TensorStudio 2.1.0

TensorStudio 2.1.0

TensorStudio 2.1.0 extends the v2 CPU-first framework foundation with a
stronger backend boundary, safer interchange utilities, reproducible vision
dataset manifests, and refreshed local benchmark results.

Highlights:

• TensorFlow-style backend metadata for allocator, runtime, logical-device,
  kernel placement, transfer, and execution-plan diagnostics.
• Native storage telemetry for allocation checkouts, active bytes, cumulative
  bytes, and peak active usage.
• Safe descriptor-only custom-kernel manifest loading, validation, discovery,
  and registration.
• ONNX Runtime named-input inference through run_onnx_inference().
• Safe metadata inspection for Keras archives, TensorFlow SavedModel
  directories, HDF5/Keras weight files, and TensorFlow Lite flatbuffers.
• Deterministic image-folder manifests and ImageManifestDataset.
• README, docs, and benchmark reports updated for 2.1.0.

Validated locally:

• python -m ruff check .
• python -m mypy python\tensorstudio
• python -m pytest -q (237 passed)
• python -m mkdocs build --strict
• python benchmark_all.py
• python benchmarks\bench_matmul.py
• python -m build
• python -m twine check dist\tensorstudio-2.1.0*
• clean wheel install smoke test

Limitations remain explicit: published wheels are CPU-execution only; CUDA,
Metal, plugin execution, and full ONNX/TensorFlow/PyTorch runtime parity remain
future work.

TensorStudio 2.0.0

TensorStudio 2.0.0 is a tested v2 foundation release. It adds concrete roadmap work while keeping unimplemented accelerator and distributed systems honest.

Highlights:

• Dataset manifests, SHA-256 file checksums, manifest validation, and a small map-style dataset cache wrapper.
• Compact attention APIs: scaled_dot_product_attention, MultiHeadSelfAttention, and TransformerEncoderBlock.
• Experimental CSR sparse tensors with dense/COO conversion and CSR sparse-dense matmul.
• Quantization calibration helpers and quantization error reporting.
• CPU from_dlpack() import for DLPack-compatible providers through NumPy.
• ONNX Runtime provider discovery and compatibility diagnostics.
• Roadmap rewritten to show remaining work only.

Validation:

• python -m pip install -e .[dev,docs]
• python test_all.py --quiet
• python -m ruff check .
• python -m mypy python\tensorstudio
• python -m pytest -q
• python -m mkdocs build --strict
• python benchmark_all.py --check-thresholds
• python benchmarks\bench_matmul.py
• python -m build
• python -m twine check dist*
• python tools\verify_artifacts.py --wheel-dir dist --sdist-dir dist

Still not claimed as complete:

• No CUDA or Metal tensor execution kernels.
• No production multi-process distributed runtime.
• No machine-code JIT/compiler.
• No full TensorFlow/PyTorch compatibility.
• No broad performance superiority claim.

TensorStudio 1.16.0

TensorStudio 1.16.0 completes the Ecosystem And Advanced Features roadmap batch.

Highlights:

• Experimental COO sparse tensors with dense conversion, coalescing, transpose, and sparse-dense matmul.
• Public dataset readers for CSV, JSONL, text-line, and LIBSVM-style files.
• Tiny model-zoo factories, language-model helpers, quantization research utilities, and a custom kernel registry.
• Distributed research helpers for single-process collectives and deterministic data-parallel planning.
• Optional ONNX Runtime adapter via tensorstudio[onnxruntime], with TensorStudio importer fallback when runtime execution fails.
• Updated docs, benchmarks, examples, and local release validation.

Validation:

• python test_all.py --quiet
• python -m mkdocs build --strict
• python benchmark_all.py --check-thresholds
• python benchmarks\bench_matmul.py
• python -m build
• python -m twine check dist*
• python tools\verify_artifacts.py --wheel-dir dist --sdist-dir dist

Limitations remain explicit: CPU tensor execution only, no CUDA/Metal kernels, no production distributed runtime, and no native full ONNX runtime.

TensorStudio 1.15.0

Changelog

Unreleased

1.15.0 - 2026-07-07

• Completed the Graph, Compiler, And Runtime Systems roadmap batch as a
  constrained, inspectable graph runtime.
• Added TensorSpec, symbolic GraphTensor tracing, trace(), and
  compile_graph() for a supported subset of TensorStudio tensor programs.
• Added JSON graph serialization with save_graph() and load_graph().
• Added basic graph optimization passes: constant folding and scalar
  multiply-add fusion.
• Added ExecutableGraph runtime execution backed by TensorStudio eager tensor
  operations.
• Added runtime profiling hooks and static memory-planning metadata for graph
  execution.
• Documented graph limitations honestly: no arbitrary Python control-flow
  capture and no machine-code JIT in this release.

1.14.0 - 2026-07-07

• Completed the safe, testable Hardware Backends roadmap batch without claiming
  CUDA or Metal execution in CPU wheels.
• Added native CPU/CUDA/Metal Device descriptors, parsing, equality, and
  backend availability metadata.
• Added top-level Python hardware helpers: device, available_devices,
  backend_info, device_count, cuda_is_available, and
  metal_is_available.
• Added explicit tensor transfer APIs: Tensor.to_device(), Tensor.cpu(),
  and Tensor.to("cpu") while preserving dtype casts like Tensor.to("float64").
• Added device= keywords to public tensor factory helpers.
• Made native storage device-aware and reject unsupported accelerator
  allocation clearly.
• Added CMake gates for experimental CUDA and Metal metadata hooks.
• Added backend benchmark coverage and hardware backend documentation.
• Documented that real CUDA/Metal kernels remain deferred until accelerator
  builds and CI can validate them.

1.13.0 - 2026-07-07

• Completed the ordered Packaging, CI, And Release Quality roadmap section as
  one release batch.
• Added tools/verify_artifacts.py for clean wheel and sdist install smoke
  tests in isolated virtual environments.
• Expanded CI with cross-platform wheel/sdist artifact smoke jobs on Windows,
  Linux, and macOS.
• Hardened wheel, TestPyPI, and PyPI workflows with clean artifact verification
  before upload.
• Added benchmark report artifacts to release workflows.
• Added GitHub Pages docs publishing automation with strict MkDocs builds.
• Expanded macOS wheel coverage to include universal2 builds where supported by
  cibuildwheel.
• Added native ABI, wheel-tag, source-build, BLAS, and platform compatibility
  documentation.

1.12.0 - 2026-07-07

• Completed the ordered Serialization And Interchange roadmap section as one
  release batch.
• Expanded NPZ archives to version 2 metadata with TensorStudio version,
  tensor counts, tensor shapes/dtypes, requires_grad flags, and user metadata.
• Added optional SafeTensors save/load support for safe tensor-only weight maps.
• Added model metadata inspection for TensorStudio NPZ files, SafeTensors,
  supported ONNX files, and trusted TensorStudio checkpoints.
• Added versioned checkpoint metadata and compatibility checks for full trusted
  checkpoints.
• Expanded ONNX export for grouped convolution metadata and ConvTranspose2d.
• Added ONNX graph metadata inspection and supported-subset ONNX import/execution
  for static graphs using TensorStudio's eager tensor ops.
• Added model-card metadata JSON export.
• Expanded interchange tests and docs for SafeTensors, metadata inspection, and
  ONNX import.

1.11.0 - 2026-07-07

• Completed the ordered Computer Vision Depth roadmap section as one release
  batch.
• Added batch-aware resize/crop/normalize helpers plus color jitter, random
  resized crop, random rotation, affine transforms, cutout, mixup, and CutMix.
• Added detection utilities for box areas, IoU variants, NMS, box
  encode/decode, coordinate conversion, and anchor generation.
• Added segmentation helpers for mask IoU, one-hot conversion, masks-to-boxes,
  nearest mask resize, and deterministic mask crops.
• Added DetectionFolder and SegmentationFolder datasets plus
  tensorstudio.data factory aliases for detection and segmentation folders.
• Added ResidualBlock, DepthwiseSeparableBlock, CompactUNet, and
  make_unet() vision model helpers using the native-backed neural-network
  layer stack.
• Added prediction drawing, mask overlay, and feature-map grid visualization
  helpers.
• Expanded vision tests and documentation for transforms, detection,
  segmentation, model blocks, and visualization.

1.10.0 - 2026-07-07

• Completed the ordered Training And Project Workflows roadmap section as one
  release batch.
• Added ArrayDataset, tensor/array/image-folder dataset factories,
  deterministic train/validation splitting, and dataset metadata summaries.
• Added tensorstudio.metrics with regression, classification, and multilabel
  metrics for small supervised workflows.
• Added trainer validation loops, scheduler stepping, callback context support,
  learning-rate logging, CSV logging, checkpoint callbacks, and early stopping.
• Added JSON, TOML, and YAML project config loading plus deterministic seeding
  across TensorStudio, NumPy, and Python random.
• Expanded full checkpoints with scheduler and epoch state and added
  resume_checkpoint() for continuing training runs.
• Added generated regression, classification, and vision project templates.
• Expanded tests and docs for project workflows, metrics, dataset creation,
  callbacks, configs, templates, and checkpoint resume.

1.9.0 - 2026-07-07

• Completed the ordered Neural Network Building Blocks roadmap section as one
  release batch.
• Added native grouped conv2d, native conv_transpose2d, and native
  embedding lookup with autograd support.
• Added Python-level Conv1d, DepthwiseConv2d, ConvTranspose2d,
  BatchNorm1d, BatchNorm2d, LayerNorm, Embedding,
  adaptive/global pooling, and additional activation modules.
• Added tensorstudio.nn.init with Xavier, Kaiming, normal, uniform, zero, and
  one initializers.
• Added label-smoothing cross entropy, focal loss, KL divergence, negative log
  likelihood, and cosine embedding loss modules and functional helpers.
• Added module buffers, buffer-aware state_dict() support, and model summary
  utilities for parameters, shapes, and estimated tensor memory.
• Expanded tests and docs for the section-5 neural-network API surface.

1.8.0 - 2026-07-07

• Completed the ordered Autograd Coverage And Hardening roadmap section as one
  release batch.
• Added retain_graph support to Tensor.backward() and
  tensorstudio.autograd.backward().
• Added graph lifecycle hardening: normal backward frees non-leaf graph history,
  repeated backward through a freed graph raises a clear error, and retained
  graphs clear intermediate gradients between backward passes.
• Added Tensor is_leaf, clear_history(), and detach_() controls for
  explicit graph lifecycle management.
• Added guarded public in-place methods zero_(), fill_(), and add_().
  They reject gradient-tracked mutation while grad mode is enabled and work
  inside tensorstudio.no_grad().
• Expanded non-scalar backward and finite-difference gradient tests for stable
  probability ops, statistics, norms, and batched matrix multiplication.
• Expanded autograd documentation with a coverage matrix, lifecycle notes, and
  explicit higher-order-gradient limitations.

1.7.0 - 2026-07-07

• Completed the ordered Core Math Expansion roadmap section as one release
  batch.
• Added native C++ logsumexp, softmax, and log_softmax operations with
  max-shifted stable numerics and autograd support.
• Added native C++ batched matrix multiplication through bmm and 3D @
  dispatch, including reverse-mode gradients for both operands.
• Added native C++ var, variance, std, all, and any operations, plus
  Tensor methods and top-level Python exports where appropriate.
• Added Tensor-level norm() and expanded tensorstudio.math with
  logsumexp, softmax, log_softmax, boolean reductions, and a documented
  practical einsum subset.
• Added seeded native random distributions: uniform, normal, randint,
  and bernoulli, with Python *_like helpers where useful.
• Switched neural-network functional softmax/log-softmax and cross entropy to
  the native stable kernels.
• Added NumPy parity and autograd tests for the expanded math surface.

1.6.0 - 2026-07-07

• Completed the ordered CPU Performance Core roadmap section as one release
  batch.
• Added optional CBLAS/Accelerate-backed matmul for contiguous float32 and
  float64 matrices when the source build environment exposes a compatible
  BLAS library and header; portable C++ kernels remain the fallback path.
• Added a small native C++ CPU thread pool, configurable with
  tensorstudio.set_num_threads() and TENSORSTUDIO_NUM_THREADS, and used it
  for large contiguous elementwise ops, reductions, matrix multiplication,
  convolution, and pooling forward kernels.
• Added SIMD-friendly typed float32/float64 contiguous kernels for common
  elementwise arithmetic and activations, while preserving mixed-dtype fallback
  behavior.
• Added a bounded C++ storage reuse pool for tensor allocations, with
  TENSORSTUDIO_DISABLE_STORAGE_POOL=1 and
  TENSORSTUDIO_STORAGE_POOL_MAX_BLOCK_BYTES controls.
• Added tensorstudio.performance_info(), get_num_threads(), and
  set_num_threads() diagnostics/configuration helpers.
• Added benchmark threshold support via benchmark_all.py --check-thresholds
  and benchmarks/thresholds.json.
• Expanded performance, CPU backend, and platform docs for BLAS selection,
  threading, storage reuse, benchmark thresholds, and honest interpretation.

1.5.1 - 2026-07-07

• Completed the next ordered correctness-roadmap item with clearer native
  shape, dtype, and indexing error messages.
• Broadcasting errors now include the mismatched axis and dimensions.
• Reshape errors now include requested...

TensorStudio 1.14.0

TensorStudio 1.14.0 completes the safe, testable Hardware Backends roadmap batch. It adds native CPU/CUDA/Metal device descriptors, backend availability metadata, explicit transfer APIs, device-aware storage validation, tensor factory device keywords, backend benchmark coverage, CMake CUDA/Metal metadata hooks, and hardware docs. Published wheels remain honest CPU-only builds: CUDA and Metal execution kernels are deferred until accelerator builds and CI can validate them. Validated locally with ruff, mypy, pytest, test_all with artifact smoke, strict MkDocs, benchmark thresholds, official build, twine check, and clean wheel/sdist verification.

TensorStudio 1.13.0

TensorStudio 1.13.0 completes the Packaging, CI, And Release Quality roadmap batch. It adds clean wheel/sdist artifact verification, cross-platform CI artifact smoke jobs, hardened wheel/TestPyPI/PyPI workflows, benchmark report artifacts, GitHub Pages docs publishing, macOS universal2 wheel coverage where supported by cibuildwheel, and ABI/platform compatibility docs. Validated locally with ruff, mypy, pytest, test_all with artifact smoke, strict MkDocs, benchmark thresholds, official build, twine check, and clean wheel/sdist artifact verification.

TensorStudio 1.12.0

TensorStudio 1.12.0 completes the Serialization And Interchange roadmap batch. It adds richer NPZ metadata, optional SafeTensors save/load and inspection, versioned checkpoint metadata, ONNX metadata inspection, grouped Conv and ConvTranspose ONNX export coverage, supported-subset ONNX import/execution for static graphs, and model-card metadata export. Validated locally with ruff, mypy, pytest, test_all, strict MkDocs, benchmark thresholds, wheel/sdist build, twine check, and clean wheel/sdist smoke installs.

TensorStudio 1.11.0

TensorStudio 1.11.0 completes the Computer Vision Depth roadmap section. It adds batch-aware image transforms, color jitter, random resized crop, rotation, affine transforms, cutout, mixup, CutMix, detection utilities, segmentation mask helpers, detection and segmentation folder datasets, ResNet-style and MobileNet-style model blocks, CompactUNet, and prediction/mask/feature-map visualization helpers. Validated locally with test_all.py --quiet, mkdocs build --strict, benchmark_all.py --check-thresholds, bench_matmul.py, twine check, and a clean wheel smoke install.

TensorStudio 1.10.0

TensorStudio 1.10.0 completes the Training And Project Workflows roadmap section. It adds dataset factories and deterministic splits, metrics, trainer validation loops, callbacks, JSON/TOML/YAML config loading, checkpoint resume helpers, deterministic seeding, generated project templates, refreshed docs, and updated local benchmark reports. Validated locally with test_all.py --quiet, benchmark_all.py --check-thresholds, bench_matmul.py, twine check, and a clean wheel smoke install.

TensorStudio 1.9.0

TensorStudio 1.9.0 completes roadmap section 5: Neural Network Building Blocks. It adds native grouped conv2d, native conv_transpose2d, native embedding lookup, Conv1d, DepthwiseConv2d, ConvTranspose2d, BatchNorm1d/2d, LayerNorm, Embedding, adaptive/global pooling, GELU/ELU/SELU/SiLU/Mish activations, nn.init initializers, expanded losses, module buffers, buffer-aware state_dict support, and model summaries. Local validation passed: ruff, mypy, pytest, examples, mkdocs build, test_all.py, python -m build, twine check, benchmark_all --check-thresholds, bench_matmul, and clean wheel smoke install.