Documentation
¶
Overview ¶
Package loader provides model weight loading functionality for Born ML framework.
This package wraps internal loader implementations and exports a clean public API for loading model weights from various formats (SafeTensors, GGUF).
Example usage:
import (
"github.com/born-ml/born/loader"
"github.com/born-ml/born/backend/cpu"
)
// Open model with auto-detection
model, err := loader.OpenModel("path/to/model.safetensors")
if err != nil {
log.Fatal(err)
}
defer model.Close()
// Get model information
fmt.Printf("Format: %s\n", model.Format())
fmt.Printf("Architecture: %s\n", model.Architecture())
// Load a specific tensor
backend := cpu.New()
tensor, err := model.LoadTensor("model.layers.0.attn.q_proj.weight", backend)
if err != nil {
log.Fatal(err)
}
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func DetectArchitecture ¶
DetectArchitecture attempts to detect model architecture from weight names. Returns "llama", "mistral", or "deepseek".
Types ¶
type ModelFormat ¶
type ModelFormat = loader.ModelFormat
ModelFormat represents the model weight format.
const ( FormatUnknown ModelFormat = loader.FormatUnknown FormatSafeTensors ModelFormat = loader.FormatSafeTensors FormatGGUF ModelFormat = loader.FormatGGUF )
Supported model formats.
type ModelReader ¶
type ModelReader = loader.ModelReader
ModelReader provides a unified interface for loading model weights. It abstracts away the underlying file format and provides consistent access to model tensors.
Note: This is a type alias because the LoadTensor method signature references internal tensor types that cannot be abstracted without a wrapper layer.
func OpenModel ¶
func OpenModel(path string) (ModelReader, error)
OpenModel opens a model file and auto-detects the format.
Supported formats:
- .safetensors (Hugging Face standard)
- .gguf (llama.cpp ecosystem)
The function automatically detects the model architecture (LLaMA, Mistral, DeepSeek) based on weight names and metadata.
Example:
model, err := loader.OpenModel("path/to/llama-7b.safetensors")
if err != nil {
log.Fatal(err)
}
defer model.Close()
fmt.Printf("Format: %s\n", model.Format()) // "SafeTensors"
fmt.Printf("Architecture: %s\n", model.Architecture()) // "llama"
// List all tensors
for _, name := range model.TensorNames() {
fmt.Println(name)
}
// Load specific tensor
backend := cpu.New()
weight, err := model.LoadTensor("model.layers.0.attn.q_proj.weight", backend)
if err != nil {
log.Fatal(err)
}
type WeightMapper ¶
type WeightMapper interface {
// MapName converts a model-specific weight name to Born standard name.
MapName(name string) (string, error)
// Architecture returns the architecture name (e.g., "llama", "mistral").
Architecture() string
}
WeightMapper maps model-specific weight names to standard Born names. Different model architectures use different naming conventions. This interface provides a way to normalize weight names.
func NewDeepSeekMapper ¶
func NewDeepSeekMapper() WeightMapper
NewDeepSeekMapper creates a weight mapper for DeepSeek models. Supports DeepSeek-V2 and DeepSeek-Coder.
func NewLLaMAMapper ¶
func NewLLaMAMapper() WeightMapper
NewLLaMAMapper creates a weight mapper for LLaMA models. Supports LLaMA, LLaMA 2, and LLaMA 3.
func NewMistralMapper ¶
func NewMistralMapper() WeightMapper
NewMistralMapper creates a weight mapper for Mistral models. Supports Mistral 7B and Mixtral 8x7B MoE.
Source Files