Add new model configurations and enforce max response length in prompts

This commit is contained in:
Simone Garau
2025-10-27 18:11:42 +01:00
parent 7098e05e44
commit 0dc90eaf33
4 changed files with 11 additions and 4 deletions

View File

@@ -28,6 +28,10 @@ models:
label: Qwen 3 (4B)
- name: qwen3:1.7b
label: Qwen 3 (1.7B)
- name: qwen3:32b
label: Qwen 3 (32B)
- name: phi4-mini:3.8b
label: Phi 4 mini (3.8b)
api:
retry_attempts: 3
@@ -40,7 +44,7 @@ api:
agents:
strategy: Conservative
team_model: qwen3:8b
team_leader_model: qwen3:8b
query_analyzer_model: qwen3:8b # query check
report_generation_model: qwen3:8b # ex predictor
team_model: qwen3:32b # the agents
team_leader_model: qwen3:32b # the team leader
query_analyzer_model: qwen3:32b # query check
report_generation_model: qwen3:32b # ex predictor

View File

@@ -54,6 +54,7 @@ Detailed Data:
3. **Always specify the data source** (which API provided the data)
4. **Report data completeness**: If user asks for 30 days but got 7, state this explicitly
5. **Current date context**: Remind that data is as of {{CURRENT_DATE}}
6. **Max response length**: Do not overcome 100 words
**ERROR HANDLING:**
- Tools failed → "Price data unavailable. Error: [details if available]"

View File

@@ -64,6 +64,7 @@ Notable Headlines:
3. **Report data staleness**: If newest article is >3 days old, flag this
4. **Cite sources**: Mention which news APIs provided the data
5. **Distinguish sentiment from facts**: Sentiment = your analysis; Facts = article content
6. **Max response length**: Do not overcome 100 words
**ERROR HANDLING:**
- No articles found → "No relevant news articles found for [QUERY]"

View File

@@ -70,6 +70,7 @@ Sample Posts (representative):
3. **Report data staleness**: If newest post is >2 days old, flag this
4. **Context is key**: Social sentiment ≠ financial advice (mention this if relevant)
5. **Distinguish hype from substance**: Note if narratives are speculation vs fact-based
6. **Max response length**: Do not overcome 100 words
**ERROR HANDLING:**
- No posts found → "No relevant social discussions found for [QUERY]"