update

src/interface.jl

@@ -2,7 +2,7 @@ module interface
     
 export addNewMessage, conversation, decisionMaker, progressValueEstimator, isterminal
 
-using JSON3, DataStructures, Dates, UUIDs, HTTP, Random, MQTTClient
+using JSON3, DataStructures, Dates, UUIDs, HTTP, Random, MQTTClient, PrettyPrinting
 using GeneralUtils
 using ..type, ..util, ..llmfunction, ..mcts
 
@@ -98,6 +98,16 @@ function decisionMaker(a::T1, state::T2)::Dict{Symbol, Any} where {T1<:agent, T2
   # (trajectories)
   # """
 
+  responseformat =
+  """
+  You should only respond in JSON format as describe below:
+  {
+    "Thought": "your reasoning",
+    "Action": {"name": "action to take", "input": "Action input"},
+    "Observation": "result of the action"
+  }
+  """
+
   _prompt = 
   """
   You are a helpful sommelier working for a wine store. 
@@ -119,12 +129,7 @@ function decisionMaker(a::T1, state::T2)::Dict{Symbol, Any} where {T1<:agent, T2
   2) chatbox[text], which you can use to interact with the user.
   3) recommendation[answer], which returns your wine reccommendation to the user. 
   
-  You should only respond in JSON format as describe below:
-  {
-    "Thought": "your reasoning",
-    "Action": {"name": "action to take", "input": "Action input"},
-    "Observation": "result of the action"
-  }
+  $responseformat
   
   Here are some examples:
   {
@@ -164,12 +169,18 @@ function decisionMaker(a::T1, state::T2)::Dict{Symbol, Any} where {T1<:agent, T2
     :msgMeta=> msgMeta,
     :payload=> Dict(
       :text=> prompt,
+      :kwargs=> Dict(
+        :max_tokens=> 512,
+        :stop=> ["<|eot_id|>"],
+      )
     )
   )
-  @show outgoingMsg
+
   _response = GeneralUtils.sendReceiveMqttMsg(outgoingMsg)
-  thoughtJsonStr = _response[:response][:text]
+  _thoughtJsonStr = _response[:response][:text]
+  thoughtJsonStr = jsoncorrection(a, _thoughtJsonStr, "")
   thoughtDict = copy(JSON3.read(thoughtJsonStr)) 
+  pprint(thoughtDict)
   return thoughtDict
 end
 
@@ -196,6 +207,17 @@ julia>
 # Signature
 """
 function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where {T1<:agent, T2<:AbstractDict}
+  responseformat =
+  """
+  You should only respond in JSON format as describe below:
+  {
+    "Thought_1": "reasoning 1",
+    "Action_1": {"name": "action to take", "input": "Action input"},
+    "Observation_1": "result of the action",
+    "Evaluation_1": {"evaluation": "your evaluation", "score": your evaluation score}
+  }
+  """
+  
   _prompt = 
   """
   Analyze the trajectories of a solution to a question answering task. The trajectories are 
@@ -211,13 +233,7 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
   yet. Do not generate additional thoughts or actions. Then ending with the correctness score s 
   where s is an integer from 1 to 10.
 
-  You should only respond in JSON format as describe below:
-  {
-    "Thought_1": "reasoning 1",
-    "Action_1": {"name": "action to take", "input": "Action input"},
-    "Observation_1": "result of the action",
-    "Evaluation_1": {"evaluation": "your evaluation", "score": your evaluation score}
-  }
+  $responseformat
 
   Here are some examples:
   {
@@ -232,6 +248,7 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
                       "score": 10}
   }
 
+  Let's begin!:
   $(JSON3.write(state[:thoughtHistory]))
   """
 
@@ -254,7 +271,8 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
   )
 
   _response = GeneralUtils.sendReceiveMqttMsg(outgoingMsg)
-  thoughtJsonStr = _response[:response][:text]
+  _thoughtJsonStr = _response[:response][:text]
+  thoughtJsonStr = jsoncorrection(a, _thoughtJsonStr, responseformat)
   thoughtDict = copy(JSON3.read(thoughtJsonStr)) 
   latestEvaluationKey, _ = 
     GeneralUtils.findHighestIndexKey(thoughtDict, "Evaluation")