fix

fix

fixfix

Author: Pouyanpi

nemoguardrails/llm/prompts/nemotron_reasoning.yml

@@ -4,12 +4,11 @@ prompts:
   - task: general
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       # by default detailed thinking is off
       # user can turn it on using the system message, not the other way
-      # - type: system
-      #   content: |
-      #     detailed thinking on
       - type: system
         content: |
           {{ general_instructions }}{% if relevant_chunks != None and relevant_chunks != '' %}
@@ -23,6 +22,8 @@ prompts:
   - task: generate_user_intent
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -46,6 +47,8 @@ prompts:
   - task: generate_next_steps
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -66,6 +69,8 @@ prompts:
   - task: generate_bot_message
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -88,11 +93,15 @@ prompts:
       - "{{ history | colang | to_intent_messages_2 }}"
 
     output_parser: "verbose_v1"
+    max_length: 16000
+    mode: "standard"
 
   # Prompt for generating the user intent, next steps and bot message in a single call.
   - task: generate_intent_steps_message
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -123,6 +132,8 @@ prompts:
   - task: generate_value
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -152,6 +163,8 @@ prompts:
   - task: generate_user_intent_from_user_action
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -182,6 +195,8 @@ prompts:
   - task: generate_user_intent_and_bot_action_from_user_action
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -222,6 +237,8 @@ prompts:
   - task: generate_value_from_instruction
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -250,6 +267,8 @@ prompts:
   - task: generate_flow_from_instructions
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -276,6 +295,8 @@ prompts:
   - task: generate_flow_from_name
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -309,6 +330,8 @@ prompts:
   - task: generate_flow_continuation
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |
@@ -340,6 +363,8 @@ prompts:
   - task: generate_flow_continuation_from_flow_nld
     models:
       - nemotron
+      - nvidia/llama-3.1-nemotron
+      - nvidia/llama-3.3-nemotron
     messages:
       - type: system
         content: |

tests/test_nemotron_prompt_modes.py

@@ -137,12 +137,12 @@ def test_prompt_selection_mechanism():
     deepseek_model = DEEPSEEK_MODEL
 
     # Nemotron model -> message-based prompt
-    nemotron_prompt = _get_prompt(task_name, nemotron_model, None, _prompts)
+    nemotron_prompt = _get_prompt(task_name, nemotron_model, "standard", _prompts)
     assert hasattr(nemotron_prompt, "messages")
     assert "nemotron" in nemotron_prompt.models
 
     # DeepSeek model -> content-based prompt
-    deepseek_prompt = _get_prompt(task_name, deepseek_model, None, _prompts)
+    deepseek_prompt = _get_prompt(task_name, deepseek_model, "standard", _prompts)
     assert hasattr(deepseek_prompt, "content")
     assert "deepseek" in deepseek_prompt.models
     assert "nemotron" not in deepseek_prompt.models