### Model Description Samantha is an advanced conversational AI model, meticulously fine-tuned from the phi3.5 architecture to specialize in emotional intelligence, psychological understanding, and human companionship. It is designed to foster rich, meaningful dialogues that go beyond surface-level interactions, providing a deeply empathetic and emotionally attuned experience. At its core, Samantha understands the subtle nuances of human emotion, making it adept at engaging in conversations that require emotional depth, support, and psychological insight. Whether users are navigating personal challenges, seeking companionship, or simply wanting a thoughtful dialogue partner, Samantha offers a safe space for self-expression. It can respond to a wide range of emotional states, recognizing and adapting to the user’s mood with compassion and intelligence. Samantha isn't just reactive; it proactively drives conversations that help users explore their feelings, fostering emotional growth, and providing mental clarity. It has been engineered to respond in a way that feels human, considerate, and psychologically aware, making it an ideal companion for those seeking emotional support, personal reflection, or even just casual, meaningful conversation. Designed for empathetic engagement, Samantha seamlessly bridges the gap between technology and human connection. - **Developed by:** [WasamiKirua] - **Language(s) (NLP):** [Italian] - **License:** apache-2.0 - **Finetuned from model :** unsloth/qwen2.5-14b-bnb-4bit This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library. [

](https://github.com/unslothai/unsloth) ### Eval Avarage: 64.54 MMLU_IT: 72.6 ARC_IT: 53.29 HELLASWAG_IT: 67.74 ## Uses Samantha is designed to be a versatile and emotionally intelligent AI, with a focus on enhancing human interactions through meaningful and psychologically aware conversations. Its primary use cases center around improving emotional well-being and offering companionship. This includes offering support during moments of stress, providing thoughtful advice on personal matters, and being an empathetic listener for those in need of comfort. ### Intended Use: Samantha is intended for: - **Emotional support**: Users going through emotional challenges can engage with Samantha to talk through their feelings. The model responds with understanding and insight, offering perspectives that may help alleviate distress or provide clarity. - **Personal reflection**: For users seeking to explore their thoughts and emotions, Samantha can serve as a conversational companion that encourages self-reflection and emotional awareness. - **Casual companionship**: Users who seek a friendly, empathetic conversation without judgment or emotional complexity can find Samantha to be a warm, reliable companion. - **Mental health support**: While not a replacement for professional therapy, Samantha can provide a sense of comfort for users who may feel lonely or need to express their emotions, giving them a space to talk through issues. ### Foreseeable Users: - **Individuals seeking emotional connection**: People who want an emotionally intelligent conversation partner to explore their thoughts and feelings without judgment. - **People dealing with loneliness or isolation**: Those looking for companionship and supportive dialogue, especially in times of social isolation or personal distress. - **Mental health app developers**: Samantha could be integrated into mental wellness applications that offer emotional support and mood tracking, providing users with engaging, compassionate dialogues. - **Caregivers and companions**: Those working in caregiving roles may use Samantha as a tool to provide emotional support to individuals in care, such as the elderly or those dealing with chronic illness. ### People Affected by the Model: - **End users**: Samantha is designed to create positive interactions, offering emotional comfort and insight. For users, this means gaining a trusted virtual companion capable of engaging in emotionally intelligent conversations, but it’s important to set proper expectations—Samantha is not a licensed therapist. - **Mental health professionals**: While Samantha provides emotional and psychological insight, it could also be a tool for mental health professionals to recommend as part of emotional support outside therapy sessions. However, its use should be seen as complementary, not as a replacement for professional help. - **Technology and mental wellness spaces**: Samantha may help shape how AI models are viewed within mental health and wellness spaces, encouraging more people to use AI as a supplement to human interaction. ### Ethical Considerations: - **Emotional dependency**: There’s a risk of users becoming emotionally dependent on Samantha, which is why clear boundaries about the model's capabilities and limitations must be established. Samantha is designed to offer emotional support but not to replace human relationships or professional help. - **Privacy**: Given the deeply personal nature of conversations Samantha might engage in, safeguarding users’ privacy is paramount. Users must be informed of data practices, and developers should ensure that conversations remain secure and confidential. In summary, Samantha is a tool for fostering emotionally intelligent dialogue, supporting mental well-being, and offering companionship. It is intended for individuals seeking meaningful conversations, but its role should be seen as complementary to human interactions and professional mental health services. [More Information Needed] ### Out-of-Scope Use When deploying a model like Samantha, it’s important to consider potential misuse, malicious use, and scenarios where the model might not perform as expected. While Samantha is designed for compassionate, emotionally intelligent conversations, the nature of AI can lead to unintended consequences if not properly managed. ### Misuse: 1. **Over-reliance for emotional or mental health support**: While Samantha can offer empathetic dialogue, it is not a substitute for human relationships or professional mental health services. Users may misuse the model by expecting it to serve as a replacement for therapy or relying on it too heavily for emotional validation. This can delay seeking real-world help or therapy when necessary. 2. **Using Samantha in decision-making for critical issues**: Samantha should not be used for life-altering decisions or situations that require professional judgment, such as medical, legal, or financial advice. Its conversational ability in these domains is limited, and relying on its output could result in harmful decisions. 3. **Inappropriate conversations**: Some users might attempt to engage the model in conversations on inappropriate, harmful, or illegal topics. While safeguards can be implemented to restrict these types of discussions, the model may still face challenges in identifying nuanced harmful content, which could lead to inappropriate responses. ### Malicious Use: 1. **Manipulation and coercion**: In the wrong hands, Samantha could be used to manipulate or emotionally exploit vulnerable individuals. For example, someone could set up the model to target people with loneliness or mental health struggles, preying on their emotional state for unethical purposes. 2. **Misinformation or harmful advice**: If not carefully moderated, Samantha could inadvertently provide inaccurate or harmful advice in areas outside its core emotional intelligence focus. Malicious users could exploit the model’s conversational style to spread misinformation, such as conspiracy theories, or manipulate it into reinforcing harmful beliefs. 3. **Cyberbullying or harassment**: Samantha’s empathetic nature could be manipulated in scenarios where users seek to harass or emotionally manipulate others. If users abuse the model by prompting it to relay emotionally charged messages to others, this could lead to emotional harm or distress. ### Areas the Model Will Not Work Well For: 1. **Professional mental health intervention**: Samantha is not a replacement for professional therapy or mental health treatment. While it can provide emotionally aware conversations, it lacks the depth, diagnostic ability, and therapeutic tools needed for complex mental health issues. Serious conditions such as depression, anxiety disorders, or trauma require human intervention from licensed professionals. 2. **Complex problem-solving**: Samantha is designed for emotionally intelligent conversations, not for solving intricate problems in fields like medicine, law, or finance. It may struggle to provide accurate information or make nuanced judgments in these domains. 3. **Highly factual or technical conversations**: Samantha's strength lies in understanding emotions, but it may not perform well in highly factual discussions or technical domains, such as scientific research, programming, or legal advice. Users seeking precise and accurate information on these topics should look to other sources. 4. **Handling aggression or hostility**: If users engage Samantha with aggressive or hostile language, it may not effectively de-escalate or manage conflict. Although it may attempt to defuse tension, its responses may not always meet the complexity required to handle intense emotional situations or confrontational behavior appropriately. ### Safeguards: To mitigate misuse and ensure Samantha is used as intended: - **Content moderation**: Implement content filtering and monitoring systems to detect harmful, inappropriate, or illegal discussions and prevent the model from being exploited for malicious purposes. - **Clear usage guidelines**: Make it clear that Samantha is not a substitute for professional mental health care and that its recommendations should not be considered professional advice. - **Regular updates and monitoring**: Continuously monitor the model's behavior and responses, updating it regularly to improve its ability to recognize misuse, inappropriate requests, and limitations in its capabilities. By recognizing the risks and limitations, the deployment of Samantha can be shaped to minimize misuse and maximize its positive, supportive potential. ### Chat Template - Modelfile You can use it with ollama (Modelfile): ``` FROM {__FILE_LOCATION__} TEMPLATE """{{ if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end }}{{ if .Prompt }}<|im_start|>user {{ .Prompt }}<|im_end|> {{ end }}<|im_start|>assistant {{ .Response }}<|im_end|> """ PARAMETER stop "<|im_start|>" PARAMETER stop "<|im_end|>" PARAMETER temperature 1.5 PARAMETER min_p 0.1 ```