sheryc/Llama-3.1-8B-Instruct-CARE
sheryc/Llama-3.1-8B-Instruct-CARE is an 8 billion parameter instruction-tuned language model based on Meta's Llama-3.1-8B-Instruct, developed by sheryc. This model is enhanced with native retrieval-augmented reasoning capabilities through the CARE framework, specifically trained to improve context fidelity and reduce hallucinations. It excels at complex reasoning tasks by explicitly integrating in-context evidence and generating structured reasoning chains with evidence citations. The model features a 128,000 token context length and is optimized for tasks requiring robust evidence-based reasoning.
Loading preview...
Llama-3.1-8B-Instruct-CARE: Context-Aware Retrieval-Enhanced Reasoning
This model, sheryc/Llama-3.1-8B-Instruct-CARE, is an 8 billion parameter instruction-tuned variant of Meta's Llama-3.1-8B-Instruct. It integrates the CARE (Context-Aware Retrieval-Enhanced reasoning) framework to provide native retrieval-augmented reasoning capabilities, significantly improving context fidelity and reducing hallucinations.
Key Capabilities & Features
- Native Retrieval-Augmented Reasoning: The model dynamically identifies and incorporates relevant evidence from the input context into its reasoning process.
- Improved Context Fidelity: Demonstrates superior adherence to provided context, even when it contradicts the model's parametric knowledge.
- Enhanced Multi-Hop Reasoning: Excels at complex reasoning tasks that require integrating multiple pieces of evidence.
- Structured Reasoning Output: Generates explicit reasoning chains using
<think>and<retrieval>tags, citing evidence directly from the input. - Advanced Training: Utilizes a two-phase training approach, combining Supervised Fine-Tuning (SFT) on HotpotQA with Reinforcement Learning (GRPO) on datasets like DROP and MS MARCO.
- Extended Context: Features a substantial 128,000 token context length.
When to Use This Model
This model is particularly well-suited for applications requiring:
- High-stakes QA systems: Where accuracy and verifiable answers based on provided context are critical.
- Complex analytical tasks: That benefit from explicit, evidence-backed reasoning steps.
- Reducing hallucinations: In scenarios where models tend to generate factually incorrect information.
- Structured output generation: For tasks needing transparent and auditable reasoning processes.
It is ideal for developers building systems where the model's ability to ground its responses in given information is paramount, offering a robust solution for evidence-based AI applications.