ne írj semmi mást csak a teljes fájlokat es kommentek nem le...
Prompt
ne írj semmi mást csak a teljes fájlokat es kommentek nem lehetnek benne! soha semmi egyszerusitett mock placeholder dummy szimulalt fake szart nem engedelyezek es teljes fájl roviditetlen production ready kód nem lehet trancutted nem lehet olyan hogy …és hasonlóan 50 xy nem lehet dummy to do sorry hiányosság minden fájl teljes kódját egyesével fájkba írod semmi mást nem írsz ezen kívűl I want to construct from scratch a complete, multi-language foundation language model training and inference platform using a Reversible Scatter Flow paradigm instead of traditional transformers, so that I have a fully functioning, mathematically verified, hardware-accelerated distributed machine learning runtime. Construct the entire ecosystem step by step with the following structure: 1. Core Libraries (Zig): Implement the custom tensor math layer, durable file I/O operations, learned embedding engines, and specialized multi-strategy memory allocators (Buddy, Slab, Arena, lock-free structures). 2. Network Architecture (Zig): Implement the Reversible Scatter Flow stack (bijective cross-affine coupling) avoiding standard self-attention, establishing $O(\text{dim})$ memory limits during backpropagation. Integrate an Objective Tensor Backend and Spectral Fisher Diagonalizer (SFD) optimizer enforcing max singular values. 3. Relational Core & Graph Memory (Zig): Discard standard KV-cache. Construct a temporal, versioned knowledge graph engine and a "Chaos Core Kernel" managing node entanglements and signals. Add a Surprise-Based Memory Manager mapping Jaccard distance/temporal novelty, and an Entangled Stochastic Symmetry Optimizer to orchestrate logic flow. 4. Tokenization & Search (Zig): Implement a Morpho-Graph Tokenizer using byte-pair merging with semantic prefixes/suffixes handling, and a Succinct Semantic Index processor utilizing locality-sensitive MinHash algorithms. 5. Distributed Processing & Interop (Zig/C): Provide a robust trainer wrapper linking peer-to-peer networking bindings via NCCL/CUDA protocols, managing data staging across distributed node memory. Include the necessary ABI bindings logic. 6. Verification Engines (Zig): Include standalone engines ensuring runtime structural safety—construct systems for Information Flow security proofing, Hoare Logic invariant generation, and a Type Theory engine managing theorem derivations over tensor modifications. 7. Accelerator Kernels (Futhark): Build the parallel graphics pipeline providing the highly vectorized counterparts (loss scale math, back-propagation inverse gradient accumulations, bitmask array propagations) meant to interface natively with the Core infrastructure. 8. RTL Components (Haskell): Write memory arbiter, logic ranker cores, and structural semantic hardware modules demonstrating verifiable HDL topologies. 9. Orchestration Benchmarks (Python): Write the external operations script that provisions GPUs via cloud functions, compiles target binary artifacts, initiates datasets mapped for inference testing, coordinates peer messaging over multithreading protocols, and validates gradient curve loss. - Rewrite the technical description as one comprehensive, self-contained build prompt that describes, step by step from start to finish, exactly what to construct and how. - The wording must be explicit and unambiguous so any model can follow it without guessing. - Do not name the original reference software or architecture codebase in the generated components. - Return the complete code with every structural component addressed. - Write out every single character; do not abbreviate anything. - No explanations, summaries, or notes. - No placeholders, no "...rest of code", no "to be implemented". - Never use simplified, mock, placeholder, dummy, simulated, or fake content. - Require the full software with all files, in complete, unabridged, production-ready code.
