31 lines
808 B
Markdown
31 lines
808 B
Markdown
|
---
|
||
|
|
base_model: HuggingFaceTB/SmolLM2-360M-Instruct
|
||
|
|
library_name: transformers
|
||
|
|
license: apache-2.0
|
||
|
|
tags:
|
||
|
|
- memory-router
|
||
|
|
- phone-assistant
|
||
|
|
- classification
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
---
|
||
|
|
|
||
|
|
## Hermie Assistant - Memory Router
|
||
|
|
|
||
|
|
# Sys prompt:
|
||
|
|
"<MEM>\n{"fact":str|null,"retrieve":bool,"tool":bool,"emotion":str}\nfact: durable personal fact, compressed. null if nothing to store."
|
||
|
|
|
||
|
|
# Output format:
|
||
|
|
{"fact":str|null,"retrieve":bool,"tool":bool,"emotion":str}
|
||
|
|
|
||
|
|
# Tool:
|
||
|
|
Basic tool intent detection. For context for larger LLM.
|
||
|
|
|
||
|
|
# Retrieve:
|
||
|
|
Binary gate that engages the embedding model for memory retrieval.
|
||
|
|
|
||
|
|
# Emotion
|
||
|
|
Emotional response for larger LLM, persistent accross chats.
|
||
|
|
|
||
|
|
|
||
|
|
Fine-tuned from [SmolLM2-360M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct) for personal assistant memory classification.
|