modelstudio 0.4.0__tar.gz → 0.5.0__tar.gz

{modelstudio-0.4.0/python/modelstudio.egg-info → modelstudio-0.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modelstudio
-Version: 0.4.0
+Version: 0.5.0
 Summary: An early-stage AI tensor framework with CPU tensors, autograd, and backend extension scaffolding.
 Author: ModelStudio Contributors
 License-Expression: MIT
@@ -31,9 +31,10 @@ Dynamic: license-file
 
 # ModelStudio
 
-ModelStudio is an early-stage AI tensor framework. Version `0.4.0` provides a
+ModelStudio is an early-stage AI tensor framework. Version `0.5.0` provides a
 CPU tensor/autograd MVP with neural-network modules, optimizers, serialization,
-basic data loading, and small LLM-oriented building blocks.
+data loading, graph tracing metadata, backend status inspection, and small
+LLM-oriented building blocks.
 
 It is not a PyTorch or TensorFlow replacement. CPU is the only working backend.
 CUDA, ROCm, and oneAPI remain explicit scaffolds until real kernels are built
@@ -59,28 +60,45 @@ python -m pip install -e ".[dev]"
 | --- | --- |
 | CPU tensors | Working MVP |
 | Autograd | Reverse-mode for core CPU ops |
-| Reductions | `sum`, `mean`, `max` with axis and keepdims; `max` is value-only |
+| Reductions | `sum`, `mean`, `max`, `all`, and `any`; `max` is value-only |
+| Comparisons | Elementwise comparisons, `equal`, `isclose`, and `allclose` |
 | Activations | ReLU, GELU, LeakyReLU, ELU, Softplus, exp, log, tanh, sigmoid, SiLU, softmax, log-softmax |
 | Losses | MSE and cross entropy with `none`, `mean`, and `sum` reductions |
+| Functional API | `modelstudio.nn.functional` wrappers for common NN operations |
 | Modules | Parameters, buffers, child traversal, state dicts, save/load |
 | Layers | Linear, Embedding, LayerNorm, RMSNorm, BatchNorm1d, Dropout, Conv1d, Conv2d, pooling, TransformerBlock |
 | Optimizers | SGD and AdamW with state serialization, parameter groups, and LR schedulers |
 | Data | Dataset, TensorDataset, random_split, DataLoader with deterministic seeded shuffle |
-| Randomness | `manual_seed`, RNG-backed `randn`, dropout, and init helpers |
+| Randomness | `manual_seed`, `ms.random`, RNG-backed creation, dropout, and init helpers |
+| Linalg | `ms.linalg.matmul`, `norm`, `vector_norm`, and `transpose` |
 | Interop | `asarray`, `from_numpy`, `to_numpy`, and `ms.numpy` |
 | Metrics | accuracy and top-k accuracy |
-| Compiler | Placeholder IR and passes |
+| Compiler | Metadata-only tracing plus placeholder IR and passes |
 
 ## Backend Status
 
-| Backend | Status |
-| --- | --- |
-| CPU | working MVP |
-| CUDA | scaffold only |
-| ROCm | scaffold only |
-| oneAPI | scaffold only |
+```python
+import modelstudio as ms
 
-Unsupported accelerator devices fail with `ModelStudioBackendUnavailable`.
+print(ms.backends.status())
+print(ms.backends.native_cpu_available())
+```
+
+Expected shape:
+
+```python
+{
+    "cpu": {"available": True, "native": False},
+    "cuda": {"available": False, "reason": "..."},
+    "rocm": {"available": False, "reason": "..."},
+    "oneapi": {"available": False, "reason": "..."},
+}
+```
+
+The production CPU path is the NumPy backend. `ms.backends.use_native_cpu(True)`
+raises `ModelStudioBackendUnavailable` unless a future optional native extension
+is actually installed. Unsupported accelerator devices fail with
+`ModelStudioBackendUnavailable`.
 
 ## Tensor Example
 
@@ -94,121 +112,68 @@ loss.backward()
 print(w.grad)
 ```
 
-## MLP Example
+## Functional API
 
 ```python
 import modelstudio as ms
 from modelstudio import nn
+from modelstudio.nn import functional as F
 
-
-class MLP(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.fc1 = nn.Linear(784, 256)
-        self.fc2 = nn.Linear(256, 10)
-
-    def forward(self, x):
-        return self.fc2(ms.gelu(self.fc1(x)))
-
-
-model = MLP()
-optimizer = ms.optim.AdamW(model.parameters(), lr=3e-4)
-x = ms.randn((16, 784))
-target = ms.randn((16, 10))
-loss = ms.mse_loss(model(x), target)
-optimizer.zero_grad()
-loss.backward()
-optimizer.step()
-```
-
-## State Dict and Save/Load
-
-```python
 model = nn.Linear(4, 2)
-ms.save(model.state_dict(), "model.ms")
-state = ms.load("model.ms")
-model.load_state_dict(state)
+x = ms.random.randn((8, 4))
+target = ms.random.randn((8, 2))
+loss = F.mse_loss(F.relu(F.linear(x, model.weight, model.bias)), target)
 ```
 
-## DataLoader
+## Tracing
 
 ```python
-from modelstudio import data
-
-dataset = data.TensorDataset(ms.randn((8, 4)), ms.arange(8))
-loader = data.DataLoader(dataset, batch_size=2, shuffle=False)
-for xb, yb in loader:
-    print(xb.shape, yb.shape)
-```
-
-## Embedding
-
-```python
-emb = nn.Embedding(num_embeddings=100, embedding_dim=32)
-tokens = ms.tensor([[1, 2, 3]], dtype=ms.int64)
-print(emb(tokens).shape)
-```
-
-## Cross Entropy
+import modelstudio as ms
+from modelstudio.nn import functional as F
 
-```python
-logits = ms.randn((4, 10), requires_grad=True)
-targets = ms.tensor([1, 2, 3, 4], dtype=ms.int64)
-loss = ms.cross_entropy(logits, targets)
-loss.backward()
+x = ms.random.randn((4, 3))
+w = ms.random.randn((3, 2))
+graph = ms.trace(lambda a, b: F.relu(a @ b), x, w)
+print(graph)
 ```
 
-## TransformerBlock
-
-```python
-block = nn.TransformerBlock(embed_dim=16, num_heads=4)
-x = ms.randn((2, 8, 16), requires_grad=True)
-y = block(x)
-print(y.shape)
-```
+Tracing captures operation names and tensor metadata. It does not optimize or
+execute graphs yet. `ms.compile(fn)` remains a documented no-op that returns the
+original callable.
 
-## 0.4.0 Training Utilities
+## Random And Linalg
 
 ```python
-ms.manual_seed(123)
-model = nn.Linear(4, 2)
-optimizer = ms.optim.AdamW(model.parameters(), lr=1e-3)
-state = {"model": model.state_dict(), "optimizer": optimizer.state_dict()}
-ms.save(state, "checkpoint.ms")
+ms.random.seed(123)
+x = ms.random.normal((4, 3), mean=0.0, std=1.0)
+w = ms.random.uniform((3, 2), low=-0.1, high=0.1)
+y = ms.linalg.matmul(x, w)
+print(ms.linalg.norm(y).item())
 ```
 
-New CPU-only helpers include `ms.concat`, `ms.stack`, `Tensor.flatten`,
-`Tensor.squeeze`, `Tensor.unsqueeze`, `nn.init`, `nn.Dropout`,
-`nn.BatchNorm1d`, `nn.Conv1d`, `nn.Conv2d`, `nn.AvgPool2d`, `nn.MaxPool2d`,
-and `nn.utils` gradient clipping.
-
-## NumPy Interop
+## Comparisons
 
 ```python
-x = ms.asarray([[1, 2, 3], [4, 5, 6]], dtype=ms.float32)
-arr = ms.to_numpy(x)
-y = ms.from_numpy(arr)
+x = ms.tensor([1.0, 2.0, 3.0])
+y = ms.tensor([1.0, 2.1, 3.0])
+print(ms.isclose(x, y, atol=0.05))
+print(ms.allclose(x, y, atol=0.05))
+print((x > 1.5).any().item())
 ```
 
-CPU uses NumPy internally. Normal examples prefer ModelStudio APIs; `ms.numpy`
-is exposed for advanced users who explicitly want NumPy access.
+Comparison and logical outputs are bool tensors and do not track gradients.
 
-## Schedulers and Metrics
+## Checkpointing
 
 ```python
+model = nn.Linear(4, 2)
 optimizer = ms.optim.AdamW(model.parameters(), lr=1e-3)
-scheduler = ms.optim.lr_scheduler.StepLR(optimizer, step_size=1, gamma=0.5)
-scheduler.step()
-
-acc = ms.metrics.accuracy(logits, targets)
+ms.save_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, extra={"epoch": 1})
+checkpoint = ms.load_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, map_location="cpu")
 ```
 
-## Checkpointing
-
-```python
-ms.save_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, scheduler=scheduler, extra={"epoch": 1})
-checkpoint = ms.load_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, scheduler=scheduler)
-```
+Checkpoint loading validates structure and model state. CPU is the only accepted
+`map_location` in the current release.
 
 ## Commands
 
@@ -226,6 +191,10 @@ python examples/numpy_interop.py
 python examples/scheduler_training.py
 python examples/checkpoint_resume.py
 python examples/metrics_demo.py
+python examples/backend_status.py
+python examples/tracing_demo.py
+python examples/functional_training.py
+python examples/random_linalg_demo.py
 python benchmarks/bench_matmul.py
 python benchmarks/bench_mlp.py
 python benchmarks/bench_attention.py
@@ -234,17 +203,24 @@ python benchmarks/bench_conv.py
 python benchmarks/bench_dropout.py
 python benchmarks/bench_creation.py
 python benchmarks/bench_manipulation.py
+python benchmarks/bench_elementwise.py
+python benchmarks/bench_trace.py
 ```
 
 ## Documentation
 
+- [Backend status](docs/backend-status.md)
+- [Tracing](docs/tracing.md)
+- [Functional API](docs/functional-api.md)
+- [Random namespace](docs/random.md)
+- [Linalg namespace](docs/linalg.md)
+- [Comparison ops](docs/comparison-ops.md)
 - [Tensor API](docs/tensor-api.md)
 - [Neural network API](docs/nn.md)
 - [Data utilities](docs/data.md)
 - [Training](docs/training.md)
 - [Modules](docs/modules.md)
 - [Serialization](docs/serialization.md)
-- [Randomness](docs/randomness.md)
 - [Native backend roadmap](docs/native-backend-roadmap.md)
 - [NumPy interop](docs/numpy-interop.md)
 - [Tensor creation](docs/tensor-creation.md)
@@ -260,6 +236,6 @@ python benchmarks/bench_manipulation.py
 ## Roadmap
 
 - Expand tensor and autograd coverage.
-- Wire native CPU kernels into Python bindings.
+- Wire optional native CPU kernels only after a safe Python extension exists.
 - Add tested CUDA, ROCm, and oneAPI packages when hardware-backed CI exists.
-- Improve compiler graph capture and lowering.
+- Improve compiler graph capture, analysis passes, and lowering.

{modelstudio-0.4.0 → modelstudio-0.5.0}/README.md

@@ -1,8 +1,9 @@
 # ModelStudio
 
-ModelStudio is an early-stage AI tensor framework. Version `0.4.0` provides a
+ModelStudio is an early-stage AI tensor framework. Version `0.5.0` provides a
 CPU tensor/autograd MVP with neural-network modules, optimizers, serialization,
-basic data loading, and small LLM-oriented building blocks.
+data loading, graph tracing metadata, backend status inspection, and small
+LLM-oriented building blocks.
 
 It is not a PyTorch or TensorFlow replacement. CPU is the only working backend.
 CUDA, ROCm, and oneAPI remain explicit scaffolds until real kernels are built
@@ -28,28 +29,45 @@ python -m pip install -e ".[dev]"
 | --- | --- |
 | CPU tensors | Working MVP |
 | Autograd | Reverse-mode for core CPU ops |
-| Reductions | `sum`, `mean`, `max` with axis and keepdims; `max` is value-only |
+| Reductions | `sum`, `mean`, `max`, `all`, and `any`; `max` is value-only |
+| Comparisons | Elementwise comparisons, `equal`, `isclose`, and `allclose` |
 | Activations | ReLU, GELU, LeakyReLU, ELU, Softplus, exp, log, tanh, sigmoid, SiLU, softmax, log-softmax |
 | Losses | MSE and cross entropy with `none`, `mean`, and `sum` reductions |
+| Functional API | `modelstudio.nn.functional` wrappers for common NN operations |
 | Modules | Parameters, buffers, child traversal, state dicts, save/load |
 | Layers | Linear, Embedding, LayerNorm, RMSNorm, BatchNorm1d, Dropout, Conv1d, Conv2d, pooling, TransformerBlock |
 | Optimizers | SGD and AdamW with state serialization, parameter groups, and LR schedulers |
 | Data | Dataset, TensorDataset, random_split, DataLoader with deterministic seeded shuffle |
-| Randomness | `manual_seed`, RNG-backed `randn`, dropout, and init helpers |
+| Randomness | `manual_seed`, `ms.random`, RNG-backed creation, dropout, and init helpers |
+| Linalg | `ms.linalg.matmul`, `norm`, `vector_norm`, and `transpose` |
 | Interop | `asarray`, `from_numpy`, `to_numpy`, and `ms.numpy` |
 | Metrics | accuracy and top-k accuracy |
-| Compiler | Placeholder IR and passes |
+| Compiler | Metadata-only tracing plus placeholder IR and passes |
 
 ## Backend Status
 
-| Backend | Status |
-| --- | --- |
-| CPU | working MVP |
-| CUDA | scaffold only |
-| ROCm | scaffold only |
-| oneAPI | scaffold only |
+```python
+import modelstudio as ms
 
-Unsupported accelerator devices fail with `ModelStudioBackendUnavailable`.
+print(ms.backends.status())
+print(ms.backends.native_cpu_available())
+```
+
+Expected shape:
+
+```python
+{
+    "cpu": {"available": True, "native": False},
+    "cuda": {"available": False, "reason": "..."},
+    "rocm": {"available": False, "reason": "..."},
+    "oneapi": {"available": False, "reason": "..."},
+}
+```
+
+The production CPU path is the NumPy backend. `ms.backends.use_native_cpu(True)`
+raises `ModelStudioBackendUnavailable` unless a future optional native extension
+is actually installed. Unsupported accelerator devices fail with
+`ModelStudioBackendUnavailable`.
 
 ## Tensor Example
 
@@ -63,121 +81,68 @@ loss.backward()
 print(w.grad)
 ```
 
-## MLP Example
+## Functional API
 
 ```python
 import modelstudio as ms
 from modelstudio import nn
+from modelstudio.nn import functional as F
 
-
-class MLP(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.fc1 = nn.Linear(784, 256)
-        self.fc2 = nn.Linear(256, 10)
-
-    def forward(self, x):
-        return self.fc2(ms.gelu(self.fc1(x)))
-
-
-model = MLP()
-optimizer = ms.optim.AdamW(model.parameters(), lr=3e-4)
-x = ms.randn((16, 784))
-target = ms.randn((16, 10))
-loss = ms.mse_loss(model(x), target)
-optimizer.zero_grad()
-loss.backward()
-optimizer.step()
-```
-
-## State Dict and Save/Load
-
-```python
 model = nn.Linear(4, 2)
-ms.save(model.state_dict(), "model.ms")
-state = ms.load("model.ms")
-model.load_state_dict(state)
+x = ms.random.randn((8, 4))
+target = ms.random.randn((8, 2))
+loss = F.mse_loss(F.relu(F.linear(x, model.weight, model.bias)), target)
 ```
 
-## DataLoader
+## Tracing
 
 ```python
-from modelstudio import data
-
-dataset = data.TensorDataset(ms.randn((8, 4)), ms.arange(8))
-loader = data.DataLoader(dataset, batch_size=2, shuffle=False)
-for xb, yb in loader:
-    print(xb.shape, yb.shape)
-```
-
-## Embedding
-
-```python
-emb = nn.Embedding(num_embeddings=100, embedding_dim=32)
-tokens = ms.tensor([[1, 2, 3]], dtype=ms.int64)
-print(emb(tokens).shape)
-```
-
-## Cross Entropy
+import modelstudio as ms
+from modelstudio.nn import functional as F
 
-```python
-logits = ms.randn((4, 10), requires_grad=True)
-targets = ms.tensor([1, 2, 3, 4], dtype=ms.int64)
-loss = ms.cross_entropy(logits, targets)
-loss.backward()
+x = ms.random.randn((4, 3))
+w = ms.random.randn((3, 2))
+graph = ms.trace(lambda a, b: F.relu(a @ b), x, w)
+print(graph)
 ```
 
-## TransformerBlock
-
-```python
-block = nn.TransformerBlock(embed_dim=16, num_heads=4)
-x = ms.randn((2, 8, 16), requires_grad=True)
-y = block(x)
-print(y.shape)
-```
+Tracing captures operation names and tensor metadata. It does not optimize or
+execute graphs yet. `ms.compile(fn)` remains a documented no-op that returns the
+original callable.
 
-## 0.4.0 Training Utilities
+## Random And Linalg
 
 ```python
-ms.manual_seed(123)
-model = nn.Linear(4, 2)
-optimizer = ms.optim.AdamW(model.parameters(), lr=1e-3)
-state = {"model": model.state_dict(), "optimizer": optimizer.state_dict()}
-ms.save(state, "checkpoint.ms")
+ms.random.seed(123)
+x = ms.random.normal((4, 3), mean=0.0, std=1.0)
+w = ms.random.uniform((3, 2), low=-0.1, high=0.1)
+y = ms.linalg.matmul(x, w)
+print(ms.linalg.norm(y).item())
 ```
 
-New CPU-only helpers include `ms.concat`, `ms.stack`, `Tensor.flatten`,
-`Tensor.squeeze`, `Tensor.unsqueeze`, `nn.init`, `nn.Dropout`,
-`nn.BatchNorm1d`, `nn.Conv1d`, `nn.Conv2d`, `nn.AvgPool2d`, `nn.MaxPool2d`,
-and `nn.utils` gradient clipping.
-
-## NumPy Interop
+## Comparisons
 
 ```python
-x = ms.asarray([[1, 2, 3], [4, 5, 6]], dtype=ms.float32)
-arr = ms.to_numpy(x)
-y = ms.from_numpy(arr)
+x = ms.tensor([1.0, 2.0, 3.0])
+y = ms.tensor([1.0, 2.1, 3.0])
+print(ms.isclose(x, y, atol=0.05))
+print(ms.allclose(x, y, atol=0.05))
+print((x > 1.5).any().item())
 ```
 
-CPU uses NumPy internally. Normal examples prefer ModelStudio APIs; `ms.numpy`
-is exposed for advanced users who explicitly want NumPy access.
+Comparison and logical outputs are bool tensors and do not track gradients.
 
-## Schedulers and Metrics
+## Checkpointing
 
 ```python
+model = nn.Linear(4, 2)
 optimizer = ms.optim.AdamW(model.parameters(), lr=1e-3)
-scheduler = ms.optim.lr_scheduler.StepLR(optimizer, step_size=1, gamma=0.5)
-scheduler.step()
-
-acc = ms.metrics.accuracy(logits, targets)
+ms.save_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, extra={"epoch": 1})
+checkpoint = ms.load_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, map_location="cpu")
 ```
 
-## Checkpointing
-
-```python
-ms.save_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, scheduler=scheduler, extra={"epoch": 1})
-checkpoint = ms.load_checkpoint("checkpoint.ms", model=model, optimizer=optimizer, scheduler=scheduler)
-```
+Checkpoint loading validates structure and model state. CPU is the only accepted
+`map_location` in the current release.
 
 ## Commands
 
@@ -195,6 +160,10 @@ python examples/numpy_interop.py
 python examples/scheduler_training.py
 python examples/checkpoint_resume.py
 python examples/metrics_demo.py
+python examples/backend_status.py
+python examples/tracing_demo.py
+python examples/functional_training.py
+python examples/random_linalg_demo.py
 python benchmarks/bench_matmul.py
 python benchmarks/bench_mlp.py
 python benchmarks/bench_attention.py
@@ -203,17 +172,24 @@ python benchmarks/bench_conv.py
 python benchmarks/bench_dropout.py
 python benchmarks/bench_creation.py
 python benchmarks/bench_manipulation.py
+python benchmarks/bench_elementwise.py
+python benchmarks/bench_trace.py
 ```
 
 ## Documentation
 
+- [Backend status](docs/backend-status.md)
+- [Tracing](docs/tracing.md)
+- [Functional API](docs/functional-api.md)
+- [Random namespace](docs/random.md)
+- [Linalg namespace](docs/linalg.md)
+- [Comparison ops](docs/comparison-ops.md)
 - [Tensor API](docs/tensor-api.md)
 - [Neural network API](docs/nn.md)
 - [Data utilities](docs/data.md)
 - [Training](docs/training.md)
 - [Modules](docs/modules.md)
 - [Serialization](docs/serialization.md)
-- [Randomness](docs/randomness.md)
 - [Native backend roadmap](docs/native-backend-roadmap.md)
 - [NumPy interop](docs/numpy-interop.md)
 - [Tensor creation](docs/tensor-creation.md)
@@ -229,6 +205,6 @@ python benchmarks/bench_manipulation.py
 ## Roadmap
 
 - Expand tensor and autograd coverage.
-- Wire native CPU kernels into Python bindings.
+- Wire optional native CPU kernels only after a safe Python extension exists.
 - Add tested CUDA, ROCm, and oneAPI packages when hardware-backed CI exists.
-- Improve compiler graph capture and lowering.
+- Improve compiler graph capture, analysis passes, and lowering.

modelstudio-0.5.0/benchmarks/bench_elementwise.py

@@ -0,0 +1,44 @@
+from __future__ import annotations
+
+import platform
+import time
+
+import modelstudio as ms
+
+
+def _time_ms(fn, warmup: int, iterations: int) -> float:
+    for _ in range(warmup):
+        fn()
+    start = time.perf_counter()
+    for _ in range(iterations):
+        fn()
+    return (time.perf_counter() - start) * 1000.0 / iterations
+
+
+def main() -> None:
+    shape = (1024, 1024)
+    warmup = 5
+    iterations = 50
+    ms.random.seed(123)
+    x = ms.random.randn(shape)
+    y = ms.random.randn(shape)
+
+    add_ms = _time_ms(lambda: x + y, warmup, iterations)
+    relu_ms = _time_ms(lambda: ms.relu(x), warmup, iterations)
+    cmp_ms = _time_ms(lambda: x > y, warmup, iterations)
+
+    print(f"Python:      {platform.python_version()}")
+    print(f"NumPy:       {ms.numpy.__version__}")
+    print(f"ModelStudio: {ms.__version__}")
+    print(f"Shape:       {shape}")
+    print(f"Warmup:      {warmup}")
+    print(f"Iterations:  {iterations}")
+    print(f"Backend:     {ms.backends.status()}")
+    print(f"add avg:     {add_ms:.3f} ms")
+    print(f"relu avg:    {relu_ms:.3f} ms")
+    print(f"compare avg: {cmp_ms:.3f} ms")
+
+
+if __name__ == "__main__":
+    main()
+

modelstudio-0.5.0/benchmarks/bench_trace.py

@@ -0,0 +1,47 @@
+from __future__ import annotations
+
+import platform
+import time
+
+import modelstudio as ms
+from modelstudio.nn import functional as F
+
+
+def _time_ms(fn, warmup: int, iterations: int) -> float:
+    for _ in range(warmup):
+        fn()
+    start = time.perf_counter()
+    for _ in range(iterations):
+        fn()
+    return (time.perf_counter() - start) * 1000.0 / iterations
+
+
+def main() -> None:
+    shape = (64, 128)
+    weight_shape = (128, 32)
+    warmup = 5
+    iterations = 100
+    x = ms.random.randn(shape)
+    w = ms.random.randn(weight_shape)
+
+    def forward(a: ms.Tensor, b: ms.Tensor) -> ms.Tensor:
+        return F.relu(a @ b)
+
+    trace_ms = _time_ms(lambda: ms.trace(forward, x, w), warmup, iterations)
+    graph = ms.trace(forward, x, w)
+
+    print(f"Python:      {platform.python_version()}")
+    print(f"NumPy:       {ms.numpy.__version__}")
+    print(f"ModelStudio: {ms.__version__}")
+    print(f"Input:       {shape}")
+    print(f"Weight:      {weight_shape}")
+    print(f"Warmup:      {warmup}")
+    print(f"Iterations:  {iterations}")
+    print(f"Backend:     {ms.backends.status()}")
+    print(f"Trace avg:   {trace_ms:.3f} ms")
+    print(f"Nodes:       {[node.op for node in graph.nodes]}")
+
+
+if __name__ == "__main__":
+    main()
+

modelstudio-0.5.0/docs/backend-status.md

@@ -0,0 +1,21 @@
+# Backend Status
+
+ModelStudio 0.5.0 keeps CPU as the only available runtime backend.
+
+```python
+import modelstudio as ms
+
+status = ms.backends.status()
+print(status["cpu"])
+```
+
+`ms.backends.status()` returns a mapping for `cpu`, `cuda`, `rocm`, and
+`oneapi`. CPU is always available through the NumPy backend. Accelerator
+backends remain explicit unavailable scaffolds and include a human-readable
+reason.
+
+`ms.backends.native_cpu_available()` checks for the optional future native CPU
+extension. `ms.backends.use_native_cpu(True)` raises
+`ModelStudioBackendUnavailable` unless that extension is installed. The NumPy
+CPU backend remains the production path.
+

modelstudio-0.5.0/docs/comparison-ops.md

@@ -0,0 +1,23 @@
+# Comparison Ops
+
+ModelStudio 0.5.0 adds elementwise tensor comparisons:
+
+```python
+x = ms.tensor([1.0, 2.0, 3.0])
+y = ms.tensor([1.0, 2.1, 3.0])
+
+print(x == y)
+print(x > 1.5)
+print(ms.isclose(x, y, atol=0.05))
+```
+
+Available helpers:
+
+- `ms.equal(x, y)` for exact shape and value equality
+- `ms.allclose(x, y, rtol=1e-5, atol=1e-8)`
+- `ms.isclose(x, y, rtol=1e-5, atol=1e-8)`
+- `x.all(axis=None, keepdims=False)` and `x.any(...)`
+- `ms.all(x, axis=None, keepdims=False)` and `ms.any(...)`
+
+Comparison and logical outputs are bool tensors and do not track gradients.
+