--- title: PRISM-Memory colorFrom: blue colorTo: green sdk: gradio sdk_version: 6.12.0 app_file: app.py pinned: false --- # PRISM-Memory Space **Hook:** Turn conversations into durable, searchable memory. This Space is the public demo for the released `PRISM-Memory 7B Adapter`, a LoRA adapter on top of `Qwen/Qwen2.5-7B-Instruct`. It matches the root repo story: - one released model - one extraction behavior - one set of confirmed benchmark results - one interactive turn-by-turn demo - one compact explanation of the synthetic training data and held-out examples Within the GitHub repo, this directory is a first-class public surface. The helper script `scripts/deploy_space.sh` bundles this folder with the release artifacts needed for Hugging Face. ## Inputs The app reads: - `results/release_summary.json` - `results/extraction_examples.json` - `results/try_it_sessions.json` - `docs/release/extraction-skill.md` - `docs/release/datasets.md` When copied into a standalone Hugging Face Space repo, keep those files beside `app.py` and `requirements.txt`, preserving the `docs/release/` and `results/` subdirectories. ## What It Shows 1. The confirmed metrics for the released model 2. A `Try It` flow with bundled multi-turn sessions and pasted transcript input 3. Side-by-side held-out extraction examples against the GPT-4.1 reference 4. A compact description of the synthetic conversation data and supervision 5. The single canonical memory extraction skill to keep The bundled example sessions use precomputed outputs from the released model. Custom pasted sessions use live model inference when the runtime can load the adapter; otherwise the Space falls back to a lightweight contract-preserving preview. ## Local Run ```bash python -m pip install -r requirements.txt python app.py ``` ## From The Repo Root ```bash bash scripts/deploy_space.sh ``` Useful bundled files: - `docs/release/extraction-skill.md` - `docs/release/datasets.md` - `docs/release/extraction-examples.md` - `docs/release/release-results.md` - `results/release_summary.json` - `results/extraction_examples.json` - `results/try_it_sessions.json`