Spaces:
Running on Zero
Running on Zero
| title: HistAgent Chat | |
| emoji: 🧬 | |
| colorFrom: indigo | |
| colorTo: red | |
| sdk: gradio | |
| sdk_version: 5.44.1 | |
| app_file: app.py | |
| pinned: false | |
| license: other | |
| models: | |
| - Qwen/Qwen3-8B | |
| - Qwen/Qwen3-Embedding-8B | |
| datasets: | |
| - wli13/HistAgent-data | |
| tags: | |
| - histopathology | |
| - spatial-transcriptomics | |
| - biology | |
| - chat | |
| - retrieval | |
| # HistAgent Chat | |
| HistAgent Chat answers questions about selected tissue spots using ranked genes, | |
| cell-state estimates, functional programs and spatial context contained in a | |
| structured evidence card. | |
| The online demonstration uses Qwen3-8B with the HistAgent evidence-card schema, | |
| system policy and evidence constraints. The interface therefore describes the | |
| language-model component as **adapted Qwen3-8B**. It does not expose | |
| chain-of-thought and is intended for research use. | |
| The Atlas Explorer tab searches a balanced public index of 100,000 measured ST | |
| spots using 4,096-dimensional Qwen3-Embedding-8B vectors and cosine similarity. | |
| The demonstration index and checksum manifest are published in the | |
| [HistAgent public evidence-bank release](https://github.com/zipging/HistAgent/releases/tag/public-evidence-bank-v1). | |
| The manuscript analyses use the complete 2.23-million-spot evidence bank. | |