Expert Streaming Engine v0.2.0

Expert Streaming Engine v0.2.0

This feature release makes ESE hardware-adaptive while preserving its defining
bounded NVMe → RAM → VRAM execution model.

Highlights

• Calibrates real CPU expert kernels and bounded expert uploads, then admits
  heterogeneous CPU/GPU MoE execution only when model-backed verification
  proves it beneficial and runtime telemetry remains within the measured
  envelope.
• Adds bounded double-buffered expert prefill streaming and adaptive
  per-device expert caches for mixed Turing/Ampere systems.
• Rebalances KV, expert-cache, MTP, and multimodal residency at idle safe
  points through reversible transactions; failed preparation or publication
  restores the exact prior physical and logical state.
• Supports configurable concurrent dense-model sessions and concurrency-aware
  Studio sweeps. MoE and hybrid modes remain single-session until equivalent
  multi-sequence parity evidence exists.
• Adds native Kimi Linear 48B-A3B GGUF execution with hybrid KDA/MLA layers,
  256-expert top-8 routing, and bounded sidecar-backed expert caching.
• Makes configuration sweeps execute a real inference workload during
  capacity validation and persist the exact multi-GPU split that was tested.

Validated hardware

The retained local evidence covers an RTX 2080 SUPER (Turing), RTX 3060 Ti and
RTX 3080 (Ampere), individually and in a heterogeneous three-GPU topology.
Ada-or-newer physical validation remains an explicit solo-maintainer waiver;
the release does not claim untested architecture-specific performance.

The Kimi Linear MXFP4_MOE gate allocated 65,536 tokens and produced a
deterministic 9.29 tok/s bounded-cache result on two Ampere GPUs. A separate
Qwen3.8 27B Q4_K_M 65,536-context profile reached 520.91 prompt tok/s and
approximately 25 generation tok/s across the three-GPU topology after its
verified tensor split was corrected to 32,32,36.

Existing model folders, app profiles, chats, sweep checkpoints, and telemetry
consent are preserved when updating from v0.1.x.

Expert Streaming Engine v0.1.2

Expert Streaming Engine v0.1.2

This hotfix restores NVIDIA GPU support to the packaged Windows ESE Studio
runtime and makes failures actionable.

Highlights

• Bundles a CUDA 12.4-enabled native Windows runtime plus the required CUDA and
  cuBLAS redistributable libraries, with CPU fallback when no NVIDIA GPU is
  available.
• Keeps hardware discovery usable on installed systems that do not have CMake
  or a compiler.
• Shows the native server diagnostic when a configuration sweep cannot start.
• Adds an editable, privacy-scrubbed GitHub bug report action to visible Studio
  errors.

Existing model folders, app profiles, chats, sweep checkpoints, and telemetry
consent are preserved when updating.

Expert Streaming Engine v0.1.1

Expert Streaming Engine v0.1.1

This maintenance release makes ESE Studio a unified installation and adds secure in-app updates.

Highlights

• Installs ESE Studio together with the ESE launcher and a matching native llama-server runtime.
• Adds signed update checks, visible download progress, verification, installation, and restart from Settings.
• Builds CUDA automatically from source when an NVIDIA toolchain is available, with a portable CPU runtime in published installers.
• Adds Linux and Windows dependency preflight and assisted installation.
• Expands the README with installation, verification, updater, and Studio screenshots.

Existing model folders, application profiles, chat history, and benchmark-sharing consent remain in place when updating.

Expert Streaming Engine v0.1.0

Expert Streaming Engine v0.1.0

The first supported ESE release combines bounded sparse-MoE inference with the
new Linux ESE Studio desktop control center.

ESE engine

• Transparent ese planning and serving for resident, hybrid, bounded-cache,
  and deferred-stream policies.
• Native global RAM/VRAM/context/KV/expert allocation with explicit reserves.
• Bounded NVMe → RAM → VRAM expert caching across one, two, or three GPUs.
• Turbo KV/TCQ/VBR foundations, transient module sharing, and adaptive
  speculation.
• Verified Turing and Ampere CUDA coverage. Ada-or-newer runtime coverage is
  not claimed because suitable hardware was unavailable.

ESE Studio for Linux

• Recursive GGUF discovery organized into collapsible model families.
• Configurable CLI applications in resizable embedded terminals, including
  automatic detection for Codex, Claude Code, OpenCode, and Hermes.
• Complete endpoint and model metadata handoff to endpoint-aware applications.
• Hardware-aware Hugging Face GGUF browsing with quant recommendations,
  resumable downloads, live speed, ETA, and cancellation.
• Real model-backed configuration sweeps that maximize safe context before
  tuning stable throughput, with advanced objectives and checkpoint/resume.
• Optional Help improve ESE sharing. Sanitized results enter a private
  collector; only groups with at least three samples can become public.

Packages

The release includes DEB and RPM packages for x86-64 Linux plus SHA256SUMS.
ESE Studio requires the ese launcher and a locally built ESE runtime for model
serving. Windows, automatic updates, signed packages, and AppImage distribution
are deferred beyond this Linux-first release.

See the main guide,
Studio guide,
and benchmark record
for installation, operation, and measured reference results.