Models¶
Sbobinator uses one required model for transcription and optional models for LLM summary. No downloads during processing.
ASR model (required)¶
| Property | Value |
|---|---|
| Hugging Face name | nvidia/parakeet-tdt-0.6b-v3 |
| Local file | models/parakeet-tdt-0.6b-v3.nemo |
| Size | ~2.5 GB |
| Engine | NVIDIA NeMo |
Uses curl (system certificates). Optional variable: NEMO_CACHE_DIR.
Cloud LLM summary (optional)¶
No model on disk. Requires an API key for the chosen provider.
| Provider | Env (alternative to secrets file) |
|---|---|
| DeepSeek | SBOBINATOR_DEEPSEEK_API_KEY |
| OpenAI | SBOBINATOR_OPENAI_API_KEY |
| Gemini | SBOBINATOR_GEMINI_API_KEY |
| Claude | SBOBINATOR_ANTHROPIC_API_KEY |
| Kimi | SBOBINATOR_KIMI_API_KEY |
UI configuration: /settings/summary
File: data/.secrets/summary_keys.json
Local LLM summary — Qwen (optional)¶
| Property | Value |
|---|---|
| Model | Qwen2.5-3B-Instruct (GGUF Q4) |
| Folder | models/qwen2.5-3b-instruct/ |
| Size | ~2 GB |
| RAM | ≥ 16 GB |
| Engine | llama.cpp (CPU) |
Requires pip install -r requirements/local.txt (includes llama-cpp-python).
Docker¶
At build time: Parakeet ASR only in /models/.
Summary: cloud API via env or data/.secrets/ on the /data volume.
Local Qwen: can be downloaded at runtime in the container with download_summary_llm.py.
Summary table¶
| Component | Path | Download | Network at runtime |
|---|---|---|---|
| Parakeet ASR | models/*.nemo |
download_model.py |
No |
| Cloud summary | — | API key | Yes |
| Local Qwen | models/qwen2.5-3b-instruct/*.gguf |
download_summary_llm.py |
No |
Deprecated¶
| Legacy | Replacement |
|---|---|
IT5 / download_summary_model.py |
Multi-provider LLM |
| LexRank extractive | Removed as product «summary» |
Licenses¶
- Parakeet: NVIDIA / NeMo
- Qwen: Alibaba model license — see Hugging Face
- Sbobinator code: proprietary license Antonio Trento (see
LICENSE)