nocmem: qa-style extraction prompt, multi-cue variants, claude history importer

- Switch extraction prompt to qa-style (80% recall vs 60% baseline)
- Semicolon-separated cues in extraction become paraphrase variants
- Add import_claude.py to bulk-import Claude Code conversation history
- Fix LLM model name in systemd service, add logging basicConfig
This commit is contained in:
Fam Zheng
2026-04-11 22:57:17 +01:00
parent daed6c9d37
commit 35cafbd4ca
4 changed files with 423 additions and 14 deletions

View File

@@ -5,12 +5,12 @@ After=network.target
[Service]
Type=simple
WorkingDirectory=/data/src/noc/mem
ExecStart=/home/fam/.local/bin/uv run uvicorn server:app --host 0.0.0.0 --port 9820
ExecStart=/home/fam/.local/bin/uv run uvicorn server:app --host 0.0.0.0 --port 9820 --log-level info
Restart=on-failure
RestartSec=5
Environment=NOCMEM_LLM_ENDPOINT=http://100.84.7.49:8000/v1
Environment=NOCMEM_LLM_MODEL=QuantTrio/gemma-4-31B-it-AWQ
Environment=NOCMEM_LLM_MODEL=gemma4-31b
Environment=NOCMEM_LLM_API_KEY=unused
Environment=NOCMEM_DATA_DIR=/data/src/noc/mem/data
Environment=NOCMEM_DEVICE=cuda