File size: 8,540 Bytes
6488982 c418e48 6488982 8a7b8dc b251d73 6488982 b251d73 6488982 b251d73 6488982 b251d73 c418e48 cc50b4f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 |
---
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
---
# 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.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](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|
|