--- base_model: unsloth/qwen2.5-7b-instruct-bnb-4bit tags: - text-generation-inference - transformers - unsloth - qwen2 - trl license: apache-2.0 language: - en model-index: - name: AetherSett 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: 53.7 name: averaged accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FAetherSett 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: 34.74 name: normalized accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FAetherSett 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: 30.74 name: exact match source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FAetherSett 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: 7.72 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FAetherSett 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: 16.21 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FAetherSett 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: 36.43 name: accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=Daemontatox%2FAetherSett name: Open LLM Leaderboard --- ![image](./image.webp) # SwiftText: Accelerating Text Generation with Optimized Efficiency - **Developed by:** Daemontatox - **License:** apache-2.0 - **Base Model:** [unsloth/qwen2.5-7b-instruct-bnb-4bit](https://huggingface.co/unsloth/qwen2.5-7b-instruct-bnb-4bit) SwiftText is a fine-tuned language model built upon the robust Qwen2.5 architecture, specifically optimized for rapid and efficient text generation. Leveraging the power of the [Unsloth](https://github.com/unslothai/unsloth) framework and Hugging Face's TRL library, SwiftText was trained in half the time of conventional methods, delivering impressive performance without compromising quality. [](https://github.com/unslothai/unsloth) ## Model Overview: Efficiency Meets Performance SwiftText embodies the principle of efficient AI development. By harnessing advanced optimization techniques, this model achieves significant reductions in training time, making powerful language models more accessible and accelerating the development cycle for text generation applications. While the primary focus was on speed, SwiftText retains the strong text generation and instruction-following capabilities of its Qwen2.5 base. ### Key Highlights: Speed and Streamlined Development - **Accelerated Training:** Trained 2x faster thanks to the innovative optimizations provided by the Unsloth framework. This rapid training cycle reduces development time and resource consumption. - **Leveraging Qwen2.5 Strength:** Built upon the capable Qwen2.5-7B-Instruct model, inheriting its proficiency in understanding and generating coherent and contextually relevant text. - **Optimized for Efficiency:** Fine-tuned with a focus on maximizing throughput and minimizing computational cost, making it suitable for a wider range of deployment scenarios. - **TRL-Enhanced:** Integrated with Hugging Face's TRL library during training, potentially benefiting from techniques that improve generation quality and alignment. ### Use Cases: Where Speed and Efficiency Matter SwiftText is well-suited for applications where rapid text generation is crucial, such as: - **Fast Content Creation:** Quickly generating drafts for articles, social media posts, and marketing materials. - **Efficient Summarization:** Rapidly summarizing documents and extracting key information. - **Responsive Chatbots:** Powering conversational agents that require quick and relevant responses. - **Prototyping and Experimentation:** Accelerating the development and testing of new features that rely on text generation. - **Educational Tools:** Providing quick explanations and generating examples for learning purposes. ### Training Methodology: Harnessing the Power of Optimization SwiftText's rapid training was achieved through the synergistic use of: - **Unsloth Framework:** A next-generation framework designed to dramatically accelerate the training of large language models through optimized memory management and computational efficiency. - **Hugging Face's TRL Library:** Utilizing tools from the TRL library may have contributed to improved generation quality and alignment through techniques like reinforcement learning. ### Getting Started with SwiftText You can easily load and use SwiftText with the Hugging Face Transformers library: ```python from transformers import AutoModelForCausalLM, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("Daemontatox/your-swifttext-model-name") # Replace with your actual model name model = AutoModelForCausalLM.from_pretrained("Daemontatox/your-swifttext-model-name", device_map="auto", load_in_4bit=True) prompt = "Write a short introduction about efficient text generation." inputs = tokenizer(prompt, return_tensors="pt").to("cuda") # Or your preferred device outputs = model.generate(**inputs) print(tokenizer.decode(outputs[0], skip_special_tokens=True)) ``` ## Acknowledgements ### We gratefully acknowledge the contributions of the Unsloth team and the Hugging Face community for their invaluable tools and resources, which were instrumental in the development of SwiftText.# [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__AetherSett-details)! Summarized results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/contents/viewer/default/train?q=Daemontatox%2FAetherSett&sort[column]=Average%20%E2%AC%86%EF%B8%8F&sort[direction]=desc)! | Metric |Value (%)| |-------------------|--------:| |**Average** | 29.92| |IFEval (0-Shot) | 53.70| |BBH (3-Shot) | 34.74| |MATH Lvl 5 (4-Shot)| 30.74| |GPQA (0-shot) | 7.72| |MuSR (0-shot) | 16.21| |MMLU-PRO (5-shot) | 36.43| # [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__AetherSett-details)! Summarized results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/contents/viewer/default/train?q=Daemontatox%2FAetherSett&sort[column]=Average%20%E2%AC%86%EF%B8%8F&sort[direction]=desc)! | Metric |Value (%)| |-------------------|--------:| |**Average** | 29.92| |IFEval (0-Shot) | 53.70| |BBH (3-Shot) | 34.74| |MATH Lvl 5 (4-Shot)| 30.74| |GPQA (0-shot) | 7.72| |MuSR (0-shot) | 16.21| |MMLU-PRO (5-shot) | 36.43|