Who contributed to Local LLMs on July 31, 2026?

12 developers shipped this update, including dhiltgen, Aleksander Grygier, timkhronos, Georgi Gerganov, Juan Pérez de Algaba, bigPYJ1151, RyanClark2k, and ZJY0516, and 4 more.

What were the notable Local LLMs updates?

ui: IndexedDB and Conversations data fixes (#26278), ui: IndexedDB and Conversations data fixes, and [CPU] Bump up CPU kernels to latest version.

Local LLMs

Ollama, llama.cpp, vLLM, and SGLang - running and serving open models locally

Pick a date

Topics: Local LLMs AI / ML Full archive →

The Wire · Showcase

LLAMA.CPP FIXES DATA CORRUPTION BUG, VLLM SHIPS CPU KERNEL OVERHAUL

By RepoJournal · Filed 07:01 UTC on July 31, 2026 · About Local LLMs

12 people shipped this

rainj-me @rainj-me 3 cited

Aleksander Grygier @allozaur 2 cited

bigPYJ1151 @bigPYJ1151 1 cited

RyanClark2k @RyanClark2k 1 cited

ZJY0516 @ZJY0516 1 cited

TheEpicDolphin @TheEpicDolphin 1 cited

vroomfondel @vroomfondel 1 cited

dhiltgen @dhiltgen

timkhronos @timkhronos

Georgi Gerganov @ggerganov

Juan Pérez de Algaba @jperezdealgaba

hzh0425 @hzh0425

llama.cpp plugged a critical conversation data loss hole, while vLLM synchronized CPU kernels with upstream sglang and shipped Kimi K3 support across the local LLM stack.

llama.cpp's IndexedDB transaction fix [1] addresses a cascading delete bug that was orphaning messages when users edited or regenerated responses. The fix makes all message-tree writes atomic, preventing torn writes that left conversations in corrupt states [2]. In the same release cycle, vLLM bumped CPU kernels to sync with sglang's latest main branch [3], incorporating MXFP4 W4A16 MoE support, AMX GDN dispatch, and RISC-V scalar optimization. vLLM also landed Marlin weight reload fixes [4] verified across GPU validation and CPU regression tests, plus native Kimi K3 model support [5] requiring flashinfer v0.6.16rc5. On the speculative decode front, vLLM's multi-layer MTP speculator [6] extends the Inkling model's 8-token capability into Model Runner V2. Meanwhile, sglang expanded Qwen3.5-MoE quantization to handle modelopt_fp4 checkpoints that partially quantize attention [7], and began wiring Rust server modules into the runtime [8] [9] [10] as part of a larger infrastructure refactor. llama.cpp also refactored its test suite to avoid redundant builds and refreshed documentation .

Action items

→ Deploy llama.cpp conversation data fix immediately if running chat UI in production ggml-org/llama.cpp [immediate]
→ Sync vLLM CPU kernels from sglang main before next batch job vllm-project/vllm [plan]
→ Test Kimi K3 with flashinfer v0.6.16rc5 if running vLLM on Qwen models vllm-project/vllm [monitor]
→ Monitor sglang Rust server wiring for stability before adopting sgl-project/sglang [monitor]

References

[1] ui: IndexedDB and Conversations data fixes (#26278) ggml-org/llama.cpp
[2] ui: IndexedDB and Conversations data fixes ↗ ggml-org/llama.cpp
[3] [CPU] Bump up CPU kernels to latest version ↗ vllm-project/vllm
[4] [Bugfix] Preserve Marlin runtime tensor storage across weight reload ↗ vllm-project/vllm
[5] [New model] Kimi K3 ↗ vllm-project/vllm
[6] [Model Runner V2][Spec Decode] Add multi-layer MTP speculator ↗ vllm-project/vllm
[7] Qwen3.5-MoE: support modelopt_fp4 checkpoints that quantize attention (+ load baked FP8 KV scales) ↗ sgl-project/sglang
[8] wire the rust server modules into lib, runtime, and tokenizer manager ↗ sgl-project/sglang
[9] add the rust server native api handlers and runtime threads ↗ sgl-project/sglang
[10] add the rust server api frame codec and http server entry ↗ sgl-project/sglang

Quick answers

What shipped in Local LLMs on July 31, 2026?: llama.cpp plugged a critical conversation data loss hole, while vLLM synchronized CPU kernels with upstream sglang and shipped Kimi K3 support across the local LLM stack. In total, 120 commits, 119 pull requests, and 10 releases landed.
Who contributed to Local LLMs on July 31, 2026?: 12 developers shipped this update, including dhiltgen, Aleksander Grygier, timkhronos, Georgi Gerganov, Juan Pérez de Algaba, bigPYJ1151, RyanClark2k, and ZJY0516, and 4 more.
What were the notable Local LLMs updates?: ui: IndexedDB and Conversations data fixes (#26278), ui: IndexedDB and Conversations data fixes, and [CPU] Bump up CPU kernels to latest version.

OLLAMA FIXES KV CACHE CORRUPTION, LLAMA.CPP HARDENS FILE SEARCH, VLLM KILLS MOE LEGACY CODE

Ollama patched a critical cache corruption bug that corrupts lazy snapshots during multi-request prefills, while llama.cpp hardened Windows junction traversal and vLLM removed 14 hardcoded MoE shards across quantization methods.

+10

llms 240 shipped

Local LLMs Aug 5

OLLAMA FIXES RACE CONDITIONS AS LLAMA.CPP SHIPS BREAKING TTS CHANGES

Ollama patched critical data races across its progress tracking system while llama.cpp introduces incompatible Qwen3-TTS support, and vLLM lands kernel optimizations that cut latency on specialized hardware.

+11

llms 246 shipped

Local LLMs Aug 4

OLLAMA REWIRES OPENAI COMPATIBILITY, LLAMA.CPP HARDENS SAMPLING

Ollama reworked its entire streaming protocol to match OpenAI's wire format chunk-for-chunk, unblocking drop-in replacement deployments.

+11

llms 230 shipped

Local LLMs Aug 3

DEEPSEEK V4 INFERENCE GETS 50% SPEEDUP; LLAMA.CPP AND VLLM BOTH SHIPPING OPTIMIZED PATHS

DeepSeek V4 speculative decode just went from nice-to-have to standard: llama.cpp landed MTP and DSpark support with roughly 50% speedup [ref:3], vLLM hardened its KV cache coordinator for the same models [ref:7], and Metal got native hyper-connection ops [ref:5].

llms 148 shipped

Elsewhere on the wire

AI Agents Aug 6

LANGCHAIN HARDENS ANTHROPIC TOOL CALLING AND GIT CREDENTIALS AS OPENHANDS SHIPS 1.10.0

LangChain fixed a critical credential leak in evaluation runs while Anthropic tool schemas get proper fallback handling, as three major agent frameworks push releases across the ecosystem.

ai-agents 54 shipped

@CachyOS Aug 6

WIKI TRANSLATION ROLLBACK CHAOS, MIRROR STACK HOLDING STEADY

CachyOS wiki maintainers are caught in a revert loop over German translations while the status infrastructure logs 55 commits without incident.

infra 57 shipped

Elixir & Phoenix Aug 6

ELIXIR BUMPS ZIZMOR SECURITY SCANNER TO 1.28.0

The Elixir repo updated its CI security action overnight, pulling in the latest zizmor static analysis tooling.

elixir 1 shipped

@dotnet Aug 6

DOTNET FOUNDATION PULLS iOS METRICS INTO FOCUS AS RUNTIME HARDENS HTTP PARSING

Performance team fixed iOS size reporting to measure only distributable packages, while runtime clamped down on RFC 9112 compliance for chunked encoding.