Spaces:
Sleeping
Sleeping
Upload run_cloud_training.py with huggingface_hub
Browse files- run_cloud_training.py +44 -13
run_cloud_training.py
CHANGED
|
@@ -24,12 +24,54 @@ from unsloth import FastLanguageModel
|
|
| 24 |
# Disable flash attention globally
|
| 25 |
os.environ["TRANSFORMERS_NO_FLASH_ATTENTION"] = "1"
|
| 26 |
|
| 27 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
try:
|
| 29 |
import subprocess
|
| 30 |
import sys
|
| 31 |
|
| 32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
logger.info("Attempting to install flash-attention...")
|
| 34 |
|
| 35 |
# Try multiple installation approaches for flash-attention
|
|
@@ -74,17 +116,6 @@ except ImportError:
|
|
| 74 |
TENSORBOARD_AVAILABLE = False
|
| 75 |
print("Tensorboard not available. Will skip tensorboard logging.")
|
| 76 |
|
| 77 |
-
# Configure logging
|
| 78 |
-
logging.basicConfig(
|
| 79 |
-
level=logging.INFO,
|
| 80 |
-
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
|
| 81 |
-
handlers=[
|
| 82 |
-
logging.StreamHandler(),
|
| 83 |
-
logging.FileHandler("training.log")
|
| 84 |
-
]
|
| 85 |
-
)
|
| 86 |
-
logger = logging.getLogger(__name__)
|
| 87 |
-
|
| 88 |
# Default dataset path - use the correct path with username
|
| 89 |
DEFAULT_DATASET = "George-API/phi4-cognitive-dataset"
|
| 90 |
|
|
|
|
| 24 |
# Disable flash attention globally
|
| 25 |
os.environ["TRANSFORMERS_NO_FLASH_ATTENTION"] = "1"
|
| 26 |
|
| 27 |
+
# Configure logging first
|
| 28 |
+
logging.basicConfig(
|
| 29 |
+
level=logging.INFO,
|
| 30 |
+
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
|
| 31 |
+
handlers=[
|
| 32 |
+
logging.StreamHandler(),
|
| 33 |
+
logging.FileHandler("training.log")
|
| 34 |
+
]
|
| 35 |
+
)
|
| 36 |
+
logger = logging.getLogger(__name__)
|
| 37 |
+
|
| 38 |
+
# Make sure torch is installed and available before proceeding
|
| 39 |
+
try:
|
| 40 |
+
logger.info("Importing torch...")
|
| 41 |
+
import torch
|
| 42 |
+
logger.info(f"PyTorch version: {torch.__version__}")
|
| 43 |
+
logger.info(f"CUDA available: {torch.cuda.is_available()}")
|
| 44 |
+
if torch.cuda.is_available():
|
| 45 |
+
logger.info(f"CUDA version: {torch.version.cuda}")
|
| 46 |
+
logger.info(f"GPU: {torch.cuda.get_device_name(0)}")
|
| 47 |
+
except ImportError:
|
| 48 |
+
logger.error("PyTorch not found. Installing torch first...")
|
| 49 |
+
try:
|
| 50 |
+
import subprocess
|
| 51 |
+
import sys
|
| 52 |
+
subprocess.check_call([sys.executable, "-m", "pip", "install", "torch"])
|
| 53 |
+
logger.info("PyTorch installed successfully. Importing...")
|
| 54 |
+
import torch
|
| 55 |
+
logger.info(f"PyTorch version: {torch.__version__}")
|
| 56 |
+
except Exception as e:
|
| 57 |
+
logger.error(f"Failed to install PyTorch: {e}")
|
| 58 |
+
logger.error("Cannot proceed without PyTorch. Exiting.")
|
| 59 |
+
raise
|
| 60 |
+
|
| 61 |
+
# Now try to install flash-attention (for systems that support it)
|
| 62 |
try:
|
| 63 |
import subprocess
|
| 64 |
import sys
|
| 65 |
|
| 66 |
+
# Make sure torch is installed before attempting flash-attn
|
| 67 |
+
try:
|
| 68 |
+
logger.info("Ensuring PyTorch is installed before flash-attention...")
|
| 69 |
+
subprocess.check_call([sys.executable, "-m", "pip", "install", "torch", "--quiet"])
|
| 70 |
+
logger.info("PyTorch installation verified")
|
| 71 |
+
except Exception as torch_error:
|
| 72 |
+
logger.warning(f"PyTorch installation check failed: {torch_error}")
|
| 73 |
+
logger.info("Will continue with flash-attention installation anyway")
|
| 74 |
+
|
| 75 |
logger.info("Attempting to install flash-attention...")
|
| 76 |
|
| 77 |
# Try multiple installation approaches for flash-attention
|
|
|
|
| 116 |
TENSORBOARD_AVAILABLE = False
|
| 117 |
print("Tensorboard not available. Will skip tensorboard logging.")
|
| 118 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
# Default dataset path - use the correct path with username
|
| 120 |
DEFAULT_DATASET = "George-API/phi4-cognitive-dataset"
|
| 121 |
|