Stablelm Tuned Alpha 3B is an open-source language model by stabilityai. Features: 3b LLM, VRAM: 14.9GB, Context: 4K, License: cc-by-nc-sa-4.0, LLM Explorer Score: 0.12, Arc: 27.8, HellaSwag: 44.1, MMLU: 23.1, GSM8K: 0.5.
Stablelm Tuned Alpha 3B Parameters and Internals
| Model Type | |
| Use Cases |
| Areas: | | open-source community, chat-like applications |
|
| Limitations: | | The model may generate biased or toxic text despite efforts in safe fine-tuning., Not intended as a replacement for human judgment |
|
| Considerations: | | Be mindful of potential bias or toxic outputs. |
|
|
| Additional Notes | | Models include a helpful hand from Dakota Mahan ([@dmayhem93](https://huggingface.co/dmayhem93)) in their development. |
|
| Supported Languages | |
| Training Details |
| Data Sources: | | tatsu-lab/alpaca, nomic-ai/gpt4all_prompt_generations, Dahoas/full-hh-rlhf, jeffwan/sharegpt_vicuna, HuggingFaceH4/databricks_dolly_15k |
|
| Methodology: | | Supervised fine-tuning on natural language datasets focused on chat and instruction-following tasks. |
|
| Context Length: | |
| Model Architecture: | | NeoX transformer architecture |
|
|
| Responsible Ai Considerations |
| Fairness: | | Models are developed to adhere to safer distributions of text but cannot mitigate all biases and toxicity. |
|
| Transparency: | | It should not be treated as a substitute for human judgment or considered a source of truth. |
|
| Accountability: | | Users are responsible for the outputs generated and should use models responsibly. |
|
| Mitigation Strategies: | | Fine-tuning on datasets aimed at improving safety, but may not remove all biases/toxicity. |
|
|
| Input Output |
| Input Format: | | Prompts formatted to <|SYSTEM|>...<|USER|>...<|ASSISTANT|>... |
|
| Accepted Modalities: | |
| Output Format: | |
|
Quantized Models of the Stablelm Tuned Alpha 3B
Best Alternatives to Stablelm Tuned Alpha 3B
Note: green Score (e.g. "73.2") means that the model is better than stabilityai/stablelm-tuned-alpha-3b.