--- base_model: - meta-llama/Llama-3.3-70B-Instruct tags: - state-of-the-art - reasoning - chain-of-thought - text-generation - transformers - llama - instruction-tuning license: apache-2.0 language: - en datasets: - Daemontatox/Deepthinking-COT - gghfez/QwQ-LongCoT-130K-cleaned pipeline_tag: text-generation library_name: transformers model-index: - name: Llama3.3-70B-CogniLink results: - task: type: text-generation name: Text Generation dataset: name: IFEval (0-Shot) type: wis-k/instruction-following-eval split: train args: num_few_shot: 0 metrics: - type: inst_level_strict_acc and prompt_level_strict_acc value: 69.31 name: averaged accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FLlama3.3-70B-CogniLink name: Open LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: BBH (3-Shot) type: SaylorTwift/bbh split: test args: num_few_shot: 3 metrics: - type: acc_norm value: 52.12 name: normalized accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FLlama3.3-70B-CogniLink name: Open LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: MATH Lvl 5 (4-Shot) type: lighteval/MATH-Hard split: test args: num_few_shot: 4 metrics: - type: exact_match value: 39.58 name: exact match source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FLlama3.3-70B-CogniLink name: Open LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: GPQA (0-shot) type: Idavidrein/gpqa split: train args: num_few_shot: 0 metrics: - type: acc_norm value: 26.06 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FLlama3.3-70B-CogniLink name: Open LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: MuSR (0-shot) type: TAUR-Lab/MuSR args: num_few_shot: 0 metrics: - type: acc_norm value: 21.4 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FLlama3.3-70B-CogniLink name: Open LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: MMLU-PRO (5-shot) type: TIGER-Lab/MMLU-Pro config: main split: test args: num_few_shot: 5 metrics: - type: acc value: 46.37 name: accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FLlama3.3-70B-CogniLink name: Open LLM Leaderboard --- ![image](./image.webp) # Model Card: CogniLink - Redefining Reasoning AI ## Overview CogniLink is a **state-of-the-art (SOTA) reasoning model**, engineered to set new benchmarks in logical problem-solving and chain-of-thought capabilities. Leveraging the power of LLaMA 3.3 70B, CogniLink excels in multi-step reasoning, inference, and real-time decision-making across diverse domains. Whether tackling mathematical proofs, legal analyses, or dynamic real-world scenarios, CogniLink ensures clarity, precision, and scalability. Designed for both **high-performance tasks** and **resource-efficient environments**, CogniLink represents the perfect fusion of innovation and practicality. --- ## Key Features - **Base Model:** [unsloth/llama-3.3-70b-instruct](https://huggingface.co/unsloth/llama-3.3-70b-instruct-bnb-4bit) - **Developed By:** Daemontatox - **License:** Apache 2.0 (open and permissive) - **Primary Language:** English - **Specialization:** Multi-domain reasoning, step-by-step logic, and advanced inference. **CogniLink is optimized for tasks requiring:** - **Reasoning Depth:** Multi-step logic with exceptional accuracy. - **Chain-of-Thought (CoT):** Built-in mechanisms to generate clear, stepwise reasoning paths. - **Resource Efficiency:** Ideal for deployment on both high-performance servers and resource-constrained devices, including edge computing platforms. --- ## Training and Optimization CogniLink’s fine-tuning was accelerated using **[Unsloth](https://github.com/unslothai/unsloth)**, enabling a **2x faster training pipeline**. The training process was powered by Hugging Face's **TRL library**, ensuring seamless instruction tuning and robust adaptability across reasoning-heavy applications. With advanced techniques like **quantization-aware training** and parameter-efficient fine-tuning, CogniLink is lightweight without compromising on performance, making it a top choice for edge deployment and embedded systems. Special thanks to **[Modal.com](https://modal.com)** for providing **H100 GPUs**, which enabled accelerated training and optimized performance for CogniLink. Their generous support significantly contributed to the model’s development and deployment readiness. --- ## Applications CogniLink is versatile and excels in various industries: ### **1. Education and Training** - Powers AI tutors for **step-by-step problem-solving** in STEM education. - Supports interactive learning tools with detailed explanations. ### **2. Research and Academia** - Assists researchers with **hypothesis testing**, complex analysis, and paper drafting. - Enhances productivity in tasks requiring deep logical reasoning. ### **3. Business Decision Support** - Real-time **scenario analysis** for strategic decision-making. - Risk assessment tools for dynamic business environments. ### **4. Legal and Policy Analysis** - Enables multi-step reasoning for **case law interpretations** and **regulatory reviews**. - Assists legal professionals with clear and logical argument generation. ### **5. Healthcare AI** - Supports diagnostics and medical workflows with robust reasoning models. - Ensures accuracy in multi-step inferential tasks like patient case reviews. --- ## Technical Specifications - **Quantization:** Fully compatible with 4-bit inference for efficient performance. - **Latency:** Optimized for real-time responses in latency-sensitive applications. - **Scalability:** Deployable on diverse hardware setups, from high-end GPUs to edge devices. --- ## Why Choose CogniLink? CogniLink isn’t just a model; it’s a **reasoning companion**. Its fine-tuned chain-of-thought design ensures not just answers, but **rational, explainable processes**, giving users the confidence and insights they need to make critical decisions. - **Transparent Reasoning:** Every decision is backed by a logical thought process. - **Versatile Applications:** From academia to business, CogniLink adapts effortlessly. - **Cutting-Edge Efficiency:** High performance meets cost-effectiveness. --- ## Get Started CogniLink is available for download and deployment. Start integrating advanced reasoning into your applications today! For inquiries, contributions, or support, visit **[Unsloth GitHub](https://github.com/unslothai/unsloth)**. **CogniLink: Connecting Intelligence with Clarity.** # [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/Daemontatox__Llama3.3-70B-CogniLink-details)! Summarized results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/contents/viewer/default/train?q=Daemontatox%2FLlama3.3-70B-CogniLink&sort[column]=Average%20%E2%AC%86%EF%B8%8F&sort[direction]=desc)! | Metric |Value (%)| |-------------------|--------:| |**Average** | 42.47| |IFEval (0-Shot) | 69.31| |BBH (3-Shot) | 52.12| |MATH Lvl 5 (4-Shot)| 39.58| |GPQA (0-shot) | 26.06| |MuSR (0-shot) | 21.40| |MMLU-PRO (5-shot) | 46.37|