|
33 | 33 |
|
34 | 34 | ## 🚀 Quick Start
|
35 | 35 |
|
36 |
| -### 1. Generate Traces |
| 36 | +### 1. Installation |
| 37 | + |
| 38 | +**Four options to install:** |
| 39 | +```bash |
| 40 | +# install nightly version |
| 41 | +pip install -U --pre tritonparse |
| 42 | +# install stable version |
| 43 | +pip install tritonparse |
| 44 | +# install from source |
| 45 | +git clone https://github.com/meta-pytorch/tritonparse.git |
| 46 | +cd tritonparse |
| 47 | +pip install -e . |
| 48 | +# pip install the latest version from github |
| 49 | +pip install git+https://github.com/meta-pytorch/tritonparse.git |
| 50 | +``` |
| 51 | + |
| 52 | +**Prerequisites:** Python ≥ 3.10, Triton ≥ 3.4.0, GPU required (NVIDIA/AMD) |
| 53 | + |
| 54 | +TritonParse relies on new features in Triton. If you're using nightly PyTorch, Triton is already included. Otherwise, install the latest Triton: |
| 55 | +```bash |
| 56 | +pip install triton |
| 57 | +``` |
| 58 | + |
| 59 | +### 2. Generate Traces |
37 | 60 |
|
38 | 61 | ```python
|
39 | 62 | import tritonparse.structured_logging
|
@@ -68,19 +91,19 @@ tritonparse.utils.unified_parse("./logs/", out="./parsed_output")
|
68 | 91 | ```
|
69 | 92 | </details>
|
70 | 93 |
|
71 |
| -### 2. Visualize Results |
| 94 | +### 3. Visualize Results |
72 | 95 |
|
73 | 96 | **Visit [https://meta-pytorch.org/tritonparse/](https://meta-pytorch.org/tritonparse/?json_url=https://meta-pytorch.org/tritonparse/dedicated_log_triton_trace_findhao__mapped.ndjson.gz)** and open your local trace files (.ndjson.gz format).
|
74 | 97 |
|
75 | 98 | > **🔒 Privacy Note**: Your trace files are processed entirely in your browser - nothing is uploaded to any server!
|
76 | 99 |
|
77 |
| -### 3. Generate Reproducers (Optional) |
| 100 | +### 4. Generate Reproducers (Optional) |
78 | 101 |
|
79 | 102 | Extract any kernel into a standalone, executable Python script for debugging or testing:
|
80 | 103 |
|
81 | 104 | ```bash
|
82 | 105 | # Generate reproducer from first launch event
|
83 |
| -tritonparse reproduce ./parsed_output/trace.ndjson.gz --line 2 --out-dir repro_output |
| 106 | +tritonparseoss reproduce ./parsed_output/trace.ndjson.gz --line 2 --out-dir repro_output |
84 | 107 |
|
85 | 108 | # Run the generated reproducer
|
86 | 109 | cd repro_output/<kernel_name>/
|
@@ -109,30 +132,6 @@ result = reproduce(
|
109 | 132 |
|
110 | 133 | </details>
|
111 | 134 |
|
112 |
| -## 🛠️ Installation |
113 |
| - |
114 |
| -**For basic usage (trace generation):** |
115 |
| -Four options: |
116 |
| -```bash |
117 |
| -# install nightly version |
118 |
| -pip install -U --pre tritonparse |
119 |
| -# install stable version |
120 |
| -pip install tritonparse |
121 |
| -# install from source |
122 |
| -git clone https://github.com/meta-pytorch/tritonparse.git |
123 |
| -cd tritonparse |
124 |
| -pip install -e . |
125 |
| -# pip install the latest version from github |
126 |
| -pip install git+https://github.com/meta-pytorch/tritonparse.git |
127 |
| -``` |
128 |
| - |
129 |
| -**Prerequisites:** Python ≥ 3.10, Triton ≥ 3.4.0, GPU required (NVIDIA/AMD) |
130 |
| - |
131 |
| -TritonParse relies on new features in Triton. Please install the latest version of Triton: |
132 |
| -```bash |
133 |
| -pip install triton |
134 |
| -``` |
135 |
| - |
136 | 135 | ## 📚 Complete Documentation
|
137 | 136 |
|
138 | 137 | | 📖 Guide | Description |
|
|
0 commit comments