Who contributed to Local LLMs on August 14, 2026?

16 developers shipped this update, including ParthSareen, dhiltgen, ngxson, arthw, malsbat, Titaniumtown, Zijun Yu, and Juan Pérez de Algaba, and 8 more.

What were the notable Local LLMs updates?

launch: add DeepSeek Harness integration, launch: add Muse Code integration, and model/renderers: match Muse Glimmer reasoning template.

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

OLLAMA LAUNCHES DEEPSEEK AND MUSE INTEGRATIONS WHILE VLLM SHIPS HARDWARE-AGNOSTIC MODELING

By RepoJournal · Filed 07:02 UTC on August 14, 2026 · About Local LLMs

16 people shipped this

dhiltgen @dhiltgen 2 cited

ParthSareen @ParthSareen 1 cited

ngxson @ngxson 1 cited

malsbat @malsbat 1 cited

Titaniumtown @Titaniumtown 1 cited

Juan Pérez de Algaba @jperezdealgaba 1 cited

bohnstingl @bohnstingl 1 cited

jasonozuzu-cohere @jasonozuzu-cohere

michaelzhang-ai @michaelzhang-ai

kpham-sgl @kpham-sgl

merrymercy @merrymercy

Ollama just opened its launch ecosystem to DeepSeek Harness and Meta's Muse Code, while vLLM pivots to HuggingFace transformers as a unified backend for tail models.

Ollama's launch system now integrates DeepSeek Harness [1] and Muse Code [2], giving developers CLI-native access to these tools with local or cloud Ollama models through a single command. The Harness integration wires web search directly into Ollama's model catalog, while Muse gets its own isolated config root to avoid collision with Meta-backed installs. Meanwhile, Ollama fixed a correctness bug in Muse Glimmer's reasoning template [3], normalizing "Reasoning effort" to "Reasoning strength" and preventing duplicate reasoning directives that could confuse the model. On the vLLM side, the project shipped a hardware-agnostic modeling layer [4] that lets tail models live in HuggingFace transformers and execute through vLLM via the transformers backend, a major shift toward reducing fragmentation across hardware. Performance gains landed across the stack: llama.cpp corrected metric counting [6] to properly track processed vs. cached tokens, SYCL added Q3_K ESIMD kernels and fused unary operations with multiplication for measurable throughput lifts [7] [8], and sglang cut diffusion denoise time by 13.1% on H100 by hoisting rope cache and fusing per-token adaLN operations [9]. Breaking security change: vLLM now enforces audio decode duration limits in NanoNemotronVL [5] to prevent resource exhaustion attacks.

Action items

→ Review vLLM's new HF transformers backend [ref:12] if you're running tail models - it may simplify your deployment stack vllm-project/vllm [plan]
→ Update to latest ollama for DeepSeek Harness and Muse Code launch integrations [ref:1] [ref:3] ollama/ollama [plan]
→ Pull llama.cpp metric fixes [ref:6] if you depend on accurate token accounting in server mode ggml-org/llama.cpp [monitor]
→ Rebuild sglang models with latest diffusion optimizations [ref:16] for 13% latency win on H100 sgl-project/sglang [plan]

References

[1] launch: add DeepSeek Harness integration ↗ ollama/ollama
[2] launch: add Muse Code integration ↗ ollama/ollama
[3] model/renderers: match Muse Glimmer reasoning template ↗ ollama/ollama
[4] Hardware-agnostic model definition via HF transformer backend (1/N) ↗ vllm-project/vllm
[5] fix(security): enforce audio decode duration limit in NanoNemotronVL (#50221) vllm-project/vllm
[6] server: refactor + correctness fixes for metrics ↗ ggml-org/llama.cpp
[7] Add DMMV ESIMD Q3_K kernel ↗ ggml-org/llama.cpp
[8] sycl: fuse UNARY(silu|sigmoid|softplus) + MUL ↗ ggml-org/llama.cpp
[9] [diffusion] Wan2.2-TI2V: fuse per-token adaLN table add into contiguous slices + hoist rope cache (denoise -13.1% H100 / -12.6% H200, bit-exact; eager beats compile) ↗ sgl-project/sglang

Quick answers

What shipped in Local LLMs on August 14, 2026?: Ollama just opened its launch ecosystem to DeepSeek Harness and Meta's Muse Code, while vLLM pivots to HuggingFace transformers as a unified backend for tail models. In total, 125 commits, 125 pull requests, and 11 releases landed.
Who contributed to Local LLMs on August 14, 2026?: 16 developers shipped this update, including ParthSareen, dhiltgen, ngxson, arthw, malsbat, Titaniumtown, Zijun Yu, and Juan Pérez de Algaba, and 8 more.
What were the notable Local LLMs updates?: launch: add DeepSeek Harness integration, launch: add Muse Code integration, and model/renderers: match Muse Glimmer reasoning template.

QWEN3.8 LANDS IN OLLAMA WITH DEVELOPER INSTRUCTION SUPPORT

Ollama shipped full Qwen3.8 support overnight, adding reasoning-effort semantics and a critical fix for coding agents that send developer instructions before user messages.

+12

llms 208 shipped

Local LLMs Aug 13

OLLAMA PATCHES CRITICAL SSRF VULNERABILITY IN MANIFEST HANDLING

A rogue OCI registry could exploit duplicate digest collisions in Ollama's blob verification to exfiltrate data from internal endpoints.

+13

llms 249 shipped

Local LLMs Aug 12

OLLAMA CUTS REPEAT PENALTY BLOAT, LLAMA.CPP SHIPS POCKET-TTS SUPPORT

Ollama kills a decade of default penalty cruft that was drowning every model, while llama.cpp gains multimodal audio chops and vLLM fixes ROCm speculative decoding corruption.

+11

llms 258 shipped

Local LLMs Aug 11

OLLAMA PATCHES CRITICAL PARSER BUG, NEMOTRON 3 LANDS WITH SPECULATIVE DECODING

Ollama shipped v0.32.7 with a critical fix for boundary token corruption in model invocation, while full MLX support for Nemotron 3 Nano Omni brings self-draft speculation to inference without a separate draft model.

+11

llms 271 shipped

Elsewhere on the wire

AI Agents Aug 15

LANGCHAIN SHIPS ENCRYPTED REASONING FIX, CLINE ADDS WEB SEARCH AND VOICE

ChatOpenAI now preserves OpenAI's encrypted reasoning from streamed responses, while Cline releases three versions with provider-executed web search and microphone transcription.

ai-agents 55 shipped

@CachyOS Aug 15

FIRMWARE AND INIT SYNC LANDS AS AUTOS ROLL FORWARD

CachyOS synced linux-firmware with Arch and bumped mkinitcpio to 41.1-1, while the package automation pipeline keeps qview and qtscrcpy current across the AUR-derived repo.

infra 41 shipped

Elixir & Phoenix Aug 15

PHOENIX LIVE VIEW SHIPS UPLOAD OVERHAUL, ELIXIR TUNES CORE PERFORMANCE

SteffenDE landed five critical fixes for LiveView's upload handler in one sweep, while the Elixir core team optimized Version comparison and regex operations.

elixir 34 shipped

@dotnet Aug 15

ANDROID DOCS CLOSE 93 PLACEHOLDER GAPS, KESTREL GAINS EXPERIMENTAL DIRECTTLS TRANSPORT

Jonathan Peppers shipped five major documentation closures across Java atomic, time, and text APIs while AspNetCore lands an opt-in TLS transport that bypasses SslStream on Linux.