# zolotukhin.ai Full Documentation Index > Agent-oriented index for the ZINC documentation set. Site: https://zolotukhin.ai LLMs index: https://zolotukhin.ai/llms.txt Zig API root: https://zolotukhin.ai/zinc/docs/zig-api Zig API JSON: https://zolotukhin.ai/zinc/docs/zig-api.json Zig API Text: https://zolotukhin.ai/zinc/docs/zig-api.txt Last updated: 2026-07-21 ## Guidance - Treat the generated Zig API as the primary and canonical reference for ZINC internals. - Treat the Serving HTTP API as the external network surface only. - Prefer the JSON export for structured ingestion and the text export for lightweight retrieval. ## Zig API Sections ### CLI & Entrypoints Startup, argument parsing, and the top-level process path that wires model loading, tokenization, and generation together. URL: https://zolotukhin.ai/zinc/docs/zig-api#cli-entrypoints Coverage: 4 modules, 16 symbols - Build Info: https://zolotukhin.ai/zinc/docs/zig-api/build-info/ Source: src/build_info.zig Summary: Build metadata exported by `build.zig` for CLI version reporting. - CLI: https://zolotukhin.ai/zinc/docs/zig-api/main/ Source: src/main.zig Summary: CLI entrypoints for configuring ZINC and starting local inference. - Zig-struct-analyzer: https://zolotukhin.ai/zinc/docs/zig-api/zig-struct-analyzer/ Source: src/zig-struct-analyzer.zig Summary: Generated struct-layout probe used by the site Zig API docs. - CLI: https://zolotukhin.ai/zinc/docs/zig-api/main/ Source: src/zinc_rt/main.zig Summary: ZINC_RT backend entrypoint. ### Model Format & Loading GGUF parsing, metadata normalization, and the runtime structures that move weights from disk into GPU-resident buffers. URL: https://zolotukhin.ai/zinc/docs/zig-api#model-format-loading Coverage: 5 modules, 59 symbols - Config: https://zolotukhin.ai/zinc/docs/zig-api/config/ Source: src/model/config.zig Summary: Platform-independent model types shared by Vulkan and Metal backends. - GGUF: https://zolotukhin.ai/zinc/docs/zig-api/gguf/ Source: src/model/gguf.zig Summary: Parse GGUF container files and expose the metadata needed by the loader. - Loader Cuda: https://zolotukhin.ai/zinc/docs/zig-api/loader-cuda/ Source: src/model/loader_cuda.zig Summary: CUDA-specific model loading — mmap the GGUF, upload every tensor to the NVIDIA device, and expose them to the CUDA forward pass. - Loader Metal: https://zolotukhin.ai/zinc/docs/zig-api/loader-metal/ Source: src/model/loader_metal.zig Summary: Metal-specific model loading — zero-copy via mmap + newBufferWithBytesNoCopy. - Loader: https://zolotukhin.ai/zinc/docs/zig-api/loader/ Source: src/model/loader.zig Summary: Build runtime model state from GGUF metadata and GPU-resident tensor buffers. ### Tokenization Prompt and output text conversion between UTF-8 strings and token IDs used by the decode loop. URL: https://zolotukhin.ai/zinc/docs/zig-api#tokenization Coverage: 1 modules, 15 symbols - Tokenizer: https://zolotukhin.ai/zinc/docs/zig-api/tokenizer/ Source: src/model/tokenizer.zig Summary: Native BPE tokenizer that reads vocabulary and merge rules from GGUF metadata. ### Decode Planning Static graph construction and dependency ordering for the per-token compute work that the runtime records and submits. URL: https://zolotukhin.ai/zinc/docs/zig-api#decode-planning Coverage: 5 modules, 57 symbols - Graph: https://zolotukhin.ai/zinc/docs/zig-api/graph/ Source: src/compute/graph.zig Summary: Represent decode work as a dependency graph that can be topologically ordered. - Architecture: https://zolotukhin.ai/zinc/docs/zig-api/architecture/ Source: src/model/architecture.zig Summary: Build static decode graphs for the supported model families. - Graph: https://zolotukhin.ai/zinc/docs/zig-api/graph/ Source: src/zinc_rt/ir/graph.zig Summary: Shape-static ZINC_RT IR graph builder. - Op: https://zolotukhin.ai/zinc/docs/zig-api/op/ Source: src/zinc_rt/ir/op.zig Summary: ZINC_RT IR opcode definitions. - Verify: https://zolotukhin.ai/zinc/docs/zig-api/verify/ Source: src/zinc_rt/ir/verify.zig Summary: ZINC_RT IR verifier entrypoints. ### Inference Runtime Decode state, pipeline ownership, command recording, and token sampling inside the active inference loop. URL: https://zolotukhin.ai/zinc/docs/zig-api#inference-runtime Coverage: 35 modules, 462 symbols - Bench Hot Decode: https://zolotukhin.ai/zinc/docs/zig-api/bench-hot-decode/ Source: src/bench_hot_decode.zig Summary: Hot-path decode kernel microbenchmarks. - Bench Support: https://zolotukhin.ai/zinc/docs/zig-api/bench-support/ Source: src/bench_support.zig Summary: Shared helpers for benchmark and standalone runner entrypoints. - Forward Cuda Gemma: https://zolotukhin.ai/zinc/docs/zig-api/forward-cuda-gemma/ Source: src/compute/forward_cuda_gemma.zig Summary: CUDA forward pass for the dense gemma4 transformer (Gemma 4 31B-it). - Forward Cuda: https://zolotukhin.ai/zinc/docs/zig-api/forward-cuda/ Source: src/compute/forward_cuda.zig Summary: CUDA forward pass for the dense `qwen35` hybrid-SSM model (Qwen 3.5 9B). - Forward Metal: https://zolotukhin.ai/zinc/docs/zig-api/forward-metal/ Source: src/compute/forward_metal.zig Summary: Metal inference engine — decode loop for Apple Silicon. - Forward Zinc Rt: https://zolotukhin.ai/zinc/docs/zig-api/forward-zinc-rt/ Source: src/compute/forward_zinc_rt.zig Summary: ZINC_RT forward-pass bring-up. - Forward: https://zolotukhin.ai/zinc/docs/zig-api/forward/ Source: src/compute/forward.zig Summary: Run the inference runtime: decode state, pipeline ownership, and token generation. - Interface: https://zolotukhin.ai/zinc/docs/zig-api/interface/ Source: src/gpu/interface.zig Summary: GPU backend abstraction — comptime-resolved, zero runtime overhead. - Memory Plan: https://zolotukhin.ai/zinc/docs/zig-api/memory-plan/ Source: src/gpu/memory_plan.zig Summary: Shared runtime memory accounting helpers for Vulkan and Metal backends. - Process Lock: https://zolotukhin.ai/zinc/docs/zig-api/process-lock/ Source: src/gpu/process_lock.zig Summary: Cross-process GPU reservation lock keyed by backend and selected device. - Batching: https://zolotukhin.ai/zinc/docs/zig-api/batching/ Source: src/zinc_rt/batching.zig Summary: Tenant-aware batch planning for ZINC_RT. - Engine: https://zolotukhin.ai/zinc/docs/zig-api/engine/ Source: src/zinc_rt/engine.zig Summary: ZINC_RT — the ZINC Runtime. - Fast Pool: https://zolotukhin.ai/zinc/docs/zig-api/fast-pool/ Source: src/zinc_rt/fast_pool.zig Summary: Low-overhead worker pool for the T-CPU decode matvec fan-out. - Dequant: https://zolotukhin.ai/zinc/docs/zig-api/dequant/ Source: src/zinc_rt/isa/cpu_zig/dequant.zig Summary: Shared scalar GGML dequantization helpers for T-CPU kernels. - Embed: https://zolotukhin.ai/zinc/docs/zig-api/embed/ Source: src/zinc_rt/isa/cpu_zig/embed.zig Summary: T-CPU EMBED implementation. - Flash Attn: https://zolotukhin.ai/zinc/docs/zig-api/flash-attn/ Source: src/zinc_rt/isa/cpu_zig/flash_attn.zig Summary: T-CPU flash attention (single-query decode) implementation. - Lm Head: https://zolotukhin.ai/zinc/docs/zig-api/lm-head/ Source: src/zinc_rt/isa/cpu_zig/lm_head.zig Summary: T-CPU LM_HEAD implementation. - Matvec: https://zolotukhin.ai/zinc/docs/zig-api/matvec/ Source: src/zinc_rt/isa/cpu_zig/matvec.zig Summary: T-CPU matrix-vector projection implementation. - Mod: https://zolotukhin.ai/zinc/docs/zig-api/mod/ Source: src/zinc_rt/isa/cpu_zig/mod.zig Summary: Pure Zig T-CPU opcode implementations. - Moe Gate Topk: https://zolotukhin.ai/zinc/docs/zig-api/moe-gate-topk/ Source: src/zinc_rt/isa/cpu_zig/moe_gate_topk.zig Summary: T-CPU MOE_GATE_TOPK implementation. - Residual Rms Norm: https://zolotukhin.ai/zinc/docs/zig-api/residual-rms-norm/ Source: src/zinc_rt/isa/cpu_zig/residual_rms_norm.zig Summary: T-CPU residual add + RMS norm implementation. - Rms Norm: https://zolotukhin.ai/zinc/docs/zig-api/rms-norm/ Source: src/zinc_rt/isa/cpu_zig/rms_norm.zig Summary: T-CPU RMS_NORM implementation. - Rope: https://zolotukhin.ai/zinc/docs/zig-api/rope/ Source: src/zinc_rt/isa/cpu_zig/rope.zig Summary: T-CPU RoPE (Rotary Positional Embedding) implementation. - Sigmoid Mul: https://zolotukhin.ai/zinc/docs/zig-api/sigmoid-mul/ Source: src/zinc_rt/isa/cpu_zig/sigmoid_mul.zig Summary: T-CPU sigmoid-gated multiply implementation. - Swiglu: https://zolotukhin.ai/zinc/docs/zig-api/swiglu/ Source: src/zinc_rt/isa/cpu_zig/swiglu.zig Summary: T-CPU SwiGLU implementation. - Vadd: https://zolotukhin.ai/zinc/docs/zig-api/vadd/ Source: src/zinc_rt/isa/cpu_zig/vadd.zig Summary: T-CPU element-wise vector addition implementation. - Kmd: https://zolotukhin.ai/zinc/docs/zig-api/kmd/ Source: src/zinc_rt/kmd.zig Summary: Thin AMDGPU kernel-driver queries used by direct ZINC_RT tiers. - Lib: https://zolotukhin.ai/zinc/docs/zig-api/lib/ Source: src/zinc_rt/lib.zig Summary: ZINC_RT reference-runtime module. - Cpu: https://zolotukhin.ai/zinc/docs/zig-api/cpu/ Source: src/zinc_rt/ring/cpu.zig Summary: T-CPU ring backend. - Cs: https://zolotukhin.ai/zinc/docs/zig-api/cs/ Source: src/zinc_rt/ring/cs.zig Summary: AMDGPU DRM command-submission (CS) path — bring-up of the RADV / radeonsi PM4 submission foundation. - Kfd: https://zolotukhin.ai/zinc/docs/zig-api/kfd/ Source: src/zinc_rt/ring/kfd.zig Summary: AMDGPU KFD (`/dev/kfd`) bring-up for the T1 PM4-direct tier. - Mod: https://zolotukhin.ai/zinc/docs/zig-api/mod/ Source: src/zinc_rt/ring/mod.zig Summary: Backend-neutral packet batch types for ZINC_RT rings. - Packet List: https://zolotukhin.ai/zinc/docs/zig-api/packet-list/ Source: src/zinc_rt/ring/packet_list.zig Summary: Dynamic packet list for building per-token decode sequences. - Packet: https://zolotukhin.ai/zinc/docs/zig-api/packet/ Source: src/zinc_rt/ring/packet.zig Summary: PM4 packet builder shared by direct AMD ZINC_RT tiers. - Umq: https://zolotukhin.ai/zinc/docs/zig-api/umq/ Source: src/zinc_rt/ring/umq.zig Summary: AMDGPU user-mode queue (T2) availability and create/free smoke gate. ### Sampling Logit post-processing, argmax helpers, and token-selection controls layered on top of the decode runtime. URL: https://zolotukhin.ai/zinc/docs/zig-api#sampling Coverage: 2 modules, 8 symbols - Argmax: https://zolotukhin.ai/zinc/docs/zig-api/argmax/ Source: src/compute/argmax.zig Summary: Wrap the GPU argmax reduction used for greedy token sampling. - Argmax: https://zolotukhin.ai/zinc/docs/zig-api/argmax/ Source: src/zinc_rt/isa/cpu_zig/argmax.zig Summary: T-CPU ARGMAX implementation. ### Shader Dispatch Typed wrappers around the compute shaders that prepare push constants, descriptor layouts, and per-op dispatch dimensions. URL: https://zolotukhin.ai/zinc/docs/zig-api#shader-dispatch Coverage: 4 modules, 152 symbols - Attention: https://zolotukhin.ai/zinc/docs/zig-api/attention/ Source: src/compute/attention.zig Summary: Wrap the flash-attention compute shader and its dispatch parameters. - DMMV: https://zolotukhin.ai/zinc/docs/zig-api/dmmv/ Source: src/compute/dmmv.zig Summary: Wrap the decode-time matrix-vector shader family used for projection ops. - Elementwise: https://zolotukhin.ai/zinc/docs/zig-api/elementwise/ Source: src/compute/elementwise.zig Summary: Wrap the fused element-wise shader family used by the decode loop. - Runtime Assets: https://zolotukhin.ai/zinc/docs/zig-api/runtime-assets/ Source: src/runtime_assets.zig Summary: Runtime asset discovery for installed and source-tree ZINC layouts. ### Hardware Detection Vendor and architecture heuristics that translate raw Vulkan properties into tuning defaults for AMD, NVIDIA, and Intel GPUs. URL: https://zolotukhin.ai/zinc/docs/zig-api#hardware-detection Coverage: 3 modules, 15 symbols - Diagnostics Metal: https://zolotukhin.ai/zinc/docs/zig-api/diagnostics-metal/ Source: src/diagnostics_metal.zig Summary: Apple Silicon diagnostics and managed-model fit reporting for Metal. - Diagnostics: https://zolotukhin.ai/zinc/docs/zig-api/diagnostics/ Source: src/diagnostics.zig Summary: Vulkan system diagnostics (`zinc --check`). - GPU Detect: https://zolotukhin.ai/zinc/docs/zig-api/gpu-detect/ Source: src/vulkan/gpu_detect.zig Summary: Inspect the selected Vulkan device and derive architecture-specific tuning defaults. ### Vulkan Runtime Low-level Vulkan setup, memory allocation, buffers, pipelines, and command submission primitives used throughout the engine. URL: https://zolotukhin.ai/zinc/docs/zig-api#vulkan-runtime Coverage: 5 modules, 55 symbols - Buffer: https://zolotukhin.ai/zinc/docs/zig-api/buffer/ Source: src/vulkan/buffer.zig Summary: Allocate Vulkan buffers used by weights, intermediates, and staging copies. - Command: https://zolotukhin.ai/zinc/docs/zig-api/command/ Source: src/vulkan/command.zig Summary: Create reusable compute command pools and command buffers. - Instance: https://zolotukhin.ai/zinc/docs/zig-api/instance/ Source: src/vulkan/instance.zig Summary: Initialize Vulkan, select a compute-capable device, and expose memory utilities. - Pipeline: https://zolotukhin.ai/zinc/docs/zig-api/pipeline/ Source: src/vulkan/pipeline.zig Summary: Load SPIR-V compute shaders into Vulkan pipelines. - Tooling: https://zolotukhin.ai/zinc/docs/zig-api/tooling/ Source: src/vulkan/tooling.zig Summary: Standalone Vulkan validation-tool imports. ### Metal Runtime Low-level Metal device discovery, buffers, pipelines, and command submission primitives used by the Apple Silicon backend. URL: https://zolotukhin.ai/zinc/docs/zig-api#metal-runtime Coverage: 6 modules, 51 symbols - Buffer: https://zolotukhin.ai/zinc/docs/zig-api/buffer/ Source: src/metal/buffer.zig Summary: Metal buffer wrapper — shared-mode GPU buffers with zero-copy mmap support. - C: https://zolotukhin.ai/zinc/docs/zig-api/c/ Source: src/metal/c.zig Summary: Shared C import for the Metal shim — all Metal modules import from here to ensure type identity across compilation units. - Command: https://zolotukhin.ai/zinc/docs/zig-api/command/ Source: src/metal/command.zig Summary: Metal command buffer wrapper — dispatch recording and GPU synchronization. - Device: https://zolotukhin.ai/zinc/docs/zig-api/device/ Source: src/metal/device.zig Summary: Metal device wrapper — macOS Apple Silicon GPU backend. - Kernel Timing: https://zolotukhin.ai/zinc/docs/zig-api/kernel-timing/ Source: src/metal/kernel_timing.zig Summary: Per-kernel Metal dispatch timing probe — default-off, env-flag-gated. - Pipeline: https://zolotukhin.ai/zinc/docs/zig-api/pipeline/ Source: src/metal/pipeline.zig Summary: Metal compute pipeline wrapper — MSL source or precompiled metallib. ### Managed Models Catalog metadata, cache management, model downloads, and active-selection helpers used by the CLI and server. URL: https://zolotukhin.ai/zinc/docs/zig-api#managed-models Coverage: 4 modules, 68 symbols - Catalog: https://zolotukhin.ai/zinc/docs/zig-api/catalog/ Source: src/model/catalog.zig Summary: Curated catalog of ZINC-supported managed GGUF models. - Hf: https://zolotukhin.ai/zinc/docs/zig-api/hf/ Source: src/model/hf.zig Summary: Resolve and download Hugging Face GGUF models for the `-hf` CLI flag. - Managed: https://zolotukhin.ai/zinc/docs/zig-api/managed/ Source: src/model/managed.zig Summary: Managed model cache, active-model selection, and download helpers. - Model Manager: https://zolotukhin.ai/zinc/docs/zig-api/model-manager/ Source: src/server/model_manager.zig Summary: Managed active-model runtime state for the HTTP server and CLI startup. ### Scheduler Continuous batching scheduler, paged KV cache management, and request lifecycle tracking for concurrent inference serving. URL: https://zolotukhin.ai/zinc/docs/zig-api#scheduler Coverage: 3 modules, 32 symbols - KV Cache: https://zolotukhin.ai/zinc/docs/zig-api/kv-cache/ Source: src/scheduler/kv_cache.zig Summary: Paged KV cache manager for concurrent request serving. - Request: https://zolotukhin.ai/zinc/docs/zig-api/request/ Source: src/scheduler/request.zig Summary: Request lifecycle management for concurrent inference serving. - Scheduler: https://zolotukhin.ai/zinc/docs/zig-api/scheduler/ Source: src/scheduler/scheduler.zig Summary: Continuous-batching scheduler groundwork for concurrent inference requests. ### API Server OpenAI-compatible HTTP server, route dispatch, SSE streaming, and session management for serving inference over the network. URL: https://zolotukhin.ai/zinc/docs/zig-api#api-server Coverage: 5 modules, 65 symbols - Http: https://zolotukhin.ai/zinc/docs/zig-api/http/ Source: src/server/http.zig Summary: Minimal HTTP/1.1 server for the OpenAI-compatible inference API. - Model Manager Metal: https://zolotukhin.ai/zinc/docs/zig-api/model-manager-metal/ Source: src/server/model_manager_metal.zig Summary: Metal-backed active-model runtime state for the HTTP server. - Model Manager Runtime: https://zolotukhin.ai/zinc/docs/zig-api/model-manager-runtime/ Source: src/server/model_manager_runtime.zig Summary: Backend-selected model manager for the HTTP server. - Routes: https://zolotukhin.ai/zinc/docs/zig-api/routes/ Source: src/server/routes.zig Summary: Route dispatcher and endpoint handlers for the OpenAI-compatible API. - Runtime: https://zolotukhin.ai/zinc/docs/zig-api/runtime/ Source: src/server/runtime.zig Summary: Backend-specific server runtime aliases and wrappers. ### Tool Calling Tool-use protocol helpers: chat-template-aware tool definitions, argument parsing, and response formatting for function-calling-capable models. URL: https://zolotukhin.ai/zinc/docs/zig-api#tool-calling Coverage: 1 modules, 19 symbols - Tool Format: https://zolotukhin.ai/zinc/docs/zig-api/tool-format/ Source: src/server/tool_format.zig Summary: Pluggable tool-calling format dispatch for chat completions. ### CUDA Runtime CUDA device discovery, context management, device buffers, NVRTC-compiled pipelines, and stream-based command submission for the NVIDIA backend. URL: https://zolotukhin.ai/zinc/docs/zig-api#cuda-runtime Coverage: 9 modules, 44 symbols - Buffer: https://zolotukhin.ai/zinc/docs/zig-api/buffer/ Source: src/cuda/buffer.zig Summary: CUDA buffer wrapper — device-local allocations with optional pinned staging. - C: https://zolotukhin.ai/zinc/docs/zig-api/c/ Source: src/cuda/c.zig Summary: Shared C import for the CUDA shim — all CUDA backend modules import from here to ensure type identity across compilation units (mirrors src/metal/c.zig). - Command: https://zolotukhin.ai/zinc/docs/zig-api/command/ Source: src/cuda/command.zig Summary: CUDA command wrapper — kernel dispatch and stream/event synchronization (mirrors src/metal/command.zig). - Device: https://zolotukhin.ai/zinc/docs/zig-api/device/ Source: src/cuda/device.zig Summary: CUDA device wrapper — NVIDIA GPU backend (mirrors src/metal/device.zig). - Pipeline: https://zolotukhin.ai/zinc/docs/zig-api/pipeline/ Source: src/cuda/pipeline.zig Summary: CUDA compute pipeline wrapper — NVRTC-compiled CUfunction (mirrors src/metal/pipeline.zig). - Smoke: https://zolotukhin.ai/zinc/docs/zig-api/smoke/ Source: src/cuda/smoke.zig Summary: Standalone smoke test for the ZINC CUDA backend Zig wrapper layer. Drives the GPU entirely through device.zig / buffer.zig / pipeline.zig / command.zig (which wrap cuda_shim.c) — proving the Zig<->CUDA seam: device select, staged buffers + H2D/D2H, NVRTC runtime compile, the buffers+push dispatch ABI, and both sync and async commit paths. - Dbg Cuda: https://zolotukhin.ai/zinc/docs/zig-api/dbg-cuda/ Source: src/dbg_cuda.zig Summary: CUDA forward-pass debug harness for the qwen35 hybrid-SSM model. Two modes: - Loadtest Cuda: https://zolotukhin.ai/zinc/docs/zig-api/loadtest-cuda/ Source: src/loadtest_cuda.zig Summary: Standalone load-test for src/model/loader_cuda.zig. - Run Cuda: https://zolotukhin.ai/zinc/docs/zig-api/run-cuda/ Source: src/run_cuda.zig Summary: Standalone CUDA greedy-decode driver for the qwen35 forward pass. ### CUDA Multi-Tenant Serving Engine Continuous-batching serving worker, request registry, per-token streaming channels, and GPU-owned decode loop for concurrent CUDA inference. URL: https://zolotukhin.ai/zinc/docs/zig-api#cuda-multi-tenant-serving-engine-effort-28-increment-3-3b Coverage: 1 modules, 12 symbols - Cuda Serve: https://zolotukhin.ai/zinc/docs/zig-api/cuda-serve/ Source: src/server/cuda_serve.zig Summary: One GPU **worker thread** runs the continuous-batching loop (admit → prefill → `decodeBatch` → evict); many transport (HTTP) handler threads `submit` requests concurrently and **stream each request's own tokens incrementally** as the worker produces them. This is the threading model proven token-identical to N isolated single-sequence runs by the `dbg_cuda serve` harness (3a); 3b factors it into a reusable engine and adds the per-token streaming registry the HTTP/SSE transport needs. ## Supplemental ZINC Docs - ZINC Technical Specification: https://zolotukhin.ai/zinc/docs/spec Summary: Architecture, GPU kernels, model support, scheduler behavior, and system design context around the runtime. - TurboQuant KV Cache Compression: https://zolotukhin.ai/zinc/docs/turboquant-spec Summary: Two-stage vector quantization, residual correction, and memory-reduction tradeoffs for the paged KV-cache path. - RDNA4 GPU Tuning Guide: https://zolotukhin.ai/zinc/docs/rdna4-tuning Summary: Performance profiling for AMD GPUs, cooperative matrix behavior, SPIR-V toolchain notes, and bandwidth tuning. - Serving HTTP API: https://zolotukhin.ai/zinc/docs/api Summary: OpenAI-compatible request and response shapes for clients talking to ZINC. This is the network API, not the Zig code reference. - AMD RDNA3/RDNA4 GPU Reference for Inference: https://zolotukhin.ai/zinc/docs/amd-gpu-reference Summary: ZINC documentation. - Apple Metal Reference: https://zolotukhin.ai/zinc/docs/apple-metal-reference Summary: ZINC documentation. - Apple Silicon Metal Enablement Notes: https://zolotukhin.ai/zinc/docs/apple-silicon-metal-enablement Summary: ZINC documentation. - Apple Silicon Architecture Reference: https://zolotukhin.ai/zinc/docs/apple-silicon-reference Summary: ZINC documentation. - Benchmarking ZINC: https://zolotukhin.ai/zinc/docs/benchmarking Summary: ZINC documentation. - CUDA backend for ZINC — design & implementation plan: https://zolotukhin.ai/zinc/docs/cuda-backend Summary: ZINC documentation. - Delta-Net: llama.cpp vs ZINC Analysis: https://zolotukhin.ai/zinc/docs/delta-net-llamacpp-analysis Summary: ZINC documentation. - Development Guide: https://zolotukhin.ai/zinc/docs/development Summary: ZINC documentation. - Getting started with ZINC: https://zolotukhin.ai/zinc/docs/getting-started Summary: ZINC documentation. - GPU Memory Scaling Plan: https://zolotukhin.ai/zinc/docs/gpu-memory-scaling-plan Summary: ZINC documentation. - ZINC hardware requirements: https://zolotukhin.ai/zinc/docs/hardware-requirements Summary: ZINC documentation. - Intel Arc B-Series GPU Reference for Inference: https://zolotukhin.ai/zinc/docs/intel-gpu-reference Summary: ZINC documentation. - Metal Performance Plan: https://zolotukhin.ai/zinc/docs/metal-performance-plan Summary: ZINC documentation. - ZINC MMQ v2 — Fused Q4_K×Q8_1 Tensor-Core GEMM (beat llama.cpp): https://zolotukhin.ai/zinc/docs/mmq-v2-design Summary: ZINC documentation. - NVIDIA Ampere/Ada/Blackwell GPU Reference for Inference: https://zolotukhin.ai/zinc/docs/nvidia-gpu-reference Summary: ZINC documentation. - Configure OpenCode with ZINC and Qwen: https://zolotukhin.ai/zinc/docs/opencode Summary: ZINC documentation. - Prefill Gap Closure Plan — ZINC CUDA vs llama.cpp: https://zolotukhin.ai/zinc/docs/prefill-gap-closure-plan Summary: ZINC documentation. - Releasing ZINC: https://zolotukhin.ai/zinc/docs/releasing Summary: ZINC documentation. - ZINC Roadmap: https://zolotukhin.ai/zinc/docs/roadmap Summary: ZINC documentation. - Running ZINC: https://zolotukhin.ai/zinc/docs/running-zinc Summary: ZINC documentation. - Verifying PR #25 on Qwen 3.6 27B: https://zolotukhin.ai/zinc/docs/testing-pr25-27b Summary: ZINC documentation. - ZINC_RT — The ZINC Runtime: https://zolotukhin.ai/zinc/docs/zinc-rt-design Summary: ZINC documentation. ## Other Surfaces - Blog: https://zolotukhin.ai/blog - RSS: https://zolotukhin.ai/feed.xml