Update README.md
Browse files
README.md
CHANGED
|
@@ -193,3 +193,14 @@ pip install -r requirements.txt
|
|
| 193 |
export HF_TOKEN=your_token_here # optional, for chat
|
| 194 |
python app.py
|
| 195 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 193 |
export HF_TOKEN=your_token_here # optional, for chat
|
| 194 |
python app.py
|
| 195 |
```
|
| 196 |
+
|
| 197 |
+
## Chat with a Groq API key (optional)
|
| 198 |
+
|
| 199 |
+
1. Create a key at console.groq.com and add it to the Space as a **secret** named `GROQ_API_KEY`
|
| 200 |
+
(Settings -> Variables and secrets). No training or model upload is needed.
|
| 201 |
+
2. The app tries Groq models first (`openai/gpt-oss-120b`, `openai/gpt-oss-20b`, `llama-3.3-70b-versatile`,
|
| 202 |
+
`llama-3.1-8b-instant`; override with a comma-separated `GROQ_MODELS` variable), then the Hugging Face
|
| 203 |
+
models if `HF_TOKEN` is set, then (English-only app) the built-in offline assistant.
|
| 204 |
+
3. Groq's free tier is rate-limited: an HTTP 429 just moves the app to the next model. Check your own limits at
|
| 205 |
+
console.groq.com/settings/limits. Chat text is sent to Groq, so mention that in any public demo.
|
| 206 |
+
4. `LOCAL_CHAT_MODEL_ID` / the fine-tuned model is no longer required; leave it unset.
|