owusu/DS-LLM-TEMPLATE-FINETUNING
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

instruct model setup

Browse Source
This commit is contained in:
Your Name
2025-08-28 17:57:59 +00:00
parent 77c563f358
commit d49b4ff2d5
55 changed files with 27760 additions and 326 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/instruct/.ipynb_checkpoints/train-checkpoint.py
+12 -11
View File
@@ -16,7 +16,7 @@ def run_training_with_config(config_path: str, dataset_path: str = None, **cli_o
if dataset_path:
print(f"Training dataset: {dataset_path}")
else:
print("Training dataset: Will use output_dir from YAML config")
print("Training dataset: Will use data_path from YAML config")
print()
# Build command
@@ -98,28 +98,28 @@ def create_training_example():
if not Path(config_path).exists():
print(f"Configuration file not found: {config_path}")
print(" Please run the data processor first to create the configuration")
print(" Please ensure the configuration file exists")
return False
print("Found required files!")
print(f" Config: {config_path}")
print(" Dataset: Will use output_dir from YAML config")
print(" Dataset: Will use data_path from YAML config")
print(" The training pipeline will automatically:")
print(" - Load conversation data from the output_dir specified in YAML")
print(" - Convert JSONL files to HuggingFace dataset format")
print(" - Load conversation data directly from JSONL file")
print(" - Convert to HuggingFace dataset format")
print(" - Apply ShareGPT standardization")
print(" - Format conversations with chat templates")
print(" - Train the model using SFTTrainer with response-only training")
print(" - Train the model using SFTTrainer")
print()
# Run training without explicit dataset path - will use YAML config
success = run_training_with_config(
config_path=config_path,
dataset_path=None, # Use output_dir from YAML config
dataset_path=None, # Use data_path from YAML config
epochs=1,
batch_size=1,
learning_rate=2e-4,
max_steps=30
max_steps=5 # Minimal steps for quick test
)
if success:
@@ -138,19 +138,20 @@ def create_quick_test():
if not Path(config_path).exists():
print(f"Configuration file not found: {config_path}")
print(" Please run the data processor first to create the configuration")
print(" Please ensure the configuration file exists")
return False
print("Running quick test with minimal training steps...")
print("This will load data directly from the JSONL file specified in config")
# Run training with very few steps for quick testing
success = run_training_with_config(
config_path=config_path,
dataset_path=None,
dataset_path=None, # Use data_path from YAML config
epochs=1,
batch_size=1,
learning_rate=2e-4,
max_steps=5 # Very few steps for quick test
max_steps=3 # Very few steps for quick test
)
if success:
scripts/instruct/train.py
+12 -11
View File
@@ -16,7 +16,7 @@ def run_training_with_config(config_path: str, dataset_path: str = None, **cli_o
if dataset_path:
print(f"Training dataset: {dataset_path}")
else:
print("Training dataset: Will use output_dir from YAML config")
print("Training dataset: Will use data_path from YAML config")
print()
# Build command
@@ -98,28 +98,28 @@ def create_training_example():
if not Path(config_path).exists():
print(f"Configuration file not found: {config_path}")
print(" Please run the data processor first to create the configuration")
print(" Please ensure the configuration file exists")
return False
print("Found required files!")
print(f" Config: {config_path}")
print(" Dataset: Will use output_dir from YAML config")
print(" Dataset: Will use data_path from YAML config")
print(" The training pipeline will automatically:")
print(" - Load conversation data from the output_dir specified in YAML")
print(" - Convert JSONL files to HuggingFace dataset format")
print(" - Load conversation data directly from JSONL file")
print(" - Convert to HuggingFace dataset format")
print(" - Apply ShareGPT standardization")
print(" - Format conversations with chat templates")
print(" - Train the model using SFTTrainer with response-only training")
print(" - Train the model using SFTTrainer")
print()
# Run training without explicit dataset path - will use YAML config
success = run_training_with_config(
config_path=config_path,
dataset_path=None, # Use output_dir from YAML config
dataset_path=None, # Use data_path from YAML config
epochs=1,
batch_size=1,
learning_rate=2e-4,
max_steps=30
max_steps=5 # Minimal steps for quick test
)
if success:
@@ -138,19 +138,20 @@ def create_quick_test():
if not Path(config_path).exists():
print(f"Configuration file not found: {config_path}")
print(" Please run the data processor first to create the configuration")
print(" Please ensure the configuration file exists")
return False
print("Running quick test with minimal training steps...")
print("This will load data directly from the JSONL file specified in config")
# Run training with very few steps for quick testing
success = run_training_with_config(
config_path=config_path,
dataset_path=None,
dataset_path=None, # Use data_path from YAML config
epochs=1,
batch_size=1,
learning_rate=2e-4,
max_steps=5 # Very few steps for quick test
max_steps=3 # Very few steps for quick test
)
if success: