API Reference - Keys & Caches

Core Functions

`kandc.init()`

Initialize a new experiment run with configuration and tracking.

kandc.init(
    project: Optional[str] = None,
    name: Optional[str] = None,
    config: Optional[Dict[str, Any]] = None,
    tags: Optional[List[str]] = None,
    notes: Optional[str] = None,
    dir: Optional[Union[str, Path]] = None,
    mode: Optional[str] = None,
    reinit: bool = False,
    open_browser: bool = True,
    capture_code: bool = True,
    code_exclude_patterns: Optional[List[str]] = None,
    **kwargs
) -> Run

Parameters:

project (str, optional): Project name for organization
name (str, optional): Run name (auto-generated if not provided)
config (dict, optional): Hyperparameters and configuration to track
tags (list, optional): Tags for filtering and organization
notes (str, optional): Optional description or notes
dir (str/Path, optional): Directory to save run data
mode (str, optional): Run mode - “online”, “offline”, or “disabled”
reinit (bool): Whether to reinitialize if already initialized
open_browser (bool): Whether to open dashboard in browser (online mode)
capture_code (bool): Whether to capture code snapshot
code_exclude_patterns (list, optional): Patterns to exclude from code capture

Returns: Run object representing the initialized experiment Examples:

# Basic initialization
run = kandc.init(
    project="optimize-transformer",
    name="test-run-1",
    config={"d_model": 64, "nhead": 4, "num_layers": 2, "seq_len": 16},
    tags=["transformer", "pytorch"],
    notes="Testing transformer architecture optimization"
)

# Disable code capture
run = kandc.init(
    project="optimize-transformer",
    capture_code=False  # No code snapshot
)

# Custom code exclusions
run = kandc.init(
    project="optimize-transformer",
    capture_code=True,
    code_exclude_patterns=[
        "*.pth",        # Model files
        "data/",        # Data directory
        "experiments/", # Experiment outputs
        "*.log"         # Log files
    ]
)

`kandc.log()`

Log metrics and data during experiment execution.

kandc.log(
    data: Dict[str, Any],
    x: Optional[float] = None
) -> None

Parameters:

data (dict): Dictionary of metrics to log
x (float, optional): Custom x-axis value (step, epoch, time, etc.)

Example:

# Basic logging
kandc.log({"loss": 0.25, "accuracy": 0.92})

# With custom x-axis
kandc.log({"val_loss": 0.18}, x=100)

# Complex data structures
kandc.log({
    "model_config": {"hidden_size": 128, "layers": 3},
    "dataset_stats": {"train_size": 50000, "val_size": 10000}
})

`kandc.finish()`

Finish the current run and ensure all data is saved and synced.

kandc.finish() -> None

Example:

try:
    kandc.init(project="my-project")
    # Your experiment code
    kandc.log({"metric": value})
finally:
    kandc.finish()  # Always finish, even on error

`kandc.get_current_run()`

Get the currently active run object.

kandc.get_current_run() -> Optional[Run]

Returns: Current Run object or None if not initialized Example:

run = kandc.get_current_run()
if run:
    print(f"Current run: {run.name}")
    print(f"Run ID: {run.id}")

`kandc.is_initialized()`

Check if kandc is currently initialized.

kandc.is_initialized() -> bool

Returns: True if initialized, False otherwise Example:

if not kandc.is_initialized():
    kandc.init(project="my-project")

PyTorch Profiling

`kandc.ProfilerWrapper`

Wrap any object with PyTorch profiler integration.

from kandc.annotators import ProfilerWrapper

wrapper = ProfilerWrapper(
    obj: Any,
    name: Optional[str] = None,
    activities: Optional[List[str]] = None,
    record_shapes: bool = True,
    profile_memory: bool = True,
    with_stack: bool = True
)

Parameters:

obj: The object to wrap and profile
name (str, optional): Name for the wrapped object in logs
activities (list, optional): Activities to profile [‘cpu’, ‘cuda’]. Defaults to both.
record_shapes (bool): Whether to record tensor shapes
profile_memory (bool): Whether to profile memory usage
with_stack (bool): Whether to record call stacks

Example:

model = MyModel()
profiled_model = ProfilerWrapper(
    model,
    name="MyModel", 
    activities=['cpu', 'cuda'],
    record_shapes=True,
    profile_memory=True
)

# All method calls are now profiled
result = profiled_model.forward(data)

`kandc.ProfilerDecorator`

Decorator for automatic PyTorch profiling of classes or functions.

from kandc.annotators import ProfilerDecorator

@ProfilerDecorator(
    name: Optional[str] = None,
    activities: Optional[List[str]] = None,
    record_shapes: bool = True,
    profile_memory: bool = True,
    with_stack: bool = True
)

Example:

@ProfilerDecorator(name="OptimizedModel", record_shapes=True)
class OptimizedModel:
    def predict(self, x):
        return x * 2
    
    def train(self, data):
        return self.predict(data)

`kandc.profile()`

Convenience function to wrap an object with PyTorch profiling.

kandc.profile(
    obj: Any,
    name: Optional[str] = None,
    activities: Optional[List[str]] = None,
    record_shapes: bool = True,
    profile_memory: bool = True,
    with_stack: bool = True
) -> ProfilerWrapper

Example:

model = MyModel()
profiled_model = kandc.profile(model, name="MyModel")

`kandc.profiler()`

Convenience function to create a PyTorch profiler decorator.

kandc.profiler(
    name: Optional[str] = None,
    activities: Optional[List[str]] = None,
    record_shapes: bool = True,
    profile_memory: bool = True,
    with_stack: bool = True
) -> ProfilerDecorator

Example:

@kandc.profiler(name="MyFunction", record_shapes=True)
def my_function(x):
    return expensive_computation(x)

Model Profiling

`kandc.capture_model_class()`

Decorator to automatically profile PyTorch model classes.

kandc.capture_model_class(
    model_name: Optional[str] = None,
    record_shapes: bool = True,
    profile_memory: bool = True,
    **profiler_kwargs: Any
) -> Callable

Parameters:

model_name (str, optional): Name for the model traces (defaults to class name)
record_shapes (bool): Whether to record tensor shapes
profile_memory (bool): Whether to profile memory usage
**profiler_kwargs: Additional PyTorch profiler arguments

Example:

@kandc.capture_model_class(
    model_name="SimpleTransformer",
    record_shapes=True,
    profile_memory=True
)
class SimpleTransformer(nn.Module):
    def __init__(self, input_dim=32, seq_len=16, d_model=64, nhead=4, num_layers=2, num_classes=10):
        super().__init__()
        self.input_dim = input_dim
        self.seq_len = seq_len
        self.d_model = d_model

        # Project input to d_model
        self.input_proj = nn.Linear(input_dim, d_model)

        # Positional encoding (learnable)
        self.pos_embedding = nn.Parameter(torch.zeros(1, seq_len, d_model))

        # Transformer encoder
        encoder_layer = nn.TransformerEncoderLayer(d_model=d_model, nhead=nhead, batch_first=True)
        self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)

        # Output head
        self.head = nn.Sequential(
            nn.LayerNorm(d_model),
            nn.Linear(d_model, num_classes)
        )

    def forward(self, x):
        # x: (batch, seq_len, input_dim)
        x = self.input_proj(x)  # (batch, seq_len, d_model)
        x = x + self.pos_embedding  # Add positional encoding
        x = self.transformer(x)  # (batch, seq_len, d_model)
        x = x.mean(dim=1)  # Pool over sequence
        x = self.head(x)   # (batch, num_classes)
        return x

`kandc.capture_model_instance()`

Wrap an existing model instance for profiling.

kandc.capture_model_instance(
    model_instance: Any,
    model_name: Optional[str] = None,
    record_shapes: bool = True,
    profile_memory: bool = True,
    **profiler_kwargs: Any
) -> Any

Parameters:

model_instance: The model instance to wrap
model_name (str, optional): Name for the model traces
record_shapes (bool): Whether to record tensor shapes
profile_memory (bool): Whether to profile memory usage
**profiler_kwargs: Additional profiler arguments

Returns: Wrapped model instance that profiles every forward pass Example:

# Existing transformer model
model = SimpleTransformer(input_dim=32, seq_len=16, d_model=64, nhead=4, num_layers=2, num_classes=10)

# Wrap for profiling
model = kandc.capture_model_instance(
    model,
    model_name="SimpleTransformer_Instance",
    record_shapes=True,
    profile_memory=True
)

# All forward passes now profiled
data = torch.randn(32, 16, 32)  # batch_size=32, seq_len=16, input_dim=32
output = model(data)

`kandc.capture_trace()`

Decorator to profile any function execution.

kandc.capture_trace(
    trace_name: Optional[str] = None,
    record_shapes: bool = False,
    profile_memory: bool = False,
    **profiler_kwargs: Any
) -> Callable

Parameters:

trace_name (str, optional): Name for the trace (defaults to function name)
record_shapes (bool): Whether to record tensor shapes
profile_memory (bool): Whether to profile memory usage
**profiler_kwargs: Additional profiler arguments

Example:

@kandc.capture_trace(
    trace_name="data_preprocessing",
    record_shapes=True
)
def preprocess_batch(images, labels):
    # Preprocessing code - automatically traced
    return processed_images, labels

`kandc.parse_model_trace()`

Parse and analyze a model trace file.

kandc.parse_model_trace(
    trace_file: str,
    model_name: str = "Unknown"
) -> Optional[Dict]

Parameters:

trace_file (str): Path to the trace file
model_name (str): Name of the model for analysis

Returns: Dictionary containing parsed trace analysis or None if parsing fails

Timing Functions

`kandc.timed()`

Decorator to time function execution.

kandc.timed(
    name: Optional[str] = None
) -> Callable

Parameters:

name (str, optional): Name for the timing record (defaults to function name)

Example:

@kandc.timed(name="random_wait")
def random_wait():
    print("⏳ Starting random wait...")
    time.sleep(random.random() * 2)
    print("✅ Random wait complete")
    return "processing_complete"

# Usage
result = random_wait()  # Timing automatically recorded

`kandc.timed_call()`

Time a function call without using a decorator.

kandc.timed_call(
    name: str,
    fn: Callable[..., Any],
    *args: Any,
    **kwargs: Any
) -> Any

Parameters:

name (str): Name for the timing record
fn (callable): Function to time
*args: Positional arguments for the function
**kwargs: Keyword arguments for the function

Returns: Result of the function call Example:

# Time an existing function
model = SimpleTransformer()
data = torch.randn(32, 16, 32)
result = kandc.timed_call("model_forward", model, data)

# Time a lambda or inline function
processed = kandc.timed_call("data_processing", lambda: time.sleep(0.1), )

API Client (Advanced)

`kandc.APIClient`

Low-level API client for direct backend communication.

client = kandc.APIClient(
    base_url: str = KANDC_BACKEND_URL,
    api_key: Optional[str] = None
)

Methods:

authenticate_with_browser(): Browser-based authentication
create_project(name, description, tags, metadata): Create new project
create_run(project_name, run_data): Create new run
log_metrics(run_id, metrics, step): Log metrics to run
create_artifact(run_id, artifact_data, file_path): Upload artifact
get_dashboard_url(project_id, run_id): Get dashboard URL

Authentication

`kandc.get_api_key()`

Get the current API key.

kandc.get_api_key() -> Optional[str]

Returns: Current API key or None if not authenticated

`kandc.ensure_authenticated()`

Ensure user is authenticated, prompting if necessary.

kandc.ensure_authenticated() -> APIClient

Returns: Authenticated API client

Run Object

The Run object represents an active experiment run.

Properties

run.id (str): Unique run identifier
run.name (str): Run name
run.project (str): Project name
run.dir (Path): Local directory for run data

Methods

run.log(data, x): Log metrics to this run
run.log_artifact(path, name): Log artifact file
run.finish(): Finish this run
run.get_dashboard_url(): Get dashboard URL for this run
run.open_dashboard(): Open dashboard in browser

Run Modes

Online Mode (Default)

kandc.init(project="my-project")  # mode="online" is default

Full cloud functionality
Real-time dashboard
Authentication required
Internet connection required

Offline Mode

kandc.init(project="my-project", mode="offline")

Local-only operation
No authentication required
No internet required
Data saved to local directory

Disabled Mode

kandc.init(project="my-project", mode="disabled")

No-op mode (zero overhead)
All kandc calls do nothing
Perfect for production deployment

Error Handling

`kandc.APIError`

Base exception for API-related errors.

`kandc.AuthenticationError`

Exception raised for authentication failures. Example:

try:
    kandc.init(project="my-project")
except kandc.AuthenticationError:
    print("Authentication failed")
except kandc.APIError as e:
    print(f"API error: {e}")

Code Snapshot Configuration

kandc automatically captures your source code for experiment reproducibility. You can control this behavior in several ways:

Disable Code Capture Completely

# Method 1: In kandc.init()
kandc.init(
    project="my-project",
    capture_code=False  # Disables all code capture
)

# Method 2: Environment variable
export KANDC_CAPTURE_CODE=false

Custom Exclude Patterns

# Exclude specific files and directories
kandc.init(
    project="my-project",
    code_exclude_patterns=[
        # Model files
        "*.pth", "*.safetensors", "*.bin", "*.ckpt",
        
        # Data directories
        "data/", "datasets/", "cache/",
        
        # Experiment outputs
        "outputs/", "results/", "logs/",
        
        # Temporary files
        "*.tmp", "*.temp", "temp_*",
        
        # Large files
        "*.csv", "*.parquet", "*.h5",
        
        # IDE files
        ".vscode/", ".idea/", "*.swp"
    ]
)

Respect .gitignore

kandc automatically respects your .gitignore file. Add patterns there to exclude them from code capture:

# .gitignore
*.pth
*.safetensors
data/
experiments/
*.log

What Gets Captured

Source Files:

Python: .py
JavaScript/TypeScript: .js, .ts, .jsx, .tsx
Other languages: .java, .cpp, .c, .go, .rs, .rb
Scripts: .sh, .bash, .zsh, .ps1, .bat
Config: .yaml, .yml, .json, .toml, .ini
Documentation: .md, .rst, .txt

Project Files:

requirements.txt, pyproject.toml
package.json, Dockerfile
.gitignore, .env.example

Environment Variables for Code Capture

# Disable code capture globally
export KANDC_CAPTURE_CODE=false

# Set default exclude patterns
export KANDC_CODE_EXCLUDE="*.pth,data/,experiments/"

# Set maximum file size (in bytes)
export KANDC_MAX_FILE_SIZE=1048576  # 1MB

Environment Variables

Configure kandc behavior via environment variables: Core Configuration:

KANDC_BACKEND_URL: Backend server URL
KANDC_PROJECT: Default project name
KANDC_MODE: Default run mode (“online”, “offline”, “disabled”)
KANDC_API_KEY: API key for authentication

Profiling Configuration:

KANDC_PROFILER_DISABLED: Disable PyTorch profiler (“1” to disable)

Code Capture Configuration:

KANDC_CAPTURE_CODE: Enable/disable code capture (“true”, “false”)
KANDC_CODE_EXCLUDE: Comma-separated exclude patterns
KANDC_MAX_FILE_SIZE: Maximum file size in bytes for code capture

Example:

# Core configuration
export KANDC_MODE="offline"
export KANDC_PROJECT="default-project"

# Disable PyTorch profiler
export KANDC_PROFILER_DISABLED=1

# Disable code capture
export KANDC_CAPTURE_CODE="false"

# Or configure code exclusions
export KANDC_CODE_EXCLUDE="*.pth,data/,experiments/,*.log"
export KANDC_MAX_FILE_SIZE="2097152"  # 2MB

python my_experiment.py

Getting Started

Reference

Support

​Core Functions

​kandc.init()

​kandc.log()

​kandc.finish()

​kandc.get_current_run()

​kandc.is_initialized()

​PyTorch Profiling

​kandc.ProfilerWrapper

​kandc.ProfilerDecorator

​kandc.profile()

​kandc.profiler()

​Model Profiling

​kandc.capture_model_class()

​kandc.capture_model_instance()

​kandc.capture_trace()

​kandc.parse_model_trace()

​Timing Functions

​kandc.timed()

​kandc.timed_call()

​API Client (Advanced)

​kandc.APIClient

​Authentication

​kandc.get_api_key()

​kandc.ensure_authenticated()

​Run Object

​Properties

​Methods

​Run Modes

​Online Mode (Default)

​Offline Mode

​Disabled Mode

​Error Handling

​kandc.APIError

​kandc.AuthenticationError

​Code Snapshot Configuration

​Disable Code Capture Completely

​Custom Exclude Patterns

​Respect .gitignore

​What Gets Captured

​Environment Variables for Code Capture

​Environment Variables

Core Functions

`kandc.init()`

`kandc.log()`

`kandc.finish()`

`kandc.get_current_run()`

`kandc.is_initialized()`

PyTorch Profiling

`kandc.ProfilerWrapper`

`kandc.ProfilerDecorator`

`kandc.profile()`

`kandc.profiler()`

Model Profiling

`kandc.capture_model_class()`

`kandc.capture_model_instance()`

`kandc.capture_trace()`

`kandc.parse_model_trace()`

Timing Functions

`kandc.timed()`

`kandc.timed_call()`

API Client (Advanced)

`kandc.APIClient`

Authentication

`kandc.get_api_key()`

`kandc.ensure_authenticated()`

Run Object

Properties

Methods

Run Modes

Online Mode (Default)

Offline Mode

Disabled Mode

Error Handling

`kandc.APIError`

`kandc.AuthenticationError`

Code Snapshot Configuration

Disable Code Capture Completely

Custom Exclude Patterns

Respect .gitignore

What Gets Captured

Environment Variables for Code Capture

Environment Variables