Model Registry

The ModelRegistry acts as the single source of truth for all components. It maps model identifiers (e.g., gpt-4o) to their respective tokenization backends, context limits, and pricing rates.

Built-in Models

LLM Context Forge ships with configurations for major production models.

Model ID	Provider	Context Window	Input (per 1M)	Output (per 1M)
gpt-4o	openai	128,000	$2.50	$10.00
gpt-4o-mini	openai	128,000	$0.15	$0.60
gpt-4	openai	8,192	$30.00	$60.00
claude-3-5-sonnet	anthropic	200,000	$3.00	$15.00
claude-3-haiku	anthropic	200,000	$0.25	$1.25
gemini-1.5-pro	google	2,000,000	$1.25	$5.00
gemini-1.5-flash	google	1,000,000	$0.075	$0.30
llama-3.1-405b	meta	128,000	$0.00	$0.00
mistral-large	mistral	32,000	$2.00	$6.00

Note: Pricing and windows are updated regularly to track provider changes.

Registering Custom Models

You can register fine-tuned, self-hosted, or new models dynamically at runtime.

Basic Registration

If your model uses a standard encoding (like cl100k_base), registration is simple:

from llm_context_forge import ModelRegistry, ModelConfig

ModelRegistry.register(
    "company-finetune-v1",
    ModelConfig(
        provider="self-hosted",
        context_window=32000,
        encoding="cl100k_base",
        input_price_1m=0.0,
        output_price_1m=0.0
    )
)

# Now other components can use it
counter = TokenCounter("company-finetune-v1")

Custom Tokenizer Backend

If you rely on a specific HuggingFace tokenizer, you can map the model to it:

ModelRegistry.register(
    "llama-3-custom",
    ModelConfig(
        provider="self-hosted",
        context_window=8192,
        backend="transformers",
        hf_repo_id="meta-llama/Meta-Llama-3-8B"
    )
)