Debate fix

README.md

@@ -126,6 +126,7 @@ challenge_final_results: bool = False
 judge_intervention: Optional[str] = None
 judge_metric: Optional[str] = None
 judge_endpoint_url: Optional[str] = None
+judge_model_name: Optional[str] = None
 judge_api_key: str = "-"
 judge_always_intervene: bool = False
 ```

mallm/coordinator.py

@@ -13,8 +13,8 @@
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.judge import Judge
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
-from mallm.discourse_policy.policy import DiscoursePolicy
+from mallm.decision_protocols.protocol import DecisionProtocol
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
 from mallm.models.Chat import Chat
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.models.discussion.SimpleResponseGenerator import SimpleResponseGenerator
@@ -273,7 +273,7 @@ def discuss(
             raise Exception(
                 f"No valid discourse policy for paradigm {config.discussion_paradigm}"
             )
-        policy: DiscoursePolicy = DISCUSSION_PARADIGMS[config.discussion_paradigm]()
+        policy: DiscussionParadigm = DISCUSSION_PARADIGMS[config.discussion_paradigm]()
 
         logger.info(
             f"""Starting discussion with coordinator {self.id}...

mallm/decision_protocol/approval_voting.py → mallm/decision_protocols/approval_voting.py

@@ -3,7 +3,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration

mallm/decision_protocol/consensus.py → mallm/decision_protocols/consensus.py

@@ -2,7 +2,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration
 from mallm.utils.types import Agreement, VotingResult, VotingResultList, WorkerFunctions

mallm/decision_protocol/consensus_voting.py → mallm/decision_protocols/consensus_voting.py

@@ -5,7 +5,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration

mallm/decision_protocol/cumulative_voting.py → mallm/decision_protocols/cumulative_voting.py

@@ -3,7 +3,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration

mallm/decision_protocol/summary.py → mallm/decision_protocols/judge.py

@@ -2,7 +2,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration
@@ -11,9 +11,9 @@
 logger = logging.getLogger("mallm")
 
 
-class Summary(DecisionProtocol):
+class Judge(DecisionProtocol):
     """
-    The Summary decision protocol creates a summary of all answers after a certain number of turns.
+    The Judge decision protocol creates a summary of all answers after a certain number of turns.
     """
 
     _name = "summary"

mallm/decision_protocol/ranked_voting.py → mallm/decision_protocols/ranked_voting.py

@@ -2,7 +2,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration

mallm/decision_protocol/simple_voting.py → mallm/decision_protocols/simple_voting.py

@@ -3,7 +3,7 @@
 from typing import Any, Optional
 
 from mallm.agents.panelist import Panelist
-from mallm.decision_protocol.protocol import DecisionProtocol
+from mallm.decision_protocols.protocol import DecisionProtocol
 from mallm.models.discussion.ResponseGenerator import ResponseGenerator
 from mallm.utils.config import Config
 from mallm.utils.enums import DecisionAlteration

mallm/discourse_policy/collective_refinement.py → mallm/discussion_paradigms/collective_refinement.py

@@ -8,7 +8,7 @@
 
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.panelist import Panelist
-from mallm.discourse_policy.policy import DiscoursePolicy
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
 from mallm.utils.types import Agreement, Memory, TemplateFilling, VotingResultList
 
 if TYPE_CHECKING:
@@ -18,7 +18,7 @@
 logger = logging.getLogger("mallm")
 
 
-class CollectiveRefinement(DiscoursePolicy):
+class CollectiveRefinement(DiscussionParadigm):
     """
     A discussion protocol where agents improve their answers through multiple rounds of feedback.
 

mallm/discourse_policy/debate.py → mallm/discussion_paradigms/debate.py

@@ -8,7 +8,7 @@
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.judge import Judge
 from mallm.agents.panelist import Panelist
-from mallm.discourse_policy.policy import DiscoursePolicy
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
 from mallm.utils.types import Agreement, TemplateFilling, VotingResultList
 
 if TYPE_CHECKING:
@@ -18,7 +18,7 @@
 logger = logging.getLogger("mallm")
 
 
-class DiscourseDebate(DiscoursePolicy):
+class DiscussionDebate(DiscussionParadigm):
     def draft_proposer_call(
         self,
         draft_proposer: DraftProposer,
@@ -120,7 +120,8 @@ def discuss(
                 logger.debug(
                     f"Discussion {coordinator.id} goes into debate round: {r!s}"
                 )
-                debate_agreements: list[Agreement] = []
+                if r == 0:
+                    debate_agreements = self.agreements
                 for i, a in enumerate(
                     coordinator.agents[1:]
                 ):  # similar to relay paradigm

mallm/discourse_policy/memory.py → mallm/discussion_paradigms/memory.py

@@ -5,15 +5,15 @@
 
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.panelist import Panelist
-from mallm.discourse_policy.policy import DiscoursePolicy
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
 from mallm.utils.types import TemplateFilling
 
 if TYPE_CHECKING:
     from mallm.coordinator import Coordinator
 logger = logging.getLogger("mallm")
 
 
-class DiscourseMemory(DiscoursePolicy):
+class DiscussionMemory(DiscussionParadigm):
     def __init__(self) -> None:
         super().__init__(
             """Paradigm: Memory

mallm/discourse_policy/policy.py → mallm/discussion_paradigms/paradigm.py

@@ -19,7 +19,7 @@
 logger = logging.getLogger("mallm")
 
 
-class DiscoursePolicy(ABC):
+class DiscussionParadigm(ABC):
     def __init__(self, paradigm_str: str = "") -> None:
         self.paradigm_str = paradigm_str
         self.decision = False

mallm/discourse_policy/relay.py → mallm/discussion_paradigms/relay.py

@@ -5,15 +5,15 @@
 
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.panelist import Panelist
-from mallm.discourse_policy.policy import DiscoursePolicy
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
 from mallm.utils.types import TemplateFilling
 
 if TYPE_CHECKING:
     from mallm.coordinator import Coordinator
 logger = logging.getLogger("mallm")
 
 
-class DiscourseRelay(DiscoursePolicy):
+class DiscussionRelay(DiscussionParadigm):
     def __init__(self) -> None:
         super().__init__(
             """Paradigm: Relay

mallm/discourse_policy/report.py → mallm/discussion_paradigms/report.py

@@ -5,15 +5,15 @@
 
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.panelist import Panelist
-from mallm.discourse_policy.policy import DiscoursePolicy
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
 from mallm.utils.types import TemplateFilling
 
 if TYPE_CHECKING:
     from mallm.coordinator import Coordinator
 logger = logging.getLogger("mallm")
 
 
-class DiscourseReport(DiscoursePolicy):
+class DiscussionReport(DiscussionParadigm):
     def __init__(self) -> None:
         super().__init__(
             """Paradigm: Report

mallm/utils/dicts.py

@@ -1,22 +1,22 @@
-from mallm.decision_protocol.approval_voting import ApprovalVoting
-from mallm.decision_protocol.consensus import (
+from mallm.decision_protocols.approval_voting import ApprovalVoting
+from mallm.decision_protocols.consensus import (
     HybridMajorityConsensus,
     MajorityConsensus,
     SupermajorityConsensus,
     UnanimityConsensus,
 )
-from mallm.decision_protocol.consensus_voting import ConsensusVoting
-from mallm.decision_protocol.cumulative_voting import CumulativeVoting
-from mallm.decision_protocol.protocol import DecisionProtocol
-from mallm.decision_protocol.ranked_voting import RankedVoting
-from mallm.decision_protocol.simple_voting import SimpleVoting
-from mallm.decision_protocol.summary import Summary
-from mallm.discourse_policy.collective_refinement import CollectiveRefinement
-from mallm.discourse_policy.debate import DiscourseDebate
-from mallm.discourse_policy.memory import DiscourseMemory
-from mallm.discourse_policy.policy import DiscoursePolicy
-from mallm.discourse_policy.relay import DiscourseRelay
-from mallm.discourse_policy.report import DiscourseReport
+from mallm.decision_protocols.consensus_voting import ConsensusVoting
+from mallm.decision_protocols.cumulative_voting import CumulativeVoting
+from mallm.decision_protocols.protocol import DecisionProtocol
+from mallm.decision_protocols.ranked_voting import RankedVoting
+from mallm.decision_protocols.simple_voting import SimpleVoting
+from mallm.decision_protocols.judge import Judge
+from mallm.discussion_paradigms.collective_refinement import CollectiveRefinement
+from mallm.discussion_paradigms.debate import DiscussionDebate
+from mallm.discussion_paradigms.memory import DiscussionMemory
+from mallm.discussion_paradigms.paradigm import DiscussionParadigm
+from mallm.discussion_paradigms.relay import DiscussionRelay
+from mallm.discussion_paradigms.report import DiscussionReport
 from mallm.models.discussion.CriticalResponseGenerator import CriticalResponseGenerator
 from mallm.models.discussion.FreeTextResponseGenerator import FreeTextResponseGenerator
 from mallm.models.discussion.ReasoningResponseGenerator import (
@@ -43,14 +43,14 @@
     "cumulative_voting": CumulativeVoting,
     "ranked_voting": RankedVoting,
     "consensus_voting": ConsensusVoting,
-    "summary": Summary,
+    "summary": Judge,
 }
 
-DISCUSSION_PARADIGMS: dict[str, type[DiscoursePolicy]] = {
-    "memory": DiscourseMemory,
-    "report": DiscourseReport,
-    "relay": DiscourseRelay,
-    "debate": DiscourseDebate,
+DISCUSSION_PARADIGMS: dict[str, type[DiscussionParadigm]] = {
+    "memory": DiscussionMemory,
+    "report": DiscussionReport,
+    "relay": DiscussionRelay,
+    "debate": DiscussionDebate,
     "collective_refinement": CollectiveRefinement,
 }
 

test/decision_protocol/test_majority_consensus.py

@@ -1,7 +1,7 @@
 from mallm.agents.draftProposer import DraftProposer
 from mallm.agents.panelist import Panelist
 from mallm.coordinator import Coordinator
-from mallm.decision_protocol.consensus import (
+from mallm.decision_protocols.consensus import (
     HybridMajorityConsensus,
     UnanimityConsensus,
 )