Merge pull request #463 from solliancenet/jdh-full-prompt

Capture and persist the full prompt with context
solliancenet · Jan 14, 2024 · e60ad0e · e60ad0e
2 parents 100fc0e + a7170f3
commit e60ad0e
Show file tree

Hide file tree

Showing 9 changed files with 19 additions and 5 deletions.
diff --git a/src/dotnet/AgentFactory/Agents/DefaultAgent.cs b/src/dotnet/AgentFactory/Agents/DefaultAgent.cs
@@ -154,6 +154,7 @@ public override async Task<CompletionResponse> GetCompletion(CompletionRequest c
             {
                 Completion = result.Completion!,
                 UserPrompt = completionRequest.UserPrompt!,
+                FullPrompt = result.FullPrompt,
                 PromptTemplate = result.PromptTemplate,
                 AgentName = result.AgentName,
                 PromptTokens = result.PromptTokens,

diff --git a/src/dotnet/AgentFactory/Models/Orchestration/LLMOrchestrationCompletionResponse.cs b/src/dotnet/AgentFactory/Models/Orchestration/LLMOrchestrationCompletionResponse.cs
@@ -19,6 +19,12 @@ public class LLMOrchestrationCompletionResponse
         [JsonProperty("user_prompt")]
         public string? UserPrompt { get; set; }
 
+        /// <summary>
+        /// The full prompt composed by the LLM.
+        /// </summary>
+        [JsonProperty("full_prompt")]
+        public string? FullPrompt { get; set; }
+
         /// <summary>
         /// The prompt template used by the LLM.
         /// </summary>

diff --git a/src/dotnet/AgentFactory/Services/LangChainService.cs b/src/dotnet/AgentFactory/Services/LangChainService.cs
@@ -66,6 +66,7 @@ public async Task<LLMOrchestrationCompletionResponse> GetCompletion(LLMOrchestra
                 {
                     Completion = completionResponse!.Completion,
                     UserPrompt = completionResponse.UserPrompt,
+                    FullPrompt = completionResponse.FullPrompt,
                     PromptTemplate = request.Agent?.PromptPrefix,
                     AgentName = request.Agent?.Name,
                     PromptTokens = completionResponse.PromptTokens,

diff --git a/src/dotnet/Common/Models/Orchestration/CompletionResponse.cs b/src/dotnet/Common/Models/Orchestration/CompletionResponse.cs
@@ -20,6 +20,12 @@ public class CompletionResponse
     [JsonProperty("user_prompt")]
     public string UserPrompt { get; set; }
 
+    /// <summary>
+    /// The full prompt composed by the LLM.
+    /// </summary>
+    [JsonProperty("full_prompt")]
+    public string? FullPrompt { get; set; }
+
     /// <summary>
     /// The prompt template used by the LLM.
     /// </summary>

diff --git a/src/dotnet/Core/Services/CoreService.cs b/src/dotnet/Core/Services/CoreService.cs
@@ -130,7 +130,7 @@ public async Task<Completion> GetChatCompletionAsync(string? sessionId, string u
             var promptMessage = new Message(sessionId, nameof(Participants.User), result.PromptTokens, userPrompt, result.UserPromptEmbedding, null, upn, _callContext.CurrentUserIdentity?.Name);
             var completionMessage = new Message(sessionId, nameof(Participants.Assistant), result.CompletionTokens, result.Completion, null, null, upn, result.AgentName);
             var completionPromptText =
-                $"User prompt: {result.UserPrompt}{Environment.NewLine}Agent: {result.AgentName}{Environment.NewLine}Prompt template: {result.PromptTemplate}";
+                $"User prompt: {result.UserPrompt}{Environment.NewLine}Agent: {result.AgentName}{Environment.NewLine}Prompt template: {(!string.IsNullOrWhiteSpace(result.FullPrompt) ? result.FullPrompt : result.PromptTemplate)}";
             var completionPrompt = new CompletionPrompt(sessionId, completionMessage.Id, completionPromptText);
             completionMessage.CompletionPromptId = completionPrompt.Id;
 

diff --git a/src/python/PythonSDK/foundationallm/langchain/agents/blob_storage_agent.py b/src/python/PythonSDK/foundationallm/langchain/agents/blob_storage_agent.py
@@ -136,7 +136,7 @@ def run(self, prompt: str) -> CompletionResponse:
             return CompletionResponse(
                 completion = rag_chain.invoke(prompt),
                 user_prompt = prompt,
-                final_prompt = self.full_prompt.text,
+                full_prompt = self.full_prompt.text,
                 completion_tokens = cb.completion_tokens,
                 prompt_tokens = cb.prompt_tokens,
                 total_tokens = cb.total_tokens,

diff --git a/src/python/PythonSDK/foundationallm/langchain/agents/csv_agent.py b/src/python/PythonSDK/foundationallm/langchain/agents/csv_agent.py
@@ -165,7 +165,7 @@ def run(self, prompt: str) -> CompletionResponse:
             return CompletionResponse(
                 completion = self.agent.run(prompt),
                 user_prompt = prompt,
-                final_prompt = self.prompt_template,
+                full_prompt = self.prompt_template,
                 completion_tokens = cb.completion_tokens,
                 prompt_tokens = cb.prompt_tokens,
                 total_tokens = cb.total_tokens,

diff --git a/src/python/PythonSDK/foundationallm/langchain/agents/search_service_agent.py b/src/python/PythonSDK/foundationallm/langchain/agents/search_service_agent.py
@@ -114,7 +114,7 @@ def run(self, prompt: str) -> CompletionResponse:
             return CompletionResponse(
                 completion = completion,
                 user_prompt = prompt,
-                final_prompt = self.full_prompt.text,
+                full_prompt = self.full_prompt.text,
                 completion_tokens = cb.completion_tokens,
                 prompt_tokens = cb.prompt_tokens,
                 total_tokens = cb.total_tokens,

diff --git a/src/python/PythonSDK/foundationallm/models/orchestration/completion_response.py b/src/python/PythonSDK/foundationallm/models/orchestration/completion_response.py
@@ -6,7 +6,7 @@ class CompletionResponse(BaseModel):
     Response from a language model.
     """
     user_prompt: str
-    final_prompt: Optional[str] = None
+    full_prompt: Optional[str] = None
     completion: Union[str, set, List[str]]
     user_prompt_embedding: Optional[List[float]] = []
     prompt_tokens: int = 0