| --- |
| language: |
| - en |
| license: apache-2.0 |
| library_name: transformers |
| tags: |
| - text-classification |
| - intent-classification |
| - query-routing |
| - agent |
| - llm-router |
| pipeline_tag: text-classification |
| --- |
| |
| # ⚡ AgentRouter |
|
|
| Ultra-fast intent classification for LLM query routing. Classifies user queries into 10 intent categories in **<5ms** on GPU. |
|
|
| Built on [MiniLM](https://huggingface.co/microsoft/MiniLM-L12-H384-uncased) (33M params) — small enough for CPU inference, fast enough for real-time routing. |
|
|
| ## 🚀 Usage |
|
|
| ```python |
| from transformers import pipeline |
| |
| router = pipeline("text-classification", model="ENTUM-AI/AgentRouter") |
| |
| router("Write a Python function to sort a list") |
| # [{'label': 'code_generation', 'score': 0.98}] |
| |
| router("Why am I getting a TypeError?") |
| # [{'label': 'code_debugging', 'score': 0.97}] |
| |
| router("Translate hello to Spanish") |
| # [{'label': 'translation', 'score': 0.99}] |
| |
| router("What is quantum computing?") |
| # [{'label': 'information_retrieval', 'score': 0.96}] |
| ``` |
|
|
| ## 🏷️ Intent Classes |
|
|
| | Intent | Description | Suggested Tools | |
| |--------|-------------|----------------| |
| | `code_generation` | Write new code | code_interpreter, file_editor | |
| | `code_debugging` | Fix bugs and errors | code_interpreter, debugger | |
| | `math_reasoning` | Solve math problems | calculator, wolfram_alpha | |
| | `creative_writing` | Write stories, poems, essays | — | |
| | `summarization` | Summarize text | file_reader | |
| | `translation` | Translate between languages | translator | |
| | `information_retrieval` | Answer questions, explain topics | knowledge_base | |
| | `data_analysis` | Analyze data, create charts | code_interpreter, data_visualizer | |
| | `web_search` | Search the web for current info | web_browser, search_engine | |
| | `general_chat` | Casual conversation | — | |
|
|
| ## 🔍 Use Cases |
|
|
| - **LLM routing** — route queries to specialized models or tools |
| - **Agent frameworks** — decide which tool to invoke |
| - **Cost optimization** — use cheap models for simple intents, expensive for complex |
| - **Latency optimization** — skip heavy pipelines for general chat |
|
|
| ## ⚠️ Limitations |
|
|
| - English only |
| - 10 fixed intent categories |
|
|
