diff --git a/.github/workflows/frontend.yml b/.github/workflows/frontend.yml index 3da8e6242..ae0a1c276 100644 --- a/.github/workflows/frontend.yml +++ b/.github/workflows/frontend.yml @@ -21,7 +21,7 @@ jobs: steps: - uses: actions/checkout@v3 with: - path: 'triton_viz' + path: 'tilelens' - name: Set up Node uses: actions/setup-node@v4 @@ -30,6 +30,6 @@ jobs: - name: Run frontend tests run: | - cd triton_viz + cd tilelens npm ci npm run test:frontend diff --git a/.github/workflows/python-app.yml b/.github/workflows/python-app.yml index 88d8482cd..1919ccd17 100644 --- a/.github/workflows/python-app.yml +++ b/.github/workflows/python-app.yml @@ -25,7 +25,7 @@ jobs: steps: - uses: actions/checkout@v3 with: - path: 'triton_viz' + path: 'tilelens' - name: Set up Python 3.12 uses: actions/setup-python@v3 @@ -39,46 +39,46 @@ jobs: - name: Lint with pre-commit run: | - cd triton_viz + cd tilelens pip install pre-commit pre-commit run --all-files - name: Install Dependencies run: | - cd triton_viz + cd tilelens uv sync --extra test - name: Upgrade Triton to latest from main run: | - cd triton_viz + cd tilelens uv pip install --pre -U triton uv pip install "git+https://github.com/triton-lang/triton.git#subdirectory=python/triton_kernels" - name: Run frontend tests run: | - cd triton_viz + cd tilelens npm install npm run test:frontend - name: Test with pytest run: | - cd triton_viz + cd tilelens uv run pytest tests -n auto --ignore=tests/end_to_end/test_triton_kernels.py - - name: Install Triton-Viz with NKI extras + - name: Install TileLens with NKI extras run: | - cd triton_viz + cd tilelens uv sync --extra test --extra nki - name: Upgrade Triton to latest from main (NKI) run: | - cd triton_viz + cd tilelens uv pip install --pre -U triton uv pip install "git+https://github.com/triton-lang/triton.git#subdirectory=python/triton_kernels" - name: Run full (Triton + NKI) pytest suite run: | - cd triton_viz + cd tilelens uv run pytest tests -m "" -n auto --ignore=tests/end_to_end/test_triton_kernels.py test-pip-install: @@ -93,7 +93,22 @@ jobs: python-version: '3.12' - name: Install with pip - run: pip install -e ".[test]" + run: pip install ".[test]" - name: Verify installation - run: python -c "import triton_viz; print(triton_viz.__version__)" + working-directory: ${{ runner.temp }} + run: | + python - <<'PY' + import tilelens + import triton_viz + from tilelens.clients import Sanitizer + from triton_viz.clients import Sanitizer as LegacySanitizer + from tilelens.core.trace import launches + from triton_viz.core.trace import launches as legacy_launches + + assert triton_viz.trace is tilelens.trace + assert triton_viz.config is tilelens.config + assert LegacySanitizer is Sanitizer + assert legacy_launches is launches + print(tilelens.__version__) + PY diff --git a/.gitignore b/.gitignore index 3688aff7c..7400c3382 100644 --- a/.gitignore +++ b/.gitignore @@ -169,6 +169,6 @@ uv.lock benchmarks/*.json # Auto-generated version file -triton_viz/version.py +tilelens/version.py .subagents/ subagent*.txt diff --git a/AGENTS.md b/AGENTS.md index c6af6c4ab..dc61bac8f 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -2,7 +2,7 @@ ## Project Structure & Module Organization -Triton-Viz is a Python package with a TypeScript web UI. Core code lives in `triton_viz/`: `core/` handles tracing and client lifecycle, `core/frontend/` owns DSL operation adapters, runtime patching, and frontend-local AST rewrites, `core/simulation/` contains NKI simulation runtimes, `clients/` contains visualizer/profiler/sanitizer logic, and `visualizer/` covers visualizer APIs. Web UI source is in `frontend/`; built assets are copied into `triton_viz/static/` and `triton_viz/templates/`. Tests live in `tests/unit/`, `tests/end_to_end/`, `tests/frontend/`, and `tests/nki/`. Examples are in `examples/`; site docs are in `docs/`. +TileLens is a Python package with a TypeScript web UI. Core code lives in `tilelens/`: `core/` handles tracing and client lifecycle, `core/frontend/` owns DSL operation adapters, runtime patching, and frontend-local AST rewrites, `core/simulation/` contains NKI simulation runtimes, `clients/` contains visualizer/profiler/sanitizer logic, and `visualizer/` covers visualizer APIs. Web UI source is in `frontend/`; built assets are copied into `tilelens/static/` and `tilelens/templates/`. Tests live in `tests/unit/`, `tests/end_to_end/`, `tests/frontend/`, and `tests/nki/`. Examples are in `examples/`; site docs are in `docs/`. ## Build, Test, and Development Commands @@ -28,4 +28,4 @@ Recent commits use concise imperative summaries, often with prefixes such as `[F ## Security & Configuration Tips -Do not commit local traces, credentials, or generated debug artifacts. Runtime behavior is controlled by environment variables such as `TRITON_VIZ_VERBOSE`, `TRITON_VIZ_PORT`, `ENABLE_SANITIZER`, `ENABLE_PROFILER`, and `SANITIZER_ENABLE_FAKE_TENSOR`; document any new variables in `README.md`. +Do not commit local traces, credentials, or generated debug artifacts. Runtime behavior is controlled by environment variables such as `TILELENS_VERBOSE`, `TILELENS_PORT`, `ENABLE_SANITIZER`, `ENABLE_PROFILER`, and `SANITIZER_ENABLE_FAKE_TENSOR`; document any new variables in `README.md`. diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 7f1794039..74a6aad5c 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -1,11 +1,11 @@ -# Contributing to Triton-Viz +# Contributing to TileLens ## Quick Start ```bash # Clone repo -git clone https://github.com/Deep-Learning-Profiling-Tools/triton-viz.git -cd triton-viz +git clone https://github.com/Deep-Learning-Profiling-Tools/tilelens.git +cd tilelens # Setup dev environment and install dependencies uv sync --extra test # append "--extra nki" if developing NKI functionality @@ -23,9 +23,9 @@ npm run test:frontend # test the web UI (if working on the visualizer ``` frontend/ # TypeScript web UI source code -examples/ # Entry points for new users to try out triton-viz functionality -docs/ # triton-viz website -triton_viz/ +examples/ # Entry points for new users to try out tilelens functionality +docs/ # tilelens website +tilelens/ core/frontend/ # DSL operation adapters and runtime patching core/simulation/ # NKI simulation runtimes clients/ # Visualizer, profiler, and sanitizer clients diff --git a/README.md b/README.md index 1cbb7ed0d..0efc0527e 100644 --- a/README.md +++ b/README.md @@ -1,5 +1,5 @@ -# Triton-Viz: A Visualization Toolkit for programming with Triton +# TileLens: A Visualization Toolkit for programming with Triton
@@ -7,9 +7,9 @@

-Welcome to Triton-Viz, a visualization and profiling toolkit designed for deep learning applications. Built with the intention of making kernel programming in tile-based DSLs like Triton more intuitive. +Welcome to TileLens, a visualization and profiling toolkit designed for deep learning applications. Built with the intention of making kernel programming in tile-based DSLs like Triton more intuitive. -Visit our [site](https://deep-learning-profiling-tools.github.io/triton-viz/) to see our tool in action! +Visit our [site](https://deep-learning-profiling-tools.github.io/tilelens/) to see our tool in action!
@@ -20,7 +20,7 @@ Visit our [site](https://deep-learning-profiling-tools.github.io/triton-viz/) to Getting Started
  • @@ -34,7 +34,7 @@ Visit our [site](https://deep-learning-profiling-tools.github.io/triton-viz/) to ## About -Triton-Viz helps developers inspect Triton kernels with visualization, profiling, and memory-safety analysis tools. It can run many examples through Triton's interpreter, so GPU access is not required for basic debugging workflows. +TileLens helps developers inspect Triton kernels with visualization, profiling, and memory-safety analysis tools. It can run many examples through Triton's interpreter, so GPU access is not required for basic debugging workflows. ## Getting Started @@ -43,27 +43,55 @@ Triton-Viz helps developers inspect Triton kernels with visualization, profiling - Python >= 3.10 -### Installation of Triton-Viz +### Installation of TileLens -> **Windows Note:** Triton-viz depends on Triton, which can only be installed on Windows Subsystem for Linux (WSL). Once installed, follow below instructions in WSL. +> **Windows Note:** TileLens depends on Triton, which can only be installed on Windows Subsystem for Linux (WSL). Once installed, follow below instructions in WSL. -Most users can install directly from PyPI: +Install TileLens from PyPI: ```sh -pip install triton-viz +pip install tilelens ``` If you want to run examples from this repo, contribute, or build the web UI, install from source instead: ```sh -git clone https://github.com/Deep-Learning-Profiling-Tools/triton-viz.git -cd triton-viz +git clone https://github.com/Deep-Learning-Profiling-Tools/tilelens.git +cd tilelens uv sync # or "uv sync --extra test" if you're running tests ``` +### Transitioning from Triton-viz to TileLens + +The GitHub repo and PyPI package are now named `tilelens`. Use `import tilelens` +in new code. Old `triton_viz` imports, including submodule imports, still work. + +If you already have `triton-viz` installed, uninstall it before installing +TileLens. The two packages share files, so keeping both installed can break +imports and CLI commands: + +```sh +pip uninstall -y triton-viz +pip install tilelens +``` + +For source installs, use `pip install .` after uninstalling the old package. +If you already installed both, uninstall `triton-viz` first, then run +`pip install --force-reinstall --no-deps tilelens` (or +`pip install --force-reinstall --no-deps .` from this repo). This restores the +shared files without reinstalling or upgrading dependencies such as Triton. +Restart Python or your notebook kernel after upgrading. + +Existing `.tvz` traces can still be loaded with `tilelens.load(...)`. +Traces saved by TileLens cannot be loaded by older Triton-Viz versions. + +CLI commands are `tile-sanitizer`, `tile-profiler`, `tile-race`, and `tile-visualizer`. +The old `triton-sanitizer`, `triton-profiler`, `triton-race-detector`, and +`triton-visualizer` commands still work. + ### Web UI Build -The PyPI package ships with prebuilt web UI assets in `triton_viz/static`, so +The PyPI package ships with prebuilt web UI assets in `tilelens/static`, so you do not need npm to run the visualizer. If you want to modify the web UI, rebuild the TS sources: @@ -77,7 +105,7 @@ npm run build:frontend For PyPI installs, install with the `nki` extra and AWS Neuron repository: ```sh -pip install triton-viz[nki] --extra-index-url https://pip.repos.neuron.amazonaws.com +pip install "tilelens[nki]" --extra-index-url https://pip.repos.neuron.amazonaws.com ``` For source installs: @@ -93,7 +121,7 @@ uv sync --extra test # tests but no NKI support ``` ### Testing -* To run core Triton-viz tests, run `pytest tests/`. +* To run core TileLens tests, run `pytest tests/`. * (if NKI installed) To run NKI-specific tests, run `pytest tests/ -m nki`. * To run all tests (Triton + NKI), run `pytest tests/ -m ""`. * To run visualizer web UI tests, run `npm run test:frontend`. @@ -111,10 +139,10 @@ Use the decorator API when writing or modifying a Triton kernel: ```py import triton import triton.language as tl -import triton_viz +import tilelens -@triton_viz.trace("sanitizer") # also supports "tracer" and "profiler" +@tilelens.trace("sanitizer") # also supports "tracer" and "profiler" @triton.jit def kernel(x_ptr, out_ptr, BLOCK: tl.constexpr): offsets = tl.arange(0, BLOCK) @@ -124,21 +152,21 @@ def kernel(x_ptr, out_ptr, BLOCK: tl.constexpr): Use the CLI wrappers to run an existing Python script without editing it. These wrappers patch plain `@triton.jit` kernels, so use them with scripts that do not -already apply `@triton_viz.trace(...)`. +already apply `@tilelens.trace(...)`. ```sh -triton-sanitizer examples/sanitizer/oob_cli.py -triton-profiler examples/profiler/load_store_cli.py -triton-visualizer trace.tvz +tile-sanitizer examples/sanitizer/oob_cli.py +tile-profiler examples/profiler/load_store_cli.py +tile-visualizer trace.tvz ``` For visualizer workflows, save a trace and launch the UI from Python: ```py -import triton_viz +import tilelens -triton_viz.save("trace.tvz") -triton_viz.launch() +tilelens.save("trace.tvz") +tilelens.launch() ``` ## DSL Frontends @@ -147,13 +175,13 @@ Triton is the default DSL frontend. NKI support is optional and selected with the `frontend` argument: ```py -triton_viz.trace("tracer") # Triton -triton_viz.trace("tracer", frontend="nki") # NKI -triton_viz.trace("tracer", frontend="nki_beta2") # NKI Beta 2 +tilelens.trace("tracer") # Triton +tilelens.trace("tracer", frontend="nki") # NKI +tilelens.trace("tracer", frontend="nki_beta2") # NKI Beta 2 ``` -The runtime integration code lives under `triton_viz/core/frontend/`. NKI -simulation runtimes live under `triton_viz/core/simulation/`. +The runtime integration code lives under `tilelens/core/frontend/`. NKI +simulation runtimes live under `tilelens/core/simulation/`. ## Analysis Clients @@ -166,24 +194,28 @@ Analyze kernels across visualization, profiling, and sanitization with a single ### Save and load traces ```py -import triton_viz +import tilelens -triton_viz.save("trace.tvz") -triton_viz.load( +tilelens.save("trace.tvz") +tilelens.load( "trace.tvz" ) # automatically clears out existing records, use kwarg "append=True" to prevent this -triton_viz.launch() +tilelens.launch() ``` -CLI: `triton-visualizer trace.tvz`. The archive is a zip file containing `manifest.json` plus `tensors.npz`, and `triton_viz.load(...)` restores the normal trace state for existing consumers. +CLI: `tile-visualizer trace.tvz`. The archive is a zip file containing `manifest.json` plus `tensors.npz`, and `tilelens.load(...)` restores the normal trace state for existing consumers. + ### Environment variables -Triton-Viz uses a small set of environment variables to configure runtime behavior. Unless noted, boolean flags are enabled only when set to `1`. +TileLens uses a small set of environment variables to configure runtime behavior. Unless noted, boolean flags are enabled only when set to `1`. + +The old names `TRITON_VIZ_VERBOSE`, `TRITON_VIZ_NUM_SMS`, and `TRITON_VIZ_PORT` +still work. If both names are set, TileLens uses the `TILELENS_*` value. -- `TRITON_VIZ_VERBOSE` (default: `0`): enable verbose logging and extra debug output. -- `TRITON_VIZ_NUM_SMS` (default: `1`): number of concurrent SMs to emulate for the CPU interpreter (min 1). -- `TRITON_VIZ_PORT` (default: `8000` with `share=True`, `5001` with `share=False`): port for the Flask server. +- `TILELENS_VERBOSE` (default: `0`): enable verbose logging and extra debug output. +- `TILELENS_NUM_SMS` (default: `1`): number of concurrent SMs to emulate for the CPU interpreter (min 1). +- `TILELENS_PORT` (default: `8000` with `share=True`, `5001` with `share=False`): port for the Flask server. - `ENABLE_SANITIZER` (default: `1`): enable the sanitizer pipeline that checks memory accesses. - `ENABLE_PROFILER` (default: `1`): enable the profiler pipeline that collects performance data. - `ENABLE_TIMING` (default: `0`): collect timing data during execution. @@ -199,7 +231,7 @@ If you're interested in fun puzzles to work with in Triton, do check out: [Trito ## License -Triton-Viz is licensed under the MIT License. See the [LICENSE](LICENSE) for details. +TileLens is licensed under the MIT License. See the [LICENSE](LICENSE) for details. ## Publication If you find this repo useful for your research, please cite our paper: diff --git a/benchmarks/bench_sanitizer.py b/benchmarks/bench_sanitizer.py index 874a32e92..056bc6bb6 100644 --- a/benchmarks/bench_sanitizer.py +++ b/benchmarks/bench_sanitizer.py @@ -35,8 +35,8 @@ import triton import triton.language as tl -import triton_viz -from triton_viz.clients.sanitizer.sanitizer import ( +import tilelens +from tilelens.clients.sanitizer.sanitizer import ( SymbolicSanitizer, _fn_symbolic_cache_set, ) @@ -77,7 +77,7 @@ def _timeout_handler(signum, frame): # --------------------------------------------------------------------------- -@triton_viz.trace(client=gemm_sanitizer) +@tilelens.trace(client=gemm_sanitizer) @triton.jit def gemm_kernel( A, @@ -107,7 +107,7 @@ def gemm_kernel( tl.store(C + C_off, accum) -@triton_viz.trace(client=gemm_oob_sanitizer) +@tilelens.trace(client=gemm_oob_sanitizer) @triton.jit def gemm_oob_kernel( A, @@ -137,7 +137,7 @@ def gemm_oob_kernel( tl.store(C + C_off, accum) -@triton_viz.trace(client=indirect_sanitizer) +@tilelens.trace(client=indirect_sanitizer) @triton.jit def indirect_load_kernel(idx_ptr, src_ptr, dst_ptr, BLOCK: tl.constexpr): pid = tl.program_id(0) @@ -147,7 +147,7 @@ def indirect_load_kernel(idx_ptr, src_ptr, dst_ptr, BLOCK: tl.constexpr): tl.store(dst_ptr + offs, vals) -@triton_viz.trace(client=nested_sanitizer) +@tilelens.trace(client=nested_sanitizer) @triton.jit def nested_loop_kernel(out_ptr, OUTER: tl.constexpr, INNER: tl.constexpr): for i in range(0, OUTER): @@ -156,7 +156,7 @@ def nested_loop_kernel(out_ptr, OUTER: tl.constexpr, INNER: tl.constexpr): tl.store(out_ptr + idx, idx) -@triton_viz.trace(client=block_ptr_sanitizer) +@tilelens.trace(client=block_ptr_sanitizer) @triton.jit def block_pointer_loop_advance_kernel(ptr, N: tl.constexpr, BLOCK: tl.constexpr): block_ptr = tl.make_block_ptr( @@ -182,7 +182,7 @@ def silu(x): return x * tl.sigmoid(x) -@triton_viz.trace(client=swiglu_fwd_sanitizer) +@tilelens.trace(client=swiglu_fwd_sanitizer) @triton.jit def swiglu_forward_kernel( a_ptr, b_ptr, c_ptr, stride, n_cols: tl.constexpr, BLOCK_SIZE: tl.constexpr @@ -203,7 +203,7 @@ def swiglu_forward_kernel( tl.store(c_ptr + col_offsets, c_row, mask=mask) -@triton_viz.trace(client=swiglu_bwd_sanitizer) +@tilelens.trace(client=swiglu_bwd_sanitizer) @triton.jit def swiglu_backward_kernel( dc_ptr, a_ptr, b_ptr, stride, n_cols: tl.constexpr, BLOCK_SIZE: tl.constexpr @@ -237,7 +237,7 @@ def swiglu_backward_kernel( # --------------------------------------------------------------------------- -@triton_viz.trace(client=cross_entropy_sanitizer) +@tilelens.trace(client=cross_entropy_sanitizer) @triton.jit def liger_cross_entropy_kernel( X_ptr, @@ -329,7 +329,7 @@ def liger_cross_entropy_kernel( tl.store(X_ptr + y, X_y) -@triton_viz.trace(client=jsd_sanitizer) +@tilelens.trace(client=jsd_sanitizer) @triton.jit def jsd_kernel( X_ptr, # input in logspace, X = log Q @@ -382,7 +382,7 @@ def jsd_kernel( tl.store(dX_ptr + offsets, dX, mask=mask) -@triton_viz.trace(client=element_mul_sanitizer) +@tilelens.trace(client=element_mul_sanitizer) @triton.jit def element_mul_kernel( X_ptr, @@ -408,7 +408,7 @@ def element_mul_kernel( # --------------------------------------------------------------------------- -@triton_viz.trace(client=flaggems_sanitizer) +@tilelens.trace(client=flaggems_sanitizer) @triton.jit def flaggems_ln_persistent_kernel( in_ptr, @@ -449,7 +449,7 @@ def flaggems_ln_persistent_kernel( tl.store(out_ptr + pid * N + n_offsets, out, mask=mask) -@triton_viz.trace(client=flaggems_sanitizer) +@tilelens.trace(client=flaggems_sanitizer) @triton.jit def flaggems_ln_multiline_kernel( in_ptr, @@ -497,7 +497,7 @@ def flaggems_ln_multiline_kernel( tl.store(out_ptr + m_offsets[:, None] * N + n_offsets, out, mask=mask) -@triton_viz.trace(client=flaggems_sanitizer) +@tilelens.trace(client=flaggems_sanitizer) @triton.jit def flaggems_ln_loop_kernel( in_ptr, @@ -580,7 +580,7 @@ def flaggems_ln_loop_kernel( tl.store(out_ptr + pid * N + n_offsets, out) -@triton_viz.trace(client=flaggems_sanitizer) +@tilelens.trace(client=flaggems_sanitizer) @triton.jit def flaggems_ln_backward_kernel( dY, @@ -647,7 +647,7 @@ def flaggems_ln_backward_kernel( tl.store(dX + cols, dx, mask=mask) -@triton_viz.trace(client=flaggems_sanitizer) +@tilelens.trace(client=flaggems_sanitizer) @triton.jit def flaggems_ln_wb_backward_kernel( dY, @@ -1409,33 +1409,33 @@ def _fljsd_run_all(templates): ( "triton_kernels._masked_compaction", "masked-compaction", - "test_triton_viz_sanitizer_masked_compaction", + "test_tilelens_sanitizer_masked_compaction", ), ( "triton_kernels._ragged_tensor_metadata_memset_compute", "ragged-metadata-make", - "test_triton_viz_sanitizer_make_ragged_tensor_metadata", + "test_tilelens_sanitizer_make_ragged_tensor_metadata", ), ( "triton_kernels._remap_ragged_tensor_metadata", "ragged-metadata-remap", - "test_triton_viz_sanitizer_remap_ragged_tensor_metadata", + "test_tilelens_sanitizer_remap_ragged_tensor_metadata", ), ( "triton_kernels._topk_forward", "topk-forward", - "test_triton_viz_sanitizer_topk_forward", + "test_tilelens_sanitizer_topk_forward", ), ( "triton_kernels._topk_backward", "topk-backward", - "test_triton_viz_sanitizer_topk_backward", + "test_tilelens_sanitizer_topk_backward", ), - ("triton_kernels._swiglu", "swiglu", "test_triton_viz_sanitizer_swiglu"), + ("triton_kernels._swiglu", "swiglu", "test_tilelens_sanitizer_swiglu"), ( "triton_kernels.upcast_mxfp4_tile", "mxfp4-tile-upcast", - "test_triton_viz_sanitizer_mxfp4_tile_upcast", + "test_tilelens_sanitizer_mxfp4_tile_upcast", ), ] diff --git a/docs/index.html b/docs/index.html index 2b7d4a3e9..45d30a310 100644 --- a/docs/index.html +++ b/docs/index.html @@ -3,7 +3,7 @@ - Triton-Viz | Visualization Toolkit for Triton GPU Programming + TileLens | Visualization Toolkit for Triton GPU Programming @@ -559,8 +559,8 @@ @@ -578,8 +578,8 @@

    Analyze and debug your tile-based DSL kernels

    Install
    - pip install triton-viz -
    @@ -587,8 +587,8 @@

    Analyze and debug your tile-based DSL kernels

    Install (NKI support)
    - pip install triton-viz[nki] --extra-index-url https://pip.repos.neuron.amazonaws.com -
    @@ -596,7 +596,7 @@

    Analyze and debug your tile-based DSL kernels

    Note: A browser with WebGL/OpenGL enabled is required (standard in modern browsers).
    - See README.md for development installs, web UI builds, DSL frontend selection, and full documentation. + See README.md for development installs, web UI builds, DSL frontend selection, and full documentation.

    @@ -637,10 +637,10 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def add_3d_slices_kernel(
     input_ptr1,
@@ -689,7 +689,7 @@ 
    Analysis Clients
    
         BLOCK_SIZE_Y=BLOCK_SIZE_Y,
         BLOCK_SIZE_Z=BLOCK_SIZE_Z,
     )
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
    @@ -713,14 +713,14 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
 BLOCK = 1024
 SHAPE = (2, 3, 4, 3, 4, 2, 4, 2, 3)
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def copy_9d_kernel(x_ptr, y_ptr, n_elements, BLOCK: tl.constexpr):
     pid = tl.program_id(0)
@@ -736,7 +736,7 @@ 
    Analysis Clients
    
     n_elements = x.numel()
     grid = (triton.cdiv(n_elements, BLOCK),)
     copy_9d_kernel[grid](x, y, n_elements, BLOCK=BLOCK)
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
    @@ -760,10 +760,10 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def add_3d_slices_kernel(
     input_ptr1,
@@ -812,7 +812,7 @@ 
    Analysis Clients
    
         BLOCK_SIZE_Y=BLOCK_SIZE_Y,
         BLOCK_SIZE_Z=BLOCK_SIZE_Z,
     )
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
    @@ -836,10 +836,10 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def matmul_kernel(
     a_ptr,
@@ -888,7 +888,7 @@ 
    Analysis Clients
    
         BLOCK_N=BLOCK_N,
         BLOCK_K=BLOCK_K,
     )
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
    @@ -912,10 +912,10 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def matmul_kernel(
     a_ptr,
@@ -964,7 +964,7 @@ 
    Analysis Clients
    
         BLOCK_N=BLOCK_N,
         BLOCK_K=BLOCK_K,
     )
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
    @@ -988,10 +988,10 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def matmul_kernel(
     a_ptr,
@@ -1040,7 +1040,7 @@ 
    Analysis Clients
    
         BLOCK_N=BLOCK_N,
         BLOCK_K=BLOCK_K,
     )
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
    @@ -1066,10 +1066,10 @@

    Analysis Clients

    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("profiler")
+@tilelens.trace("profiler")
 @triton.jit
 def simple_kernel(
     x_ptr,
@@ -1164,10 +1164,10 @@ 
    Analysis Clients
    
                                                                             
    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("sanitizer")
+@tilelens.trace("sanitizer")
 @triton.jit
 def gemm_kernel(
     A, B, C, M: tl.constexpr, N: tl.constexpr, K: tl.constexpr, TILE_SIZE: tl.constexpr
@@ -1225,10 +1225,10 @@ 
    Analysis Clients
    
                                                                             
    import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("sanitizer")
+@tilelens.trace("sanitizer")
 @triton.jit
 def gemm_kernel(
     A, B, C, M: tl.constexpr, N: tl.constexpr, K: tl.constexpr, TILE_SIZE: tl.constexpr
@@ -1274,7 +1274,7 @@ 
    Analysis Clients
    
                                                                             
    TERMINAL OUTPUT
    
                                                                             
    🚨 ILLEGAL MEMORY ACCESS DETECTED 🚨
 ━━━ Code Context ━━━
-File: triton-viz/examples/sanitizer/gemm_oob.py
+File: tilelens/examples/sanitizer/gemm_oob.py
 Function: gemm_kernel
 Line 25:
   22 │     for k_block in range(K // TILE_SIZE):
@@ -1385,27 +1385,27 @@ 
    Analysis Clients
    
     
     
    
         
    Sponsored by
    
-        
    Special thanks to AMD for supporting Triton-Viz.
    
+        
    Special thanks to AMD for supporting TileLens.
    
     
    
     
     
    
         
    Ready to visualize your kernels?
    
-        
    Get started with Triton-Viz in minutes. No GPU required for development.
    
+        
    Get started with TileLens in minutes. No GPU required for development.
    
         
    
-            View on GitHub
-            Browse Examples
+            View on GitHub
+            Browse Examples
         
    
     
    
     
     
@@ -1456,12 +1456,12 @@ 
    Ready to visualize your kernels?
    
         }
         return highlighted;
     }
-    function highlightTritonVizLines() {
+    function highlightTileLensLines() {
         const patterns = [
-            'import triton_viz',
-            'from triton_viz',
-            '@triton_viz.trace',
-            'triton_viz.launch',
+            'import tilelens',
+            'from tilelens',
+            '@tilelens.trace',
+            'tilelens.launch',
         ];
         document.querySelectorAll('pre code.language-python').forEach(code => {
             const lines = code.textContent.split('\n');
@@ -1478,7 +1478,7 @@ 
    Ready to visualize your kernels?
    
             code.innerHTML = updated.join('\n');
         });
     }
-    highlightTritonVizLines();
+    highlightTileLensLines();
 
 
 
diff --git a/examples/multithread.py b/examples/multithread.py
index 39c3223c3..d7349aaee 100644
--- a/examples/multithread.py
+++ b/examples/multithread.py
@@ -60,7 +60,7 @@ def main():
     parser.add_argument(
         "--viz",
         action="store_true",
-        help="Enable triton-viz tracing (uses tracer client).",
+        help="Enable tilelens tracing (uses tracer client).",
     )
     parser.add_argument(
         "--num-sms",
@@ -73,16 +73,16 @@ def main():
 
     trace_decorator = None
     if args.viz:
-        import triton_viz
+        import tilelens
 
-        triton_viz.config.num_sms = max(1, args.num_sms)
-        trace_decorator = triton_viz.trace("tracer")
+        tilelens.config.num_sms = max(1, args.num_sms)
+        trace_decorator = tilelens.trace("tracer")
 
     producer_consumer, racing_threads = make_kernels(trace_decorator)
 
     device = (
         "cpu" if args.viz else "cuda"
-    )  # triton-viz only supports CPU, triton only supports CUDA (TRITON_INTERPRET=1 runs on CPU but hangs)
+    )  # tilelens only supports CPU, triton only supports CUDA (TRITON_INTERPRET=1 runs on CPU but hangs)
 
     # Use smaller workloads on CPU to keep interpreter runs snappy.
     pc_N = 16384 * 16 if device == "cuda" else 2048
diff --git a/examples/nki/matmul.py b/examples/nki/matmul.py
index e0a7c94be..d3f702033 100644
--- a/examples/nki/matmul.py
+++ b/examples/nki/matmul.py
@@ -1,11 +1,11 @@
 from neuronxcc import nki
 import neuronxcc.nki.language as nl
 
-import triton_viz
+import tilelens
 import numpy as np
 import math
 
-TRITON_VIZ_ENABLED = True  # True = enable visualizer, False = run compiler
+TILELENS_ENABLED = True  # True = enable visualizer, False = run compiler
 TRANSPOSED = False  # True = run matmul_kernel, False = run matmul_kernel_lhsT
 
 
@@ -137,12 +137,12 @@ def _run_demo():
     result = np.empty((M, N), dtype=lhs_small.dtype)
     kernel_args = (lhs_small, rhs_small, result)
 
-    if TRITON_VIZ_ENABLED:
+    if TILELENS_ENABLED:
         print("Executing matmul_kernel with NKI interpreter...")
-        traced_kernel = triton_viz.trace("tracer", frontend="nki")(kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki")(kernel)
         kernel_instance = traced_kernel[kernel_grid]
         kernel_instance(*kernel_args)
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         print("Executing NKI JIT-ed matmul kernel...")
         compiled_kernel = nki.jit(kernel, kernel_return=False)
diff --git a/examples/nki/rmsnorm.py b/examples/nki/rmsnorm.py
index 87c8603ca..d5f2bd87e 100644
--- a/examples/nki/rmsnorm.py
+++ b/examples/nki/rmsnorm.py
@@ -1,11 +1,11 @@
 import neuronxcc.nki as nki
 import neuronxcc.nki.language as nl
-from triton_viz.clients import Tracer
-from triton_viz.core.trace import launches
+from tilelens.clients import Tracer
+from tilelens.core.trace import launches
 import math
 import numpy as np
 import torch
-import triton_viz
+import tilelens
 
 
 def nki_rmsnorm_kernel(a_tensor, g_tensor, result):
@@ -77,7 +77,7 @@ def torch_rmsnorm_kernel(a_tensor, g_tensor):
 
 
 def _run_demo():
-    triton_viz_enabled = True
+    tilelens_enabled = True
     kernel_grid = (1, 1, 1)
     b_dim, d_dim = 32, 32
     a_tensor = torch.arange(b_dim * d_dim).float().view(b_dim, d_dim)
@@ -85,9 +85,9 @@ def _run_demo():
     result = torch.empty_like(a_tensor).numpy()
     kernel_args = (a_tensor.numpy(), g_tensor.numpy(), result)
 
-    if triton_viz_enabled:
+    if tilelens_enabled:
         print("Executing kernel with NKI interpreter...")
-        traced_kernel = triton_viz.trace(client=Tracer(), frontend="nki")(
+        traced_kernel = tilelens.trace(client=Tracer(), frontend="nki")(
             nki_rmsnorm_kernel
         )
         kernel_instance = traced_kernel[kernel_grid]
@@ -107,7 +107,7 @@ def _run_demo():
                     print(f"  masks shape: {record.masks.shape}")
 
         try:
-            triton_viz.launch(share=False)
+            tilelens.launch(share=False)
         except Exception as e:
             print(f"\nError during visualization: {e}")
             import traceback
diff --git a/examples/nki/rope.py b/examples/nki/rope.py
index 0b959895c..b84a70411 100644
--- a/examples/nki/rope.py
+++ b/examples/nki/rope.py
@@ -3,9 +3,9 @@
 import neuronxcc.nki.language as nl
 import torch
 from neuronxcc import nki
-import triton_viz
-from triton_viz.clients import Tracer
-from triton_viz.core.trace import launches
+import tilelens
+from tilelens.clients import Tracer
+from tilelens.core.trace import launches
 import numpy as np
 
 os.environ["NEURON_FRAMEWORK_DEBUG"] = "1"
@@ -209,11 +209,11 @@ def rotate_half(x):
 
 
 def _run_demo():
-    triton_viz_enabled = True
+    tilelens_enabled = True
     h_dim, s_dim, d_dim = 2, 4, 8
     kernel_grid = (h_dim,)
 
-    if triton_viz_enabled:
+    if tilelens_enabled:
         q = torch.randn(h_dim, s_dim, d_dim)
         k = torch.randn(h_dim, s_dim, d_dim)
         position_ids = torch.arange(s_dim)
@@ -231,9 +231,7 @@ def _run_demo():
         )
 
         print("Executing rotary embedding kernel with NKI interpreter...")
-        traced_kernel = triton_viz.trace(client=Tracer(), frontend="nki")(
-            nki_rope_kernel
-        )
+        traced_kernel = tilelens.trace(client=Tracer(), frontend="nki")(nki_rope_kernel)
         kernel_instance = traced_kernel[kernel_grid]
         kernel_instance(*kernel_args)
 
@@ -243,7 +241,7 @@ def _run_demo():
             print(f"Number of records: {len(launch.records)}")
 
         try:
-            triton_viz.launch(share=False)
+            tilelens.launch(share=False)
         except Exception as e:
             print(f"Visualization error: {e}")
     else:
diff --git a/examples/nki/softmax.py b/examples/nki/softmax.py
index fb051b7a8..80e0b1bdb 100644
--- a/examples/nki/softmax.py
+++ b/examples/nki/softmax.py
@@ -1,9 +1,9 @@
 from neuronxcc import nki
 import neuronxcc.nki.language as nl
 
-import triton_viz
-from triton_viz.clients import Tracer
-from triton_viz.core.trace import launches
+import tilelens
+from tilelens.clients import Tracer
+from tilelens.core.trace import launches
 import numpy as np
 import math
 
@@ -31,17 +31,15 @@ def softmax_kernel(in_tensor, out_tensor):
 
 
 def _run_demo():
-    triton_viz_enabled = True
+    tilelens_enabled = True
     kernel_grid = (1, 1, 1)
     x_small = np.random.rand(16, 32).astype(np.float32)
     y_small = np.empty(x_small.shape, dtype=x_small.dtype)
     kernel_args = (x_small, y_small)
 
-    if triton_viz_enabled:
+    if tilelens_enabled:
         print("Executing softmax_kernel with NKI interpreter...")
-        traced_kernel = triton_viz.trace(client=Tracer(), frontend="nki")(
-            softmax_kernel
-        )
+        traced_kernel = tilelens.trace(client=Tracer(), frontend="nki")(softmax_kernel)
         kernel_instance = traced_kernel[kernel_grid]
         kernel_instance(*kernel_args)
 
@@ -59,7 +57,7 @@ def _run_demo():
                     print(f"  masks shape: {record.masks.shape}")
 
         try:
-            triton_viz.launch(share=False)
+            tilelens.launch(share=False)
         except Exception as e:
             print(f"\nError during visualization: {e}")
             import traceback
diff --git a/examples/nki_beta2/add.py b/examples/nki_beta2/add.py
index 8ee7683bc..4ea494922 100644
--- a/examples/nki_beta2/add.py
+++ b/examples/nki_beta2/add.py
@@ -4,8 +4,8 @@
 import numpy as np
 import torch
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 
 
 def nki_tensor_add_kernel(a_input, b_input, result):
@@ -63,14 +63,14 @@ def _run_demo():
     kernel_args = (a, b, result)
     expected = a + b
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(kernel)
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected, result)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:  # Note: no official NKI Beta 2 CPU interpreter exists so run on XLA
         result = _run_with_xla(kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected, result)
diff --git a/examples/nki_beta2/matmul.py b/examples/nki_beta2/matmul.py
index dac4c695e..c09575405 100644
--- a/examples/nki_beta2/matmul.py
+++ b/examples/nki_beta2/matmul.py
@@ -4,8 +4,8 @@
 
 import numpy as np
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 
 
 def matmul_kernel(lhsT, rhs, result):
@@ -109,14 +109,14 @@ def _run_demo():
     kernel_args = (lhs_small, rhs_small, result)
     expected = lhs_small.T @ rhs_small
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(kernel)
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected, result)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         result = _run_with_xla(kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected, result)
diff --git a/examples/nki_beta2/mlp.py b/examples/nki_beta2/mlp.py
index 59686287a..db8a1809e 100644
--- a/examples/nki_beta2/mlp.py
+++ b/examples/nki_beta2/mlp.py
@@ -4,8 +4,8 @@
 
 import numpy as np
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 TILE_M = 128
 TILE_K = 128
 TILE_H = 128
@@ -170,14 +170,14 @@ def _run_demo():
     kernel_args = (x_t, w1_padded, w2_padded, out)
     expected = np.maximum(x @ w1, 0.0) @ w2
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(mlp_kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(mlp_kernel)
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected, out[:batch, :out_dim], atol=1e-4, rtol=1e-4)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         out = _run_with_xla(mlp_kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected, out[:batch, :out_dim], atol=1e-4, rtol=1e-4)
diff --git a/examples/nki_beta2/rmsnorm.py b/examples/nki_beta2/rmsnorm.py
index 6a497cf18..811a18ff1 100644
--- a/examples/nki_beta2/rmsnorm.py
+++ b/examples/nki_beta2/rmsnorm.py
@@ -4,8 +4,8 @@
 
 import numpy as np
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 
 
 def rmsnorm_kernel(x, gamma, out, eps=1e-6):
@@ -94,14 +94,14 @@ def _run_demo():
     kernel_args = (x, gamma, out)
     expected = _numpy_rmsnorm(x, gamma)
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(rmsnorm_kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(rmsnorm_kernel)
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected, out)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         out = _run_with_xla(rmsnorm_kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected, out)
diff --git a/examples/nki_beta2/rope.py b/examples/nki_beta2/rope.py
index f388baa05..ee7917521 100644
--- a/examples/nki_beta2/rope.py
+++ b/examples/nki_beta2/rope.py
@@ -4,8 +4,8 @@
 
 import numpy as np
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 
 
 def _rotate_half(x):
@@ -216,15 +216,15 @@ def _run_demo():
         head_dim,
     )
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(rope_kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(rope_kernel)
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected_q, out_q4d)
         assert np.allclose(expected_k, out_k4d)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         out_q4d, out_k4d = _run_with_xla(rope_kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected_q, out_q4d)
diff --git a/examples/nki_beta2/softmax.py b/examples/nki_beta2/softmax.py
index 4cc695289..e120da954 100644
--- a/examples/nki_beta2/softmax.py
+++ b/examples/nki_beta2/softmax.py
@@ -4,8 +4,8 @@
 
 import numpy as np
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 
 
 def softmax_kernel(x, out):
@@ -83,14 +83,14 @@ def _run_demo():
     kernel_args = (x, out)
     expected = _numpy_softmax(x)
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(softmax_kernel)
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(softmax_kernel)
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected, out)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         out = _run_with_xla(softmax_kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected, out)
diff --git a/examples/nki_beta2/tiled_attention.py b/examples/nki_beta2/tiled_attention.py
index 91ee36ce9..b8fddacf6 100644
--- a/examples/nki_beta2/tiled_attention.py
+++ b/examples/nki_beta2/tiled_attention.py
@@ -4,8 +4,8 @@
 import nki.language as nl
 import numpy as np
 
-TRITON_VIZ_ENABLED = True
-PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before Triton-Viz simulation for stronger semantic checks.
+TILELENS_ENABLED = True
+PRE_TRACE = True  # Run NKI Beta 2 compiler tracing before TileLens simulation for stronger semantic checks.
 
 
 def tiled_attention_kernel(
@@ -372,16 +372,16 @@ def _run_demo():
     )
     expected4d = _numpy_tiled_attention(q4d, k4d, v4d)
 
-    if TRITON_VIZ_ENABLED:
-        import triton_viz
+    if TILELENS_ENABLED:
+        import tilelens
 
-        traced_kernel = triton_viz.trace("tracer", frontend="nki_beta2")(
+        traced_kernel = tilelens.trace("tracer", frontend="nki_beta2")(
             tiled_attention_kernel
         )
         traced_kernel[kernel_grid](*kernel_args, pre_trace=PRE_TRACE)
         assert np.allclose(expected4d, out4d)
         print("☑️ Actual equals expected!")
-        triton_viz.launch(share=False)
+        tilelens.launch(share=False)
     else:
         out4d = _run_with_xla(tiled_attention_kernel, kernel_grid, *kernel_args)
         assert np.allclose(expected4d, out4d)
diff --git a/examples/profiler/load_store.py b/examples/profiler/load_store.py
index e9710de50..1e15e9499 100644
--- a/examples/profiler/load_store.py
+++ b/examples/profiler/load_store.py
@@ -1,11 +1,11 @@
 import torch
 import triton
 import triton.language as tl
-import triton_viz
-from triton_viz.core.config import config as cfg
+import tilelens
+from tilelens.core.config import config as cfg
 
 
-@triton_viz.trace("profiler")
+@tilelens.trace("profiler")
 @triton.jit
 def simple_kernel(
     x_ptr,
diff --git a/examples/sanitizer/gemm.py b/examples/sanitizer/gemm.py
index 18126591a..2fb7e7a26 100644
--- a/examples/sanitizer/gemm.py
+++ b/examples/sanitizer/gemm.py
@@ -3,10 +3,10 @@
 import triton
 import triton.language as tl
 
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("sanitizer")
+@tilelens.trace("sanitizer")
 @triton.jit
 def gemm_kernel(
     A, B, C, M: tl.constexpr, N: tl.constexpr, K: tl.constexpr, TILE_SIZE: tl.constexpr
diff --git a/examples/sanitizer/gemm_oob.py b/examples/sanitizer/gemm_oob.py
index 6cae95075..d594500d5 100644
--- a/examples/sanitizer/gemm_oob.py
+++ b/examples/sanitizer/gemm_oob.py
@@ -3,10 +3,10 @@
 import triton
 import triton.language as tl
 
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("sanitizer")
+@tilelens.trace("sanitizer")
 @triton.jit
 def gemm_kernel(
     A, B, C, M: tl.constexpr, N: tl.constexpr, K: tl.constexpr, TILE_SIZE: tl.constexpr
diff --git a/examples/sanitizer/gluon_tma_oob.py b/examples/sanitizer/gluon_tma_oob.py
index 7f9eb8409..39976a389 100644
--- a/examples/sanitizer/gluon_tma_oob.py
+++ b/examples/sanitizer/gluon_tma_oob.py
@@ -3,8 +3,8 @@
 from triton.experimental.gluon import language as gl
 from triton.experimental.gluon.language.nvidia.hopper import mbarrier, tma
 
-import triton_viz
-from triton_viz.clients.sanitizer.sanitizer import Sanitizer
+import tilelens
+from tilelens.clients.sanitizer.sanitizer import Sanitizer
 
 
 BLOCK_M = 16
@@ -40,7 +40,7 @@ def gluon_tma_oob_kernel(
 def run(abort_on_error: bool = True):
     x = torch.zeros((BLOCK_M, BLOCK_N), dtype=torch.float32)
     layout = gl.NVMMASharedLayout.get_default_for([BLOCK_M, BLOCK_N], gl.float32)
-    kernel = triton_viz.trace(
+    kernel = tilelens.trace(
         client=Sanitizer(abort_on_error=abort_on_error),
         frontend="gluon",
     )(gluon_tma_oob_kernel)
diff --git a/examples/sanitizer/oob_tensor_view.py b/examples/sanitizer/oob_tensor_view.py
index 6ffdd6373..defeafbb5 100644
--- a/examples/sanitizer/oob_tensor_view.py
+++ b/examples/sanitizer/oob_tensor_view.py
@@ -2,7 +2,7 @@
 
 When a masked ``tl.load`` uses a logical length larger than the view's
 element count, the JIT performs a "wild pointer read" past the underlying
-storage. This example wraps the kernel with the triton-viz sanitizer so the
+storage. This example wraps the kernel with the tilelens sanitizer so the
 OOB access is caught symbolically instead of silently returning OOB memory.
 """
 
@@ -11,10 +11,10 @@
 import triton
 import triton.language as tl
 
-import triton_viz
+import tilelens
 
 
-@triton_viz.trace("sanitizer")
+@tilelens.trace("sanitizer")
 @triton.jit
 def oob_load_kernel(
     vals_ptr,
diff --git a/examples/visualizer/3dims.py b/examples/visualizer/3dims.py
index b58683b77..ada043888 100644
--- a/examples/visualizer/3dims.py
+++ b/examples/visualizer/3dims.py
@@ -3,15 +3,15 @@
 import triton.language as tl
 
 
-import triton_viz
-from triton_viz.clients import Tracer
+import tilelens
+from tilelens.clients import Tracer
 
 BLOCK_SIZE_X = 32
 BLOCK_SIZE_Y = 8
 BLOCK_SIZE_Z = 4
 
 
-@triton_viz.trace(client=Tracer())
+@tilelens.trace(client=Tracer())
 @triton.jit
 def add_3d_slices_kernel(
     input_ptr1,
@@ -109,7 +109,7 @@ def add_3d_slices(input1, input2, output):
 
     # Call the kernel
     add_3d_slices(input1, input2, output)
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
 
     # Verify the result
     expected_output = input1 + input2
diff --git a/examples/visualizer/9dims_copy.py b/examples/visualizer/9dims_copy.py
index 26a7f173c..210746c3d 100644
--- a/examples/visualizer/9dims_copy.py
+++ b/examples/visualizer/9dims_copy.py
@@ -1,14 +1,14 @@
 import torch
 import triton
 import triton.language as tl
-import triton_viz
+import tilelens
 
 
 BLOCK = 1024
 SHAPE = (2, 3, 4, 3, 4, 2, 4, 2, 3)
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def copy_9d_kernel(
     x_ptr,
@@ -26,7 +26,7 @@ def copy_9d_kernel(
 
 
 def run_demo() -> None:
-    """Run a rank-9 copy example and launch Triton Viz."""
+    """Run a rank-9 copy example and launch TileLens."""
     torch.manual_seed(0)
     x = torch.randn(SHAPE, device="cpu")
     y = torch.empty_like(x)
@@ -41,7 +41,7 @@ def run_demo() -> None:
     )
 
     assert torch.allclose(y, x), "copy kernel output mismatch"
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
 
 
 if __name__ == "__main__":
diff --git a/examples/visualizer/indirect_load.py b/examples/visualizer/indirect_load.py
index d967ced72..11e75837b 100644
--- a/examples/visualizer/indirect_load.py
+++ b/examples/visualizer/indirect_load.py
@@ -2,13 +2,13 @@
 import triton
 import triton.language as tl
 
-import triton_viz
+import tilelens
 
 N = 64
 BLOCK = 16
 
 
-@triton_viz.trace("tracer")
+@tilelens.trace("tracer")
 @triton.jit
 def indirect_load_kernel(x_ptr, rand_ptr, out_ptr, BLOCK: tl.constexpr):
     offs = tl.arange(0, BLOCK)
@@ -24,7 +24,7 @@ def run():
     rand = torch.randperm(N, dtype=torch.int32, device=device)
     out = torch.zeros_like(x)
     indirect_load_kernel[(1,)](x, rand, out, BLOCK=BLOCK)
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
 
 
 if __name__ == "__main__":
diff --git a/examples/visualizer/matmul.py b/examples/visualizer/matmul.py
index cd8827d77..095965db3 100644
--- a/examples/visualizer/matmul.py
+++ b/examples/visualizer/matmul.py
@@ -2,12 +2,12 @@
 import triton
 import triton.language as tl
 
-import triton_viz
-from triton_viz.clients import Tracer
+import tilelens
+from tilelens.clients import Tracer
 
 
 # Simple matmul kernel producing a C = A @ B (fp32, small sizes for demo)
-@triton_viz.trace(client=Tracer())
+@tilelens.trace(client=Tracer())
 @triton.jit
 def matmul_kernel(
     a_ptr,
@@ -86,7 +86,7 @@ def run_demo():
     assert torch.allclose(c, ref, atol=1e-3), "matmul result mismatch"
 
     # Launch viz UI in blocking (share=True) mode
-    triton_viz.launch(share=False, port=5001)
+    tilelens.launch(share=False, port=5001)
 
 
 if __name__ == "__main__":
diff --git a/frontend/GLOSSARY.md b/frontend/GLOSSARY.md
index 04fbb450c..69b54b25b 100644
--- a/frontend/GLOSSARY.md
+++ b/frontend/GLOSSARY.md
@@ -1,6 +1,6 @@
 # Glossary
 
-This glossary defines the canonical UI terms, data types, and events used across the Triton-Viz frontend.
+This glossary defines the canonical UI terms, data types, and events used across the TileLens frontend.
 
 ## UI Surfaces
 - Control Panel: left sidebar with program controls, operation controls, and the code peek panel.
diff --git a/frontend/components/op_workspace.ts b/frontend/components/op_workspace.ts
index 007e7001b..e5ac7f81f 100644
--- a/frontend/components/op_workspace.ts
+++ b/frontend/components/op_workspace.ts
@@ -79,7 +79,7 @@ export class OpWorkspace {
 
         this.ensureGlobalCodeToggle();
         try {
-            window.__tritonVizActiveBlock = this;
+            window.__tileLensActiveBlock = this;
         } catch (error) {}
     }
 
@@ -303,7 +303,7 @@ export class OpWorkspace {
         const canReuseViz = !!options.preserveViewState && isSameTab && !!this.contentArea.__vizGetState;
         const preserveCodePanel = options.refreshCodePanel === false;
         if (preserveCodePanel) {
-            try { window.__tritonVizPreserveCodePanel = true; } catch (error) {}
+            try { window.__tileLensPreserveCodePanel = true; } catch (error) {}
         }
         const isTensorOp = op.type === 'Dot' || op.type === 'Load' || op.type === 'Store' || op.type === 'Transfer';
         const reuseTensorView = canReuseViz && isTensorOp && this.lastOpType === op.type;
@@ -314,7 +314,7 @@ export class OpWorkspace {
             }
         } finally {
             if (preserveCodePanel) {
-                try { window.__tritonVizPreserveCodePanel = false; } catch (error) {}
+                try { window.__tileLensPreserveCodePanel = false; } catch (error) {}
             }
         }
         if (!reuseTensorView) {
@@ -366,7 +366,7 @@ export class OpWorkspace {
     }
 
     ensureGlobalCodeToggle(): void {
-        if (window.__tritonVizCodeToggle) return;
+        if (window.__tileLensCodeToggle) return;
         let panel: HTMLDivElement | null = null;
         let visible = false;
 
@@ -380,7 +380,7 @@ export class OpWorkspace {
         };
 
         const getActiveUuid = (): string | null => {
-            const activeBlock = window.__tritonVizActiveBlock;
+            const activeBlock = window.__tileLensActiveBlock;
             const candidate = window.current_op_uuid ?? activeBlock?.blockData?.[0]?.uuid;
             return candidate ?? null;
         };
@@ -461,16 +461,16 @@ export class OpWorkspace {
             return visible;
         };
 
-        window.__tritonVizCodeToggle = togglePanel;
-        window.__tritonVizCodeHide = hidePanel;
-        window.__tritonVizCodeVisible = () => visible;
+        window.__tileLensCodeToggle = togglePanel;
+        window.__tileLensCodeHide = hidePanel;
+        window.__tileLensCodeVisible = () => visible;
     }
 
     syncCodePanel(forceShow = false): void {
-        if (!window.__tritonVizCodeToggle) return;
-        const isVisible = window.__tritonVizCodeVisible ? window.__tritonVizCodeVisible() : false;
+        if (!window.__tileLensCodeToggle) return;
+        const isVisible = window.__tileLensCodeVisible ? window.__tileLensCodeVisible() : false;
         if (!forceShow && !isVisible) return;
-        const result = window.__tritonVizCodeToggle(true);
+        const result = window.__tileLensCodeToggle(true);
         const setOpControlState = window.setOpControlState;
         if (!setOpControlState) return;
         if (result && typeof (result as any).then === 'function') {
diff --git a/frontend/components/tensor_view.ts b/frontend/components/tensor_view.ts
index 1b2123069..9f547dd2f 100644
--- a/frontend/components/tensor_view.ts
+++ b/frontend/components/tensor_view.ts
@@ -1932,8 +1932,8 @@ export function createTensorVisualization(
             state.rafId = requestAnimationFrame(() => { state.rafId = null; orbitControls.update(); syncClipPlanes(); renderer.render(scene, camera); if (state.renderPending) { state.renderPending = false; ctx.requestRender(); } });
         };
         const handleTextSync = (): void => { ctx.requestRender(); };
-        window.addEventListener('triton-viz-text-sync', handleTextSync);
-        disposer.add(() => window.removeEventListener('triton-viz-text-sync', handleTextSync));
+        window.addEventListener('tilelens-text-sync', handleTextSync);
+        disposer.add(() => window.removeEventListener('tilelens-text-sync', handleTextSync));
         let lastWidth = 0;
         let lastHeight = 0;
         const resizeRenderer = (): void => {
@@ -2538,7 +2538,7 @@ export function createTensorVisualization(
                 requestRender();
                 return state.colorizeOn;
             },
-            toggleShowCode: () => window.__tritonVizCodeToggle?.(),
+            toggleShowCode: () => window.__tileLensCodeToggle?.(),
             toggleHistogram: (): boolean => {
                 const isVisible = vizCache.histogramUI.overlay.style.display === 'block';
                 if (isVisible) {
diff --git a/frontend/core/api.ts b/frontend/core/api.ts
index 0bbca07f3..35ae41ad6 100644
--- a/frontend/core/api.ts
+++ b/frontend/core/api.ts
@@ -8,7 +8,7 @@ export type RequestOptions = {
 
 const getDefaultBase = (): string => {
     if (typeof globalThis === 'undefined') return '';
-    const globalBase = (globalThis as typeof globalThis & { __TRITON_VIZ_API__?: string }).__TRITON_VIZ_API__;
+    const globalBase = (globalThis as typeof globalThis & { __TILELENS_API__?: string }).__TILELENS_API__;
     if (globalBase) return globalBase;
     if (typeof window === 'undefined' || !window.location) return '';
     // derive a stable base from the current page location for proxy paths
diff --git a/frontend/core/logger.ts b/frontend/core/logger.ts
index 9f112b90b..1eb99f1ab 100644
--- a/frontend/core/logger.ts
+++ b/frontend/core/logger.ts
@@ -1,4 +1,4 @@
-const LOG_PREFIX = '[triton-viz]';
+const LOG_PREFIX = '[tilelens]';
 
 export function logAction(action: string, details: Record = {}): void {
     console.info(LOG_PREFIX, action, details);
diff --git a/frontend/core/visualization.ts b/frontend/core/visualization.ts
index 976a18c73..fa32c8314 100644
--- a/frontend/core/visualization.ts
+++ b/frontend/core/visualization.ts
@@ -30,7 +30,7 @@ let maxZ = 0;
 const PROGRAM_AXES: Array = ['x', 'y', 'z'];
 const appDisposer = createDisposer();
 let pidDisposer = createDisposer();
-const THEME_STORAGE_KEY = 'triton-viz-theme';
+const THEME_STORAGE_KEY = 'tilelens-theme';
 const DEV_OVERLAY_QUERY_KEY = 'dev';
 const DEV_OVERLAY_TOGGLE = 'KeyD';
 let devOverlayEnabled = false;
@@ -69,7 +69,7 @@ function setOpControlState(nextState: Partial> = {}):
         ...nextState,
     };
     setToggles(toggles);
-    try { window.__tritonVizOpState = { ...toggles }; } catch (err) {}
+    try { window.__tileLensOpState = { ...toggles }; } catch (err) {}
     updateOpControls(toggles);
 }
 
@@ -82,9 +82,9 @@ function resetOpControls(): void {
     opControls.handlers = null;
     closeDownloadModal();
     const nextState = resetToggles();
-    try { window.__tritonVizOpState = { ...nextState.toggles }; } catch (err) {}
-    if (window.__tritonVizCodeHide) {
-        window.__tritonVizCodeHide();
+    try { window.__tileLensOpState = { ...nextState.toggles }; } catch (err) {}
+    if (window.__tileLensCodeHide) {
+        window.__tileLensCodeHide();
     }
     updateOpControls(nextState.toggles);
 }
diff --git a/frontend/templates/index.html b/frontend/templates/index.html
index b120159c4..f2e2144a7 100644
--- a/frontend/templates/index.html
+++ b/frontend/templates/index.html
@@ -11,7 +11,7 @@