Timothyemmanuel/Arakandar

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 20, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Timothyemmanuel/Arakandar is a 3.1 billion parameter causal language model designed as a local market research assistant. This base model, with a 32,768 token context window and grouped-query attention, is specifically engineered to explain structured market evidence. It integrates with a deterministic LightGBM signal model and purpose-built tools to provide grounded explanations of market values, news, and workflow, rather than generating autonomous trading advice.

Loading preview...

Arakandar 3B Base Model

Arakandar is a 3.1 billion parameter causal language model developed by Timothyemmanuel, serving as the core component of a local market research assistant application. This base model is designed to explain structured market evidence, integrating with a deterministic LightGBM signal model and other tools.

Key Capabilities

  • Market Evidence Explanation: Focuses on explaining supplied market values, news evidence, workflow, and memory.
  • Grounded Responses: Explicitly designed not to invent prices, news, probabilities, or trades, ensuring explanations are based solely on provided data.
  • Integration with Deterministic Signals: Works in conjunction with a LightGBM signal model, where the LightGBM remains authoritative for 'BUY', 'HOLD', or 'SELL' decisions.
  • Large Context Window: Features a 32,768 token context window, allowing for comprehensive analysis of market data.
  • Efficient Architecture: Utilizes grouped-query attention with 16 query heads and 2 key/value heads across 36 layers.

Usage and Specialization

This repository provides the base model. For conversation specialization, a separate LoRA adapter (e.g., Timothyemmanuel/Arakandar) is distributed and should be loaded on top of this base model. The model requires approximately 10-12 GB of VRAM for practical inference on an NVIDIA GPU. The application combines this language model with a deterministic LightGBM signal and purpose-built tools to provide market insights.