Embedders Reference

Embedders convert text chunks into vector representations for semantic search. LlamaFarm supports multiple embedding providers to match your infrastructure and requirements.

Quick Start

Embedders are configured within embedding_strategies in your database definition:

rag:
  databases:
    - name: my_database
      embedding_strategies:
        - name: default_embeddings
          type: OllamaEmbedder
          config:
            model: nomic-embed-text

---

UniversalEmbedder

Flexible embedder that works with Ollama and falls back to HuggingFace models.

Best for: Most use cases, automatic model management

- name: default_embeddings
  type: UniversalEmbedder
  config:
    model: nomic-ai/nomic-embed-text-v2-moe
    base_url: http://localhost:11434
    dimension: 768

Options

Option	Type	Default	Required	Description
model	string	nomic-ai/nomic-embed-text-v2-moe	No	HuggingFace model ID
base_url	string	http://localhost:11434	No	Ollama API endpoint
dimension	integer	768	No	Embedding dimension (128-4096)
batch_size	integer	16	No	Batch processing size (1-128)
timeout	integer	60	No	Request timeout in seconds

---

OllamaEmbedder

Native Ollama embedder with auto-pull support.

Best for: Ollama-based deployments, local inference

- name: ollama_embeddings
  type: OllamaEmbedder
  config:
    model: nomic-embed-text:latest
    base_url: http://localhost:11434
    auto_pull: true

Options

Option	Type	Default	Required	Description
model	string	nomic-embed-text	No	Ollama model name
base_url	string	http://localhost:11434	No	Ollama API endpoint
dimension	integer	768	No	Embedding dimension (128-4096)
batch_size	integer	16	No	Batch processing size (1-128)
timeout	integer	60	No	Request timeout in seconds
auto_pull	boolean	true	No	Auto-pull missing models

Recommended Models

# Pull embedding models
ollama pull nomic-embed-text          # 768 dim, general purpose
ollama pull mxbai-embed-large         # 1024 dim, higher quality
ollama pull all-minilm                # 384 dim, fast

---

HuggingFaceEmbedder (Coming Soon)

Coming Soon

HuggingFaceEmbedder is planned but not yet implemented. Use UniversalEmbedder or OllamaEmbedder instead.

Direct HuggingFace model loading with GPU/MPS support.

Best for: Custom models, GPU acceleration, offline use

- name: hf_embeddings
  type: HuggingFaceEmbedder
  config:
    model_name: sentence-transformers/all-MiniLM-L6-v2
    device: auto
    normalize_embeddings: true

Planned Options

Option	Type	Default	Required	Description
model_name	string	sentence-transformers/all-MiniLM-L6-v2	No	HuggingFace model ID
device	string	auto	No	cpu, cuda, mps, or auto
batch_size	integer	32	No	Batch size (1-256)
normalize_embeddings	boolean	true	No	L2 normalize embeddings
show_progress_bar	boolean	false	No	Show progress bar
cache_folder	string	null	No	Model cache directory

Recommended Models (for when available)

Model	Dimensions	Speed	Quality
all-MiniLM-L6-v2	384	Fast	Good
all-mpnet-base-v2	768	Medium	High
BAAI/bge-base-en-v1.5	768	Medium	High
BAAI/bge-large-en-v1.5	1024	Slow	Highest

---

SentenceTransformerEmbedder (Coming Soon)

Coming Soon

SentenceTransformerEmbedder is planned but not yet implemented. Use UniversalEmbedder or OllamaEmbedder instead.

Sentence Transformers library integration.

Best for: Sentence-level embeddings, specialized models

- name: st_embeddings
  type: SentenceTransformerEmbedder
  config:
    model_name: sentence-transformers/all-MiniLM-L6-v2
    device: cpu

Planned Options

Option	Type	Default	Required	Description
model_name	string	sentence-transformers/all-MiniLM-L6-v2	No	Model name
device	string	cpu	No	cpu, cuda, or mps

---

OpenAIEmbedder (Coming Soon)

Coming Soon

OpenAIEmbedder is planned but not yet implemented. Use UniversalEmbedder or OllamaEmbedder instead.

OpenAI embedding API integration.

Best for: OpenAI API users, high-quality embeddings

- name: openai_embeddings
  type: OpenAIEmbedder
  config:
    model: text-embedding-3-small
    api_key: ${OPENAI_API_KEY}

Options

Option	Type	Default	Required	Description
model	string	text-embedding-3-small	No	OpenAI embedding model
api_key	string	-	Yes	OpenAI API key
base_url	string	null	No	Custom API base URL
organization	string	null	No	OpenAI organization ID
batch_size	integer	100	No	Batch size (1-2048)
max_retries	integer	3	No	Maximum retry attempts
timeout	integer	60	No	Request timeout in seconds

Available Models

Model	Dimensions	Max Tokens	Use Case
text-embedding-3-small	1536	8191	Cost-effective
text-embedding-3-large	3072	8191	Highest quality
text-embedding-ada-002	1536	8191	Legacy

---

Multiple Embedding Strategies

Configure multiple embedders for different content types:

embedding_strategies:
  - name: general
    type: OllamaEmbedder
    config:
      model: nomic-embed-text
    priority: 0

  - name: code
    type: HuggingFaceEmbedder
    config:
      model_name: microsoft/codebert-base
    condition: "doc.type == 'code'"
    priority: 1

default_embedding_strategy: general

Dimension Matching

Ensure embedding dimensions match your vector store configuration:

databases:
  - name: my_db
    type: ChromaStore
    config:
      embedding_dimension: 768  # Must match embedder output

    embedding_strategies:
      - name: default
        type: OllamaEmbedder
        config:
          model: nomic-embed-text
          dimension: 768  # Match store dimension

Next Steps

• Extractors Reference - Add metadata extraction
• Databases Reference - Configure vector stores
• RAG Guide - Full RAG overview