4 Commits

Include renames

Author	SHA1	Message	Date
西风逍遥游	92d62b2f66	feat(tensor): add quantized ONNX inference with reusable Vulkan sessions	2026-09-10 20:01:21 +08:00
西风逍遥游	112ace39c2	feat: integrate EveScript across engine tooling	2026-08-26 21:56:34 +08:00
西风逍遥游	f7962f43fe	feat(examples): developer-experience and AI-native showcase examples ... - devlab: flagship --debug example (console/snapshot/AI panel/pause/hot reload/error slice), make devlab target with RUN_ARGS support - ai-game: agent-drivable live game + agent_demo.py (pause/read/mutate/ verify/snapshot/restore/screenshot loop) - ai-editor: headless eve mcp editor host demo (JSON View + ViewModel MVVM) - examples: fill all missing READMEs + config.nut, add examples/ index - load.nut: add key_just_pressed/clampf/persist helpers; dedupe 19 examples - rpg/cardgame: move JSON definitions to data files - fix eve mcp crash: ModuleManager::requireAll() instantiates modules in the headless host (IEditorHost etc. were null) - graphics: actionable screenshot error when readback was just enabled - CI: smoke all examples (devlab with --debug) + DevLab hot-reload test Fixes surfaced by the new CI smoke (examples had never been smoke-tested): - scene_director kit: camera slot doubled as function + instance and was nulled by reset(); move the live Camera3D into cameraObj - readTextFile used non-existent readn()/len(); use file.read() (rpg/cardgame/hex-levels) - daynight used non-existent keyboard.wasPressed() -> key_just_pressed - outline passed ints to float-only drawSolidRect; linear-structures slider passed an int value - scene-ecs beginNode() requires (id, name); ai-stage guarded compilestring - smoke_examples.sh: timeout -k 3 so SIGTERM-ignoring hosts cannot hang CI	2026-08-21 19:18:52 +08:00
西风逍遥游	cff9bb4d0c	tensor: AITemplate-style compile pipeline, gpgpu Sequence, mini-LLM demo and weight-only quantization ... Refactor the tensor module around AITemplate-style static compilation: - Optimizer: DCE, constant folding, elementwise chain fusion, matmul/conv bias+activation epilogue fusion, liveness-based static memory planning, topological group execution order. - KernelGen: shape-baked GLSL per fused group; two-pass softmax/layernorm; shared-memory SDPA; naive + 16x16 tiled matmul templates. - GpuBackend: arena buffer allocation per memory plan, matmul autotuning, single command-buffer submission via the new gpgpu::Sequence (uploads, dispatches and downloads recorded once per run, Kompute-style). - shaderc in-process GLSL->SPIR-V on Windows (no glslc.exe dependency), with glslc spawn fallback. - CPU reference kernels shared by eager and the compiled interpreter; no-Vulkan fallback keeps compiled graphs runnable headless. Neural-network coverage: softmax/logsoftmax, layernorm/rmsnorm (with weight/bias), gelu/silu, conv1d/conv2d, pool, embedding, concat/slice, axis reduces, argmax, cast, fused SDPA (+masked), resize2d, batched matmul, seeded random. Weight-only quantization (activations stay fp32): - fp16 (IEEE half), fp8 (e4m3), fp4 (e2m1), int8, int4 with per-group block scales (MXFP-style: scale = maxAbs / format max magnitude). - Quantized weights capture into graph Const nodes; CPU interpreter dequantizes at materialization; GPU matmul/embedding kernels dequantize inline from packed uint buffers (naive matmul variant). - TF::quantizeWeight API + script binding. Mini-LLM validation (TinyStories-gpt2-3M, 8 layers d=64): - test/tensor_llm.cpp builds the full GPT-2 graph from engine ops and matches numpy reference: probs within 3e-7, greedy generation 24/24 identical (~100 ms/token CPU), compiled path exact. - tensor.llm.quantizedDtypes: fp16/int8/fp8/int4 all keep 16/16 greedy top-1 (fp4 16/16 with proper block scaling), weights 27.6 MB -> 3.3-13 MB. - scripts/export_tiny_gpt2.py converts a PyTorch .bin checkpoint (no torch dependency) plus GPT-2 tokenizer into engine assets + numpy reference. Tests: gpgpu 8/8, tensor 31/31 (original suite) + tensor.llm.* + tensor.quant.* Assets in test/assets/mini_llm are gitignored (regenerate via the script).	2026-08-18 17:42:38 +08:00