lzw1008/Emollama-7b
Emollama-7b, developed by lzw1008 as part of the EmoLLMs project, is a 7 billion parameter instruction-following large language model fine-tuned from Meta's LLaMA2-7B. It specializes in comprehensive affective analysis, including sentiment polarity, categorical emotions, sentiment strength, and emotion intensity. The model is trained on the full AAID instruction tuning data and supports a 4096 token context length.
Loading preview...
Emollama-7b: Affective Analysis LLM
Emollama-7b is a 7 billion parameter model from the EmoLLMs project by lzw1008, designed for comprehensive affective analysis. It is fine-tuned from the LLaMA2-7B foundation model using the full AAID instruction tuning dataset.
Key Capabilities
- Affective Classification: Performs tasks such as sentimental polarity and categorical emotion identification.
- Affective Regression: Handles tasks like sentiment strength and emotion intensity scoring.
- Instruction Following: Capable of understanding and executing instructions for various affective analysis tasks.
Use Cases
Emollama-7b is particularly well-suited for applications requiring detailed emotional and sentiment understanding from text. Examples include:
- Emotion Intensity Scoring: Assigning numerical values (e.g., 0 to 1) for specific emotions like joy.
- Sentiment Strength Evaluation: Quantifying the valence intensity of a writer's mental state (e.g., 0 for most negative to 1 for most positive).
- Sentiment Classification: Categorizing text into ordinal classes representing degrees of positive, neutral, or negative sentiment.
- Emotion Classification: Identifying the presence of specific emotions (e.g., anger, joy, sadness) or classifying text as 'neutral or no emotion'.
Ethical Considerations
The developers note potential biases, incorrect predictions, and over-generalization in LLMs, advising caution when applying the model to real-world affective analysis systems.