items

References

  1. [1] v0.32.5 ↗ ollama/ollama
  2. [2] model: Add MiniMax-M3 (MSA: MiniMax Sparse Attention) support (#24908) ↗ ggml-org/llama.cpp
  3. [3] ui: detect the conversation import format from file contents (#26121) ↗ ggml-org/llama.cpp
  4. [4] mtmd: Add Vision Support for Minimax-M3 (#25113) ↗ ggml-org/llama.cpp
  5. [5] server + ui: fix stream routes for model names containing a slash (#26137) ↗ ggml-org/llama.cpp
  6. [6] Add MiniMax-M3 (MSA: MiniMax Sparse Attention) support ↗ ggml-org/llama.cpp
  7. [7] [Core][KV-transfer] MoRIIO: heterogeneous TP<->DP prefill/decode read routing ↗ vllm-project/vllm
  8. [8] [LoRA] Support LoRA under the breakable/full prefill CUDA graph ↗ sgl-project/sglang
  9. [9] model: support EmbeddingGemma ↗ sgl-project/sglang

$ ls local-llm/reviews/ # the step-back reads

the week in review Sparse attention lands across vLLM, SGLang, llama.cpp the month in review Speculative decoding consolidates across the stack as DeepSeek V4 lands
Topics: Local LLMs AI / ML Full archive → Weekly reviews →

Quick answers

What shipped in Local LLMs on July 27, 2026?
The inference stack gets a new model architecture and a fundamental shift in how prefill and decode can run on different hardware configurations. In total, 60 commits, 60 pull requests, and 5 releases landed.
Who contributed to Local LLMs on July 27, 2026?
14 developers shipped this update, including github-actions[bot], dhiltgen, timkhronos, Pascal, AndreasKaratzas, walterbm, edwinlim0919, and zxd1997066, and 6 more.
What were the notable Local LLMs updates?
v0.32.5, model: Add MiniMax-M3 (MSA: MiniMax Sparse Attention) support (#24908), and ui: detect the conversation import format from file contents (#26121).
♥

$ log Local LLMs --recent

Metal MoE/SSM fusion lands, MiMo-V2.5 fp8 sharding fixed 158 shipped Ollama exposes thinking levels, llama.cpp widens Hexagon and architecture coverage 250 shipped GGUF offset corruption fixed, vLLM restores Dynamo KV metadata 244 shipped MLX engine graduates out of x/, ROCm fixes silent attention corruption 268 shipped

$ log --all # elsewhere on the wire

@openclaw Cold-start Gateway diagnostics stop crying wolf, doctor repairs plugin drift Agentic Coding Cline desktop bundles shrink, goose unblocks its own clicks

$ ls showcases/

Rails Django Laravel Node.js Go Vercel Vue.js Kubernetes HashiCorp Arch Linux Linux Anthropic

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.