Spaces:
Running
Running
deploy: phase 3 BYOK backend (Dockerfile.hf, FastAPI on 7860)
Browse files- core/agents/router.py +21 -1
- core/agents/synthesizer.py +9 -0
core/agents/router.py
CHANGED
|
@@ -185,7 +185,27 @@ async def call_llm_stream(
|
|
| 185 |
yield token
|
| 186 |
except Exception as exc:
|
| 187 |
logger.error("call_llm_stream_failed", error=str(exc))
|
| 188 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 189 |
|
| 190 |
|
| 191 |
def _get_routing_prompt(query: str) -> str:
|
|
|
|
| 185 |
yield token
|
| 186 |
except Exception as exc:
|
| 187 |
logger.error("call_llm_stream_failed", error=str(exc))
|
| 188 |
+
# Map common failure modes to actionable user-facing copy. Bare
|
| 189 |
+
# "[Error generating response]" left visitors guessing whether
|
| 190 |
+
# the bug was in the corpus, the prompt, or the LLM provider.
|
| 191 |
+
err_text = str(exc).lower()
|
| 192 |
+
if "429" in err_text or "rate" in err_text or "quota" in err_text:
|
| 193 |
+
yield (
|
| 194 |
+
"Sorry — the demo's shared Groq quota is exhausted for "
|
| 195 |
+
"this hour. Paste your own Groq / OpenAI / Anthropic key "
|
| 196 |
+
"via the 🔑 button at the top of the page for an "
|
| 197 |
+
"unthrottled bucket."
|
| 198 |
+
)
|
| 199 |
+
elif "timeout" in err_text or "connect" in err_text:
|
| 200 |
+
yield (
|
| 201 |
+
"The LLM provider didn't respond in time. Try again in a "
|
| 202 |
+
"few seconds, or paste your own key via the 🔑 button."
|
| 203 |
+
)
|
| 204 |
+
else:
|
| 205 |
+
yield (
|
| 206 |
+
"I couldn't reach the LLM provider this turn. Try again, "
|
| 207 |
+
"or paste your own key via the 🔑 button at the top."
|
| 208 |
+
)
|
| 209 |
|
| 210 |
|
| 211 |
def _get_routing_prompt(query: str) -> str:
|
core/agents/synthesizer.py
CHANGED
|
@@ -352,6 +352,13 @@ def _compute_synthesis_confidence(
|
|
| 352 |
def _add_disclaimers(response: str, sensitivity_level: str) -> str:
|
| 353 |
"""Add disclaimers to the response based on sensitivity level.
|
| 354 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 355 |
Args:
|
| 356 |
response: The generated response text.
|
| 357 |
sensitivity_level: The sensitivity level of the documents used.
|
|
@@ -359,6 +366,8 @@ def _add_disclaimers(response: str, sensitivity_level: str) -> str:
|
|
| 359 |
Returns:
|
| 360 |
Response text with appropriate disclaimers appended.
|
| 361 |
"""
|
|
|
|
|
|
|
| 362 |
if sensitivity_level == "high":
|
| 363 |
disclaimer = (
|
| 364 |
"\n\n---\n"
|
|
|
|
| 352 |
def _add_disclaimers(response: str, sensitivity_level: str) -> str:
|
| 353 |
"""Add disclaimers to the response based on sensitivity level.
|
| 354 |
|
| 355 |
+
In BYOK demo mode the UI already shows a ``sensitivity: <level>``
|
| 356 |
+
badge under every answer, so appending a verbal disclaimer doubles
|
| 357 |
+
the signal and clutters short answers ("Use MFA. -- Note: this
|
| 358 |
+
response references documents with moderate sensitivity..."). The
|
| 359 |
+
badge is the canonical surface for the BYOK demo; the verbal
|
| 360 |
+
disclaimer stays on for production deploys.
|
| 361 |
+
|
| 362 |
Args:
|
| 363 |
response: The generated response text.
|
| 364 |
sensitivity_level: The sensitivity level of the documents used.
|
|
|
|
| 366 |
Returns:
|
| 367 |
Response text with appropriate disclaimers appended.
|
| 368 |
"""
|
| 369 |
+
if settings.byok_mode:
|
| 370 |
+
return response
|
| 371 |
if sensitivity_level == "high":
|
| 372 |
disclaimer = (
|
| 373 |
"\n\n---\n"
|