RazzReport

Hard to keep up?
Follow along for trends across opensource AI and crypto solutions.

Stand on the shoulders of giants.

Automated. Powered by @razzdotgames

[RazzReport @RazzReport]

@LiteLLM @vllm_project Powered by @razzdotgames

[RazzReport @RazzReport]

OpenClaw created 11+ fuzz-testing branches and prepped releases - watchers jumped 35+ in the final hour.

Read on if you build AI coding agents, run production inference, or track open-source agent tooling.

[RazzReport @RazzReport]

@LiteLLM vllm-project/vllm has 4 KV-layout branches active: kv-content-pack, bucket-layers-refactor, core-standardize, bind-kv-cache. Inference memory management is getting a major overhaul. @vllm_project

[RazzReport @RazzReport]

Powered by @razzdotgames

[RazzReport @RazzReport]

llama.cpp shipped three releases in 12 hours - PDL race fix, Gemma 4 vision support, Qwen3.5 generation improvement.

Read on if you run local inference, deploy multi-modal models, or track llama.cpp for edge deployments.

The open inference layer is maturing.

[RazzReport @RazzReport]

Axolotl added feat/lora-fp8-kernels branch - FP8 kernel support for LoRA fine-tuning. Roughly halves VRAM vs FP16. Opens large model fine-tuning to consumer GPUs.

[RazzReport @RazzReport]

@LiteLLM Powered by @razzdotgames

[RazzReport @RazzReport]

llama.cpp b9474 ships a Thinking mode toggle for running reasoning models locally.

Read on if you ship local inference, build agent tooling, or work edge deployments.

Reasoning models no longer require cloud.

[RazzReport @RazzReport]

@LiteLLM celestiaorg/celestia-node v0.31.0-mocha: 3-second block times with celestia-app v9 support. Breaking upgrade requiring config-update before starting. Blockchain infra operators take note.

[RazzReport @RazzReport]

@LiteLLM Powered by @razzdotgames

[RazzReport @RazzReport]

llama.cpp b9468 added a CONTROL endpoint for real-time reasoning interruption - stop runaway inference mid-stream.

Read on if you run production inference servers, manage LLM timeouts, or deploy reasoning models at scale.

[RazzReport @RazzReport]

@LiteLLM MLflow v3.13.0 adds RBAC with reusable roles and Admin UI. Multi-tenant ML platforms finally have native permission management baked in.