Upload folder using huggingface_hub
Browse files- INSTALL.md +75 -0
- INSTALL_AR.md +75 -0
- INSTALL_backup.md +76 -0
- MY_WHEELI_NOTES.md +39 -0
- MY_WHEEli_NOTES.md +39 -0
- PyTorch_ROCm_Build_Notes.md +53 -0
- README.md +7 -136
- README_AR.md +105 -0
- README_backup.md +108 -0
- build_torch.sh +30 -0
- built_with_fame_amd_torch.png +3 -0
- fame_git_reset_push.sh +29 -0
- fame_hf_upload.py +14 -0
- fame_torch_freeze.txt +31 -0
- install_torch.sh +24 -0
- test_torch.sh +16 -0
INSTALL.md
ADDED
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Fame PyTorch ROCm Kit – Installation Instructions 🚀
|
2 |
+
|
3 |
+
This file explains how to manually install the Fame-built PyTorch ROCm version.
|
4 |
+
|
5 |
+
---
|
6 |
+
|
7 |
+
## 🛠️ 1. Prepare Environment
|
8 |
+
|
9 |
+
Activate your Python environment (recommended: venv):
|
10 |
+
|
11 |
+
```bash
|
12 |
+
source ~/rocm_env/bin/activate
|
13 |
+
```
|
14 |
+
|
15 |
+
Install CMake if missing:
|
16 |
+
|
17 |
+
```bash
|
18 |
+
pip install cmake
|
19 |
+
```
|
20 |
+
|
21 |
+
Ensure HIP and ROCm paths are correctly set (`ROC_ENABLE_PRE_VEGA=1`, etc).
|
22 |
+
|
23 |
+
---
|
24 |
+
|
25 |
+
## 🛠️ 2. Build PyTorch
|
26 |
+
|
27 |
+
Navigate to the Fame Kit folder:
|
28 |
+
|
29 |
+
```bash
|
30 |
+
cd fame-pytorch-kit/
|
31 |
+
```
|
32 |
+
|
33 |
+
Run the build script:
|
34 |
+
|
35 |
+
```bash
|
36 |
+
bash build_torch.sh
|
37 |
+
```
|
38 |
+
|
39 |
+
---
|
40 |
+
|
41 |
+
## 📦 3. Install the Wheel
|
42 |
+
|
43 |
+
Navigate to the dist folder:
|
44 |
+
|
45 |
+
```bash
|
46 |
+
cd pytorch/dist/
|
47 |
+
```
|
48 |
+
|
49 |
+
Install the Wheel manually:
|
50 |
+
|
51 |
+
```bash
|
52 |
+
pip install torch-2.8.0a0+gitc402b3b-cp312-cp312-linux_x86_64.whl
|
53 |
+
|
54 |
+
```
|
55 |
+
|
56 |
+
---
|
57 |
+
|
58 |
+
## ✅ 4. Verify Installation
|
59 |
+
|
60 |
+
Test your installation:
|
61 |
+
|
62 |
+
```bash
|
63 |
+
python -c "import torch; print(torch.cuda.is_available())"
|
64 |
+
python -c "import torch; print(torch.cuda.get_device_name(0))"
|
65 |
+
python -c "import torch; print(torch.version.hip)"
|
66 |
+
```
|
67 |
+
|
68 |
+
You should see:
|
69 |
+
- `True` (GPU detected)
|
70 |
+
- Your GPU model name
|
71 |
+
- HIP version string
|
72 |
+
|
73 |
+
---
|
74 |
+
|
75 |
+
# Built with Fire. Built with Fame. 🛡️🔥
|
INSTALL_AR.md
ADDED
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Fame PyTorch ROCm Kit - تعليمات التثبيت 🚀
|
2 |
+
|
3 |
+
يشرح هذا الملف كيفية تثبيت إصدار Fame المُصمم من PyTorch ROCm يدويًا.
|
4 |
+
|
5 |
+
--
|
6 |
+
|
7 |
+
## 🛠️ 1. إعداد البيئة
|
8 |
+
|
9 |
+
فعّل بيئة بايثون (موصى به: venv):
|
10 |
+
|
11 |
+
```bash
|
12 |
+
source ~/rocm_env/bin/activate
|
13 |
+
```
|
14 |
+
|
15 |
+
ثبّت CMake في حال عدم توفره:
|
16 |
+
|
17 |
+
```bash
|
18 |
+
pip install cmake
|
19 |
+
```
|
20 |
+
|
21 |
+
تأكد من ضبط مسارات HIP وROCm بشكل صحيح (`ROC_ENABLE_PRE_VEGA=1`، إلخ).
|
22 |
+
|
23 |
+
---
|
24 |
+
|
25 |
+
## 🛠️ ٢. بناء PyTorch
|
26 |
+
|
27 |
+
انتقل إلى مجلد Fame Kit:
|
28 |
+
|
29 |
+
```bash
|
30 |
+
cd fame-pytorch-kit/
|
31 |
+
```
|
32 |
+
|
33 |
+
شغّل سكربت البناء:
|
34 |
+
|
35 |
+
```bash
|
36 |
+
bash build_torch.sh
|
37 |
+
```
|
38 |
+
|
39 |
+
---
|
40 |
+
|
41 |
+
## 📦 ٣. تثبيت Wheel
|
42 |
+
|
43 |
+
انتقل إلى مجلد dist:
|
44 |
+
|
45 |
+
```bash
|
46 |
+
cd pytorch/dist/
|
47 |
+
```
|
48 |
+
|
49 |
+
ثبّت Wheel يدويًا:
|
50 |
+
|
51 |
+
```bash
|
52 |
+
pip install torch-2.8.0a0+gitc402b3b-cp312-cp312-linux_x86_64.whl
|
53 |
+
|
54 |
+
```
|
55 |
+
|
56 |
+
---
|
57 |
+
|
58 |
+
## ✅ ٤. التحقق من التثبيت
|
59 |
+
|
60 |
+
اختبر التثبيت:
|
61 |
+
|
62 |
+
```bash
|
63 |
+
python -c "import torch; print(torch.cuda.is_available())"
|
64 |
+
python -c "import torch; print(torch.cuda.get_device_name(0))"
|
65 |
+
python -c "import torch; print(torch.version.hip)"
|
66 |
+
```
|
67 |
+
|
68 |
+
يجب أن يظهر لك:
|
69 |
+
- `True` (تم اكتشاف وحدة معالجة الرسومات)
|
70 |
+
- اسم طراز وحدة معالجة الرسومات لديك
|
71 |
+
- سلسلة إصدار HIP
|
72 |
+
|
73 |
+
---
|
74 |
+
|
75 |
+
# صُنع ببراعة. صُنع بشهرة. 🛡️🔥
|
INSTALL_backup.md
ADDED
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Fame PyTorch ROCm Kit – Installation Instructions 🚀
|
2 |
+
|
3 |
+
This file explains how to manually install the Fame-built PyTorch ROCm version.
|
4 |
+
|
5 |
+
---
|
6 |
+
|
7 |
+
## 🛠️ 1. Prepare Environment
|
8 |
+
|
9 |
+
Activate your Python environment (recommended: venv):
|
10 |
+
|
11 |
+
```bash
|
12 |
+
source ~/rocm_env/bin/activate
|
13 |
+
```
|
14 |
+
|
15 |
+
Install CMake if missing:
|
16 |
+
|
17 |
+
```bash
|
18 |
+
pip install cmake
|
19 |
+
```
|
20 |
+
|
21 |
+
Ensure HIP and ROCm paths are correctly set (`ROC_ENABLE_PRE_VEGA=1`, etc).
|
22 |
+
|
23 |
+
---
|
24 |
+
|
25 |
+
## 🛠️ 2. Build PyTorch
|
26 |
+
|
27 |
+
Navigate to the Fame Kit folder:
|
28 |
+
|
29 |
+
```bash
|
30 |
+
cd fame-pytorch-kit/
|
31 |
+
```
|
32 |
+
|
33 |
+
Run the build script:
|
34 |
+
|
35 |
+
```bash
|
36 |
+
bash build_torch.sh
|
37 |
+
```
|
38 |
+
|
39 |
+
---
|
40 |
+
|
41 |
+
## 📦 3. Install the Wheel
|
42 |
+
|
43 |
+
Navigate to the dist folder:
|
44 |
+
|
45 |
+
```bash
|
46 |
+
cd pytorch/dist/
|
47 |
+
```
|
48 |
+
|
49 |
+
Install the Wheel manually:
|
50 |
+
|
51 |
+
```bash
|
52 |
+
pip install torch-2.8.0a0+gitxxxxxx-cp312-cp312-linux_x86_64.whl
|
53 |
+
```
|
54 |
+
|
55 |
+
_(Replace `gitxxxxxx` with your actual build identifier.)_
|
56 |
+
|
57 |
+
---
|
58 |
+
|
59 |
+
## ✅ 4. Verify Installation
|
60 |
+
|
61 |
+
Test your installation:
|
62 |
+
|
63 |
+
```bash
|
64 |
+
python -c "import torch; print(torch.cuda.is_available())"
|
65 |
+
python -c "import torch; print(torch.cuda.get_device_name(0))"
|
66 |
+
python -c "import torch; print(torch.version.hip)"
|
67 |
+
```
|
68 |
+
|
69 |
+
You should see:
|
70 |
+
- `True` (GPU detected)
|
71 |
+
- Your GPU model name
|
72 |
+
- HIP version string
|
73 |
+
|
74 |
+
---
|
75 |
+
|
76 |
+
# Built with Fire. Built with Fame. 🛡️🔥
|
MY_WHEELI_NOTES.md
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# 🛡️ MY_WHEEli_NOTES.md
|
2 |
+
|
3 |
+
## 📅 Build-Info
|
4 |
+
- Build Datum: 2025-04-28
|
5 |
+
- Build Maschine: horscht-X670E-AORUS-XTREME
|
6 |
+
- ROCm Version: 6.2 Nightly
|
7 |
+
- Python Version: 3.12.7
|
8 |
+
- CMake Version: 4.0.0
|
9 |
+
- HIP Compiler: hipcc (clang-17 alias)
|
10 |
+
|
11 |
+
## ⚙️ Build Schritte
|
12 |
+
1. ROCm Environment gesetzt (ROCM_PATH, HIP_PATH, etc.)
|
13 |
+
2. git clean -xfd
|
14 |
+
3. Externer Build-Ordner `~/build-fame-pytorch/pytorch-build`
|
15 |
+
4. cmake .. -DCMAKE_CXX_COMPILER=hipcc
|
16 |
+
5. make -j$(nproc)
|
17 |
+
6. python3 setup.py bdist_wheel
|
18 |
+
7. pip install dist/*.whl
|
19 |
+
|
20 |
+
## 📦 Wheel Datei
|
21 |
+
- Name: torch-2.8.0a0+gitc402b3b-py3.12-linux-x86_64.whl
|
22 |
+
- Speicherort: ~/build-fame-pytorch/pytorch/dist/
|
23 |
+
|
24 |
+
## 🔥 Wichtig:
|
25 |
+
- Nicht `git clean -xfd` ausführen ohne vorher `/dist/` zu sichern!
|
26 |
+
- Vor jedem Build `build_torch.sh` neu starten
|
27 |
+
- Immer ROCm-Variablen setzen
|
28 |
+
|
29 |
+
## 🎯 Direkter Nutzen
|
30 |
+
- Volle PyTorch GPU Power auf AMD ROCm mit RX 7900 XTX
|
31 |
+
- Eigene Kontrolle über Build-Optionen
|
32 |
+
- Unabhängig von NVIDIA CUDA-Ökosystem
|
33 |
+
|
34 |
+
## ✨ Indirekter Nutzen
|
35 |
+
- ROCm-Umgebungen verstehen
|
36 |
+
- Fehlerquellen systematisch lösen
|
37 |
+
- Reverse Engineering & eigenes Research-Fundament
|
38 |
+
|
39 |
+
# 🏆 Projekt-Name: Fame PyTorch ROCm Build
|
MY_WHEEli_NOTES.md
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# 🛡️ MY_WHEEli_NOTES.md
|
2 |
+
|
3 |
+
## 📅 Build-Info
|
4 |
+
- Build Datum: 2025-04-28
|
5 |
+
- Build Maschine: horscht-X670E-AORUS-XTREME
|
6 |
+
- ROCm Version: 6.2 Nightly
|
7 |
+
- Python Version: 3.12.7
|
8 |
+
- CMake Version: 4.0.0
|
9 |
+
- HIP Compiler: hipcc (clang-17 alias)
|
10 |
+
|
11 |
+
## ⚙️ Build Schritte
|
12 |
+
1. ROCm Environment gesetzt (ROCM_PATH, HIP_PATH, etc.)
|
13 |
+
2. git clean -xfd
|
14 |
+
3. Externer Build-Ordner `~/build-fame-pytorch/pytorch-build`
|
15 |
+
4. cmake .. -DCMAKE_CXX_COMPILER=hipcc
|
16 |
+
5. make -j$(nproc)
|
17 |
+
6. python3 setup.py bdist_wheel
|
18 |
+
7. pip install dist/*.whl
|
19 |
+
|
20 |
+
## 📦 Wheel Datei
|
21 |
+
- Name: torch-2.8.0a0+gitc402b3b-py3.12-linux-x86_64.whl
|
22 |
+
- Speicherort: ~/build-fame-pytorch/pytorch/dist/
|
23 |
+
|
24 |
+
## 🔥 Wichtig:
|
25 |
+
- Nicht `git clean -xfd` ausführen ohne vorher `/dist/` zu sichern!
|
26 |
+
- Vor jedem Build `build_torch.sh` neu starten
|
27 |
+
- Immer ROCm-Variablen setzen
|
28 |
+
|
29 |
+
## 🎯 Direkter Nutzen
|
30 |
+
- Volle PyTorch GPU Power auf AMD ROCm mit RX 7900 XTX
|
31 |
+
- Eigene Kontrolle über Build-Optionen
|
32 |
+
- Unabhängig von NVIDIA CUDA-Ökosystem
|
33 |
+
|
34 |
+
## ✨ Indirekter Nutzen
|
35 |
+
- ROCm-Umgebungen verstehen
|
36 |
+
- Fehlerquellen systematisch lösen
|
37 |
+
- Reverse Engineering & eigenes Research-Fundament
|
38 |
+
|
39 |
+
# 🏆 Projekt-Name: Fame PyTorch ROCm Build
|
PyTorch_ROCm_Build_Notes.md
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
🔥 Fame PyTorch ROCm Build Notes
|
2 |
+
Stand: April 2025
|
3 |
+
|
4 |
+
📦 Umgebung
|
5 |
+
|
6 |
+
Komponente Version
|
7 |
+
Betriebssystem Ubuntu 24.10 (6.11 Kernel)
|
8 |
+
Python 3.12.7 (venv: rocm_env)
|
9 |
+
Compiler gcc/g++ 14.2.0
|
10 |
+
CMake 4.0.0
|
11 |
+
Ninja installed (lokal via pip)
|
12 |
+
ROCm 6.2 Nightly (gfx1100 support)
|
13 |
+
micro als Editor (micro)
|
14 |
+
🔥 Relevante Einstellungen
|
15 |
+
|
16 |
+
Variable Wert
|
17 |
+
USE_ROCM 1 (aktiviert)
|
18 |
+
USE_CUDA 0 (deaktiviert)
|
19 |
+
CMAKE_ARGS -DROCM_ARCH=gfx1100
|
20 |
+
USE_NUMPY 1 (aktiviert)
|
21 |
+
USE_NCCL 0 (abgeschaltet, no clone)
|
22 |
+
USE_MKL 0 (no Intel MKL)
|
23 |
+
USE_BLAS 0 (Fallback Eigen)
|
24 |
+
USE_OPENMP 1 (aktiv)
|
25 |
+
USE_MPI 0 (kein MPI)
|
26 |
+
USE_MAGMA 0 (kein MAGMA)
|
27 |
+
🔨 Submodule Handling
|
28 |
+
Submodules nicht vollständig geclont (manuelle Rescue: SSH Clone von wichtigsten Modulen).
|
29 |
+
|
30 |
+
Einige Submodules wurden bewusst ignoriert (DCGM, civetweb, manche TensorPipe-Tiefmodule).
|
31 |
+
|
32 |
+
Keine Abhängigkeit mehr von rekursivem HTTPS/SSH Git beim Build.
|
33 |
+
|
34 |
+
PeachPy Submodule (NNPACK) braucht six (manuell installiert: pip install six).
|
35 |
+
|
36 |
+
🗃️ Probleme, die gefixt wurden
|
37 |
+
|
38 |
+
Problem Lösung
|
39 |
+
Git Submodule SSH/HTTPS-Konflikte Global Patch auf git@github.com:
|
40 |
+
NCCL Zwangs-Clone checkout_nccl() Funktion disabled
|
41 |
+
PeachPy Dependency (six) fehlt pip install six
|
42 |
+
BLAS fallback auf Eigen automatisch von Build erkannt
|
43 |
+
📌 Tarballs / Snapshots
|
44 |
+
|
45 |
+
Backup Status
|
46 |
+
fame-pytorch-full-backup.tar.gz Erstellt
|
47 |
+
🏁 Status
|
48 |
+
python setup.py bdist_wheel läuft aktiv.
|
49 |
+
|
50 |
+
Erwartete Build-Dauer: 60–120 Minuten (abhängig von CPU/GPU).
|
51 |
+
|
52 |
+
Wheel wird erzeugt:
|
53 |
+
/home/horscht/build-fame-pytorch/pytorch/dist/torch-2.8.0a0+gitc402b3b-*.whl
|
README.md
CHANGED
@@ -1,28 +1,7 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
language:
|
4 |
-
- en
|
5 |
-
- ar
|
6 |
-
tags:
|
7 |
-
- rocm
|
8 |
-
- amd
|
9 |
-
- gpu
|
10 |
-
- build-environment
|
11 |
-
- pytorch
|
12 |
-
- custom-build
|
13 |
-
- deep-learning
|
14 |
-
- machine-learning
|
15 |
-
- llm-enigneering
|
16 |
-
- arabic
|
17 |
-
- multilingual
|
18 |
-
- bilingual
|
19 |
-
---
|
20 |
-
# Fame PyTorch ROCm Kit 🔥🛡️
|
21 |
-
|
22 |
-
Built with Courage. Built with Fire. Built with Fame.
|
23 |
-
|
24 |
|
25 |
-
|
26 |
|
27 |
---
|
28 |
|
@@ -46,8 +25,8 @@ You will find everything needed to build, install, and test PyTorch from source,
|
|
46 |
| `MY_WHEELI_NOTES.md` | Notes related to the Wheel build |
|
47 |
| `PyTorch_ROCm_Build_Notes.md` | Extended ROCm build notes |
|
48 |
| `README.md` | This document – the Fame Masterplan |
|
49 |
-
---
|
50 |
|
|
|
51 |
|
52 |
## 🛠️ Build Instructions
|
53 |
|
@@ -124,114 +103,6 @@ Fame Kit proudly built by [sbeierle](https://github.com/sbeierle) with 🛡️ C
|
|
124 |
---
|
125 |
|
126 |
# Built with Fire. Built with Fame. 🛡️🔥
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
# Fame PyTorch ROCm Kit 🔥
|
133 |
-
|
134 |
-
مُصمم بشجاعة. مُصمم ببراعة. مُصمم بشغف.
|
135 |
-
|
136 |
-
---
|
137 |
-
|
138 |
-
## 🚀 حول
|
139 |
-
|
140 |
-
يوفر هذا المشروع مجموعة بناء كاملة ومنظمة لتجميع نسخة PyTorch ROCm عاملة على وحدات معالجة الرسومات AMD - دون عناء استخدام وحدات ROCm أو HIP أو OpenMP أو Git الفرعية.
|
141 |
-
|
142 |
-
ستجد كل ما تحتاجه لبناء PyTorch وتثبيته واختباره من المصدر، مع إنشاء حزمة `.whl` صالحة لجهازك المحلي.
|
143 |
-
|
144 |
-
---
|
145 |
-
|
146 |
-
## 📋 هيكل المشروع
|
147 |
-
|
148 |
-
| ملف/مجلد | الغرض |
|
149 |
-
|:-----------|:--------|
|
150 |
-
| `build_torch.sh` | سكربت لبناء PyTorch من المصدر |
|
151 |
-
| `install_torch.sh` | سكربت تثبيت ملف Wheel المُولّد |
|
152 |
-
| `test_torch.sh` | سكربت لاختبار استيراد PyTorch وحالة ROCm |
|
153 |
-
| `fame_torch_freeze.txt` | تجميد البيئة (إصدارات الحزمة) |
|
154 |
-
| `pytorch-wheel/` | (اختياري) المجلد الذي يحتوي على ملف `.whl` |
|
155 |
-
| `MY_WHEELI_NOTES.md` | ملاحظات متعلقة ببناء Wheel |
|
156 |
-
| `PyTorch_ROCm_Build_Notes.md` | ملاحظات بناء ROCm الموسعة |
|
157 |
-
| `README.md` | هذه الوثيقة - الخطة الرئيسية لـ Fame |
|
158 |
-
|
159 |
-
---
|
160 |
-
|
161 |
-
## 🛠️ تعليمات البناء
|
162 |
-
|
163 |
-
1. **تفعيل بيئة ROCm**
|
164 |
-
|
165 |
-
```bash
|
166 |
-
source ~/rocm_env/bin/activate
|
167 |
-
```
|
168 |
-
|
169 |
-
2. **انتقل إلى مجلد المشروع**
|
170 |
-
|
171 |
-
```bash
|
172 |
-
cd fame-pytorch-kit/
|
173 |
-
```
|
174 |
-
|
175 |
-
3. **بناء PyTorch**
|
176 |
-
|
177 |
-
```bash
|
178 |
-
bash build_torch.sh
|
179 |
-
```
|
180 |
-
|
181 |
-
سيكون ملف `.whl` المُنشأ موجودًا في `pytorch/dist/`.
|
182 |
-
|
183 |
-
---
|
184 |
-
|
185 |
-
## 📦 تعليمات التثبيت
|
186 |
-
|
187 |
-
### تثبيت سريع
|
188 |
-
|
189 |
-
```bash
|
190 |
-
bash install_torch.sh
|
191 |
-
```
|
192 |
-
|
193 |
-
### تثبيت يدوي
|
194 |
-
|
195 |
-
```bash
|
196 |
-
cd pytorch/dist/
|
197 |
-
pip install torch-2.8.0a0+gitc402b3b-cp312-cp312-linux_x86_64.whl
|
198 |
-
|
199 |
-
```
|
200 |
-
|
201 |
-
تأكد من تفعيل بيئة بايثون (venv) لديك!
|
202 |
-
|
203 |
-
---
|
204 |
-
|
205 |
-
## ✅ تعليمات الاختبار
|
206 |
-
|
207 |
-
```bash
|
208 |
-
bash test_torch.sh
|
209 |
-
```
|
210 |
-
|
211 |
-
أو يدويًا:
|
212 |
-
|
213 |
-
```bash
|
214 |
-
python -c "import torch; print(torch.cuda.is_available())"
|
215 |
-
python -c "import torch; print(torch.cuda.get_device_name(0))"
|
216 |
-
python -c "import torch; print(torch.version.hip)"
|
217 |
-
```
|
218 |
-
|
219 |
-
المتوقع: `True`، اسم وحدة معالجة الرسومات، سلسلة إصدار HIP.
|
220 |
-
|
221 |
-
---
|
222 |
-
|
223 |
-
## 📜 الترخيص
|
224 |
-
|
225 |
-
هذا المشروع مرخص بموجب ترخيص معهد ماساتشوستس للتكنولوجيا (MIT).
|
226 |
-
|
227 |
-
---
|
228 |
-
|
229 |
-
## 🤝 شكر وتقدير
|
230 |
-
|
231 |
-
مجموعة الشهرة مصممة بفخر من قِبل [sbeierle](https://github.com/sbeierle) بشجاعة 🛡️، وحماس 🔥، وشهرة 🚀.
|
232 |
-
|
233 |
-
---
|
234 |
-
|
235 |
-
# صُنعت بالنار. صُنعت بالشهرة. 🛡️🔥
|
236 |
-
# Fame PyTorch ROCm Kit 🔥
|
237 |
-
|
|
|
1 |
+
<<<<<<< HEAD
|
2 |
+
# Fame PyTorch ROCm Kit 🔥
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
+
Built with Courage. Built with Fire. Built with FAME.
|
5 |
|
6 |
---
|
7 |
|
|
|
25 |
| `MY_WHEELI_NOTES.md` | Notes related to the Wheel build |
|
26 |
| `PyTorch_ROCm_Build_Notes.md` | Extended ROCm build notes |
|
27 |
| `README.md` | This document – the Fame Masterplan |
|
|
|
28 |
|
29 |
+
---
|
30 |
|
31 |
## 🛠️ Build Instructions
|
32 |
|
|
|
103 |
---
|
104 |
|
105 |
# Built with Fire. Built with Fame. 🛡️🔥
|
106 |
+
=======
|
107 |
+
# fame-pytorch-kit
|
108 |
+
>>>>>>> cbd4317 (first commit)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
README_AR.md
ADDED
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Fame PyTorch ROCm Kit 🔥
|
2 |
+
|
3 |
+
مُصمم بشجاعة. مُصمم ببراعة. مُصمم بشغف.
|
4 |
+
|
5 |
+
---
|
6 |
+
|
7 |
+
## 🚀 حول
|
8 |
+
|
9 |
+
يوفر هذا المشروع مجموعة بناء كاملة ومنظمة لتجميع نسخة PyTorch ROCm عاملة على وحدات معالجة الرسومات AMD - دون عناء استخدام وحدات ROCm أو HIP أو OpenMP أو Git الفرعية.
|
10 |
+
|
11 |
+
ستجد كل ما تحتاجه لبناء PyTorch وتثبيته واختباره من المصدر، مع إنشاء حزمة `.whl` صالحة لجهازك المحلي.
|
12 |
+
|
13 |
+
---
|
14 |
+
|
15 |
+
## 📋 هيكل المشروع
|
16 |
+
|
17 |
+
| ملف/مجلد | الغرض |
|
18 |
+
|:-----------|:--------|
|
19 |
+
| `build_torch.sh` | سكربت لبناء PyTorch من المصدر |
|
20 |
+
| `install_torch.sh` | سكربت تثبيت ملف Wheel المُولّد |
|
21 |
+
| `test_torch.sh` | سكربت لاختبار استيراد PyTorch وحالة ROCm |
|
22 |
+
| `fame_torch_freeze.txt` | تجميد البيئة (إصدارات الحزمة) |
|
23 |
+
| `pytorch-wheel/` | (اختياري) المجلد الذي يحتوي على ملف `.whl` |
|
24 |
+
| `MY_WHEELI_NOTES.md` | ملاحظات متعلقة ببناء Wheel |
|
25 |
+
| `PyTorch_ROCm_Build_Notes.md` | ملاحظات بناء ROCm الموسعة |
|
26 |
+
| `README.md` | هذه الوثيقة - الخطة الرئيسية لـ Fame |
|
27 |
+
|
28 |
+
---
|
29 |
+
|
30 |
+
## 🛠️ تعليمات البناء
|
31 |
+
|
32 |
+
1. **تفعيل بيئة ROCm**
|
33 |
+
|
34 |
+
```bash
|
35 |
+
source ~/rocm_env/bin/activate
|
36 |
+
```
|
37 |
+
|
38 |
+
2. **انتقل إلى مجلد المشروع**
|
39 |
+
|
40 |
+
```bash
|
41 |
+
cd fame-pytorch-kit/
|
42 |
+
```
|
43 |
+
|
44 |
+
3. **بناء PyTorch**
|
45 |
+
|
46 |
+
```bash
|
47 |
+
bash build_torch.sh
|
48 |
+
```
|
49 |
+
|
50 |
+
سيكون ملف `.whl` المُنشأ موجودًا في `pytorch/dist/`.
|
51 |
+
|
52 |
+
---
|
53 |
+
|
54 |
+
## 📦 تعليمات التثبيت
|
55 |
+
|
56 |
+
### تثبيت سريع
|
57 |
+
|
58 |
+
```bash
|
59 |
+
bash install_torch.sh
|
60 |
+
```
|
61 |
+
|
62 |
+
### تثبيت يدوي
|
63 |
+
|
64 |
+
```bash
|
65 |
+
cd pytorch/dist/
|
66 |
+
pip install torch-2.8.0a0+gitc402b3b-cp312-cp312-linux_x86_64.whl
|
67 |
+
|
68 |
+
```
|
69 |
+
|
70 |
+
تأكد من تفعيل بيئة بايثون (venv) لديك!
|
71 |
+
|
72 |
+
---
|
73 |
+
|
74 |
+
## ✅ تعليمات الاختبار
|
75 |
+
|
76 |
+
```bash
|
77 |
+
bash test_torch.sh
|
78 |
+
```
|
79 |
+
|
80 |
+
أو يدويًا:
|
81 |
+
|
82 |
+
```bash
|
83 |
+
python -c "import torch; print(torch.cuda.is_available())"
|
84 |
+
python -c "import torch; print(torch.cuda.get_device_name(0))"
|
85 |
+
python -c "import torch; print(torch.version.hip)"
|
86 |
+
```
|
87 |
+
|
88 |
+
المتوقع: `True`، اسم وحدة معالجة الرسومات، سلسلة إصدار HIP.
|
89 |
+
|
90 |
+
---
|
91 |
+
|
92 |
+
## 📜 الترخيص
|
93 |
+
|
94 |
+
هذا المشروع مرخص بموجب ترخيص معهد ماساتشوستس للتكنولوجيا (MIT).
|
95 |
+
|
96 |
+
---
|
97 |
+
|
98 |
+
## 🤝 شكر وتقدير
|
99 |
+
|
100 |
+
مجموعة الشهرة مصممة بفخر من قِبل [sbeierle](https://github.com/sbeierle) بشجاعة 🛡️، وحماس 🔥، وشهرة 🚀.
|
101 |
+
|
102 |
+
---
|
103 |
+
|
104 |
+
# صُنعت بالنار. صُنعت بالشهرة. 🛡️🔥
|
105 |
+
# Fame PyTorch ROCm Kit 🔥
|
README_backup.md
ADDED
@@ -0,0 +1,108 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
<<<<<<< HEAD
|
2 |
+
# Fame PyTorch ROCm Kit 🔥
|
3 |
+
|
4 |
+
Built with Courage. Built with Fire. Built with FAME.
|
5 |
+
|
6 |
+
---
|
7 |
+
|
8 |
+
## 🚀 About
|
9 |
+
|
10 |
+
This project provides a complete and clean build kit for compiling a working PyTorch ROCm version on AMD GPUs – without the usual headaches involving ROCm, HIP, OpenMP, or Git submodules.
|
11 |
+
|
12 |
+
You will find everything needed to build, install, and test PyTorch from source, generating a valid `.whl` package for your local machine.
|
13 |
+
|
14 |
+
---
|
15 |
+
|
16 |
+
## 📋 Project Structure
|
17 |
+
|
18 |
+
| File/Folder | Purpose |
|
19 |
+
|:------------|:--------|
|
20 |
+
| `build_torch.sh` | Script to build PyTorch from source |
|
21 |
+
| `install_torch.sh` | Script to install the generated Wheel file |
|
22 |
+
| `test_torch.sh` | Script to test PyTorch import and ROCm status |
|
23 |
+
| `fame_torch_freeze.txt` | Environment freeze (package versions) |
|
24 |
+
| `pytorch-wheel/` | (Optional) Folder containing the `.whl` file |
|
25 |
+
| `MY_WHEELI_NOTES.md` | Notes related to the Wheel build |
|
26 |
+
| `PyTorch_ROCm_Build_Notes.md` | Extended ROCm build notes |
|
27 |
+
| `README.md` | This document – the Fame Masterplan |
|
28 |
+
|
29 |
+
---
|
30 |
+
|
31 |
+
## 🛠️ Build Instructions
|
32 |
+
|
33 |
+
1. **Activate your ROCm environment**
|
34 |
+
|
35 |
+
```bash
|
36 |
+
source ~/rocm_env/bin/activate
|
37 |
+
```
|
38 |
+
|
39 |
+
2. **Navigate to the project folder**
|
40 |
+
|
41 |
+
```bash
|
42 |
+
cd fame-pytorch-kit/
|
43 |
+
```
|
44 |
+
|
45 |
+
3. **Build PyTorch**
|
46 |
+
|
47 |
+
```bash
|
48 |
+
bash build_torch.sh
|
49 |
+
```
|
50 |
+
|
51 |
+
The generated `.whl` file will be located in `pytorch/dist/`.
|
52 |
+
|
53 |
+
---
|
54 |
+
|
55 |
+
## 📦 Install Instructions
|
56 |
+
|
57 |
+
### Quick Install
|
58 |
+
|
59 |
+
```bash
|
60 |
+
bash install_torch.sh
|
61 |
+
```
|
62 |
+
|
63 |
+
### Manual Install
|
64 |
+
|
65 |
+
```bash
|
66 |
+
cd pytorch/dist/
|
67 |
+
pip install torch-2.8.0a0+gitc402b3b-cp312-cp312-linux_x86_64.whl
|
68 |
+
|
69 |
+
```
|
70 |
+
|
71 |
+
Make sure your Python environment (venv) is activated!
|
72 |
+
|
73 |
+
---
|
74 |
+
|
75 |
+
## ✅ Test Instructions
|
76 |
+
|
77 |
+
```bash
|
78 |
+
bash test_torch.sh
|
79 |
+
```
|
80 |
+
|
81 |
+
Or manually:
|
82 |
+
|
83 |
+
```bash
|
84 |
+
python -c "import torch; print(torch.cuda.is_available())"
|
85 |
+
python -c "import torch; print(torch.cuda.get_device_name(0))"
|
86 |
+
python -c "import torch; print(torch.version.hip)"
|
87 |
+
```
|
88 |
+
|
89 |
+
Expected: `True`, GPU Name, HIP version string.
|
90 |
+
|
91 |
+
---
|
92 |
+
|
93 |
+
## 📜 License
|
94 |
+
|
95 |
+
This project is licensed under the MIT License.
|
96 |
+
|
97 |
+
---
|
98 |
+
|
99 |
+
## 🤝 Credits
|
100 |
+
|
101 |
+
Fame Kit proudly built by [sbeierle](https://github.com/sbeierle) with 🛡️ Courage, 🔥 Fire, and 🚀 Fame.
|
102 |
+
|
103 |
+
---
|
104 |
+
|
105 |
+
# Built with Fire. Built with Fame. 🛡️🔥
|
106 |
+
=======
|
107 |
+
# fame-pytorch-kit
|
108 |
+
>>>>>>> cbd4317 (first commit)
|
build_torch.sh
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
# === Fame Kit Build Script ===
|
3 |
+
# (c) by Horscht & Bro 2025
|
4 |
+
|
5 |
+
set -e
|
6 |
+
|
7 |
+
echo "🚀 Setting up ROCm environment..."
|
8 |
+
export ROCM_PATH=/opt/rocm
|
9 |
+
export HIP_PATH=/opt/rocm
|
10 |
+
export PATH=$ROCM_PATH/bin:$ROCM_PATH/llvm/bin:$PATH
|
11 |
+
export LD_LIBRARY_PATH=$ROCM_PATH/lib:$ROCM_PATH/lib64:$LD_LIBRARY_PATH
|
12 |
+
export CMAKE_PREFIX_PATH=$ROCM_PATH
|
13 |
+
|
14 |
+
echo "🧹 Cleaning previous build..."
|
15 |
+
cd ~/build-fame-pytorch/pytorch
|
16 |
+
git clean -xfd
|
17 |
+
|
18 |
+
echo "📂 Preparing build directory..."
|
19 |
+
cd ~/build-fame-pytorch
|
20 |
+
rm -rf pytorch-build
|
21 |
+
mkdir pytorch-build
|
22 |
+
cd pytorch-build
|
23 |
+
|
24 |
+
echo "⚙️ Configuring CMake..."
|
25 |
+
cmake ../pytorch -DCMAKE_CXX_COMPILER=hipcc -DCMAKE_CXX_FLAGS="--rocm-path=/opt/rocm"
|
26 |
+
|
27 |
+
echo "🔨 Building PyTorch (this will take ~90 min)..."
|
28 |
+
make -j$(nproc)
|
29 |
+
|
30 |
+
echo "✅ Build completed successfully!"
|
built_with_fame_amd_torch.png
ADDED
![]() |
Git LFS Details
|
fame_git_reset_push.sh
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
|
3 |
+
# Fame Git Full Reset & Force Push
|
4 |
+
|
5 |
+
echo "🔧 Entferne alten Remote-Status..."
|
6 |
+
git remote remove origin
|
7 |
+
|
8 |
+
echo "🔧 Setze korrekte Remote URL..."
|
9 |
+
git remote add origin git@github.com:sbeierle/fame-pytorch-kit.git
|
10 |
+
|
11 |
+
echo "🔧 Holen und resetten..."
|
12 |
+
git fetch origin
|
13 |
+
git reset --hard origin/main
|
14 |
+
|
15 |
+
echo "🔧 Vorbereitung abgeschlossen. Force Push bereit."
|
16 |
+
|
17 |
+
# Jetzt neue Änderungen hinzufügen
|
18 |
+
echo "🔧 Stage alle lokalen Dateien..."
|
19 |
+
git add .
|
20 |
+
|
21 |
+
# Commit neu setzen
|
22 |
+
echo "🔧 Commit neue Version..."
|
23 |
+
git commit -m '🔥 Fame Full Reset Commit'
|
24 |
+
|
25 |
+
# Force Push durchführen
|
26 |
+
echo "🚀 Fame Force Push nach GitHub..."
|
27 |
+
git push -u origin main --force
|
28 |
+
|
29 |
+
echo "✅ Fame Git Reset abgeschlossen."
|
fame_hf_upload.py
ADDED
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# fame_hf_upload.py
|
2 |
+
from huggingface_hub import HfApi
|
3 |
+
|
4 |
+
api = HfApi()
|
5 |
+
|
6 |
+
api.upload_folder(
|
7 |
+
folder_path=".",
|
8 |
+
repo_id="sbeierle/fame-pytorch-kit",
|
9 |
+
repo_type="model",
|
10 |
+
path_in_repo=".",
|
11 |
+
ignore_patterns=["venv/*", ".git/*", "*.pyc", "__pycache__/*"]
|
12 |
+
)
|
13 |
+
|
14 |
+
print("✅ Fame Kit erfolgreich auf Huggingface hochgeladen!")
|
fame_torch_freeze.txt
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
accelerate==1.6.0
|
2 |
+
certifi==2025.4.26
|
3 |
+
charset-normalizer==3.4.1
|
4 |
+
filelock==3.13.1
|
5 |
+
fsspec==2024.6.1
|
6 |
+
huggingface-hub==0.30.2
|
7 |
+
idna==3.10
|
8 |
+
Jinja2==3.1.4
|
9 |
+
MarkupSafe==2.1.5
|
10 |
+
mpmath==1.3.0
|
11 |
+
networkx==3.3
|
12 |
+
numpy==1.26.4
|
13 |
+
packaging==25.0
|
14 |
+
pillow==11.0.0
|
15 |
+
psutil==7.0.0
|
16 |
+
pytorch-triton-rocm==3.2.0
|
17 |
+
PyYAML==6.0.2
|
18 |
+
regex==2024.11.6
|
19 |
+
requests==2.32.3
|
20 |
+
safetensors==0.5.3
|
21 |
+
setuptools==80.0.0
|
22 |
+
six==1.17.0
|
23 |
+
sympy==1.14.0
|
24 |
+
tokenizers==0.21.1
|
25 |
+
torch==2.6.0+rocm6.1
|
26 |
+
torchvision==0.21.0+rocm6.1
|
27 |
+
tqdm==4.67.1
|
28 |
+
transformers==4.51.3
|
29 |
+
typing_extensions==4.12.2
|
30 |
+
urllib3==2.4.0
|
31 |
+
wheel==0.45.1
|
install_torch.sh
ADDED
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
# === Fame Kit Install Script ===
|
3 |
+
|
4 |
+
set -e
|
5 |
+
|
6 |
+
echo "🚀 Setting up ROCm environment (again, safe side)..."
|
7 |
+
export ROCM_PATH=/opt/rocm
|
8 |
+
export HIP_PATH=/opt/rocm
|
9 |
+
export PATH=$ROCM_PATH/bin:$ROCM_PATH/llvm/bin:$PATH
|
10 |
+
export LD_LIBRARY_PATH=$ROCM_PATH/lib:$ROCM_PATH/lib64:$LD_LIBRARY_PATH
|
11 |
+
export CMAKE_PREFIX_PATH=$ROCM_PATH
|
12 |
+
|
13 |
+
echo "📦 Building PyTorch Wheel..."
|
14 |
+
cd ~/build-fame-pytorch/pytorch
|
15 |
+
export USE_ROCM=1
|
16 |
+
export USE_CUDA=0
|
17 |
+
export CMAKE_ARGS="-DROCM_ARCH=gfx1100"
|
18 |
+
|
19 |
+
python3 setup.py bdist_wheel
|
20 |
+
|
21 |
+
echo "📂 Installing Wheel..."
|
22 |
+
pip install dist/*.whl
|
23 |
+
|
24 |
+
echo "✅ Installation successful!"
|
test_torch.sh
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
# === Fame Kit Test Script ===
|
3 |
+
|
4 |
+
set -e
|
5 |
+
|
6 |
+
echo "🧪 Importing torch and checking ROCm support..."
|
7 |
+
|
8 |
+
python3 -c "
|
9 |
+
import torch
|
10 |
+
print('Torch Version:', torch.__version__)
|
11 |
+
print('CUDA available:', torch.cuda.is_available())
|
12 |
+
print('HIP version (ROCm):', getattr(torch.version, 'hip', 'N/A'))
|
13 |
+
print('Built with ROCm:', torch.version.hip is not None)
|
14 |
+
"
|
15 |
+
|
16 |
+
echo "✅ Test completed!"
|