Refactoring some LLM handlers

trustgraph-bedrock/trustgraph/model/text_completion/bedrock/llm.py

@@ -6,22 +6,14 @@ Input is prompt, output is response. Mistral is default.
 
 import boto3
 import json
-from prometheus_client import Histogram
 import os
 import enum
 
-from .... schema import TextCompletionRequest, TextCompletionResponse, Error
-from .... schema import text_completion_request_queue
-from .... schema import text_completion_response_queue
-from .... log_level import LogLevel
-from .... base import ConsumerProducer
 from .... exceptions import TooManyRequests
+from .... base import LlmService, LlmResult
 
-module = "text-completion"
+default_ident = "text-completion"
 
-default_input_queue = text_completion_request_queue
-default_output_queue = text_completion_response_queue
-default_subscriber = module
 default_model = 'mistral.mistral-large-2407-v1:0'
 default_temperature = 0.0
 default_max_output = 2048
@@ -149,16 +141,12 @@ class Cohere(ModelHandler):
 
 Default=Mistral
 
-class Processor(ConsumerProducer):
+class Processor(LlmService):
 
     def __init__(self, **params):
 
         print(params)
     
-        input_queue = params.get("input_queue", default_input_queue)
-        output_queue = params.get("output_queue", default_output_queue)
-        subscriber = params.get("subscriber", default_subscriber)
-
         model = params.get("model", default_model)
         temperature = params.get("temperature", default_temperature)
         max_output = params.get("max_output", default_max_output)
@@ -185,30 +173,12 @@ class Processor(ConsumerProducer):
 
         super(Processor, self).__init__(
             **params | {
-                "input_queue": input_queue,
-                "output_queue": output_queue,
-                "subscriber": subscriber,
-                "input_schema": TextCompletionRequest,
-                "output_schema": TextCompletionResponse,
                 "model": model,
                 "temperature": temperature,
                 "max_output": max_output,
             }
         )
 
-        if not hasattr(__class__, "text_completion_metric"):
-            __class__.text_completion_metric = Histogram(
-                'text_completion_duration',
-                'Text completion duration (seconds)',
-                buckets=[
-                    0.25, 0.5, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0,
-                    8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0,
-                    17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0,
-                    30.0, 35.0, 40.0, 45.0, 50.0, 60.0, 80.0, 100.0,
-                    120.0
-                ]
-            )
-
         self.model = model
         self.temperature = temperature
         self.max_output = max_output
@@ -257,30 +227,21 @@ class Processor(ConsumerProducer):
 
         return Default
                         
-    async def handle(self, msg):
+    async def generate_content(self, system, prompt):
-
-        v = msg.value()
-
-        # Sender-produced ID
-
-        id = msg.properties()["id"]
-
-        print(f"Handling prompt {id}...", flush=True)
 
         try:
 
-            promptbody = self.variant.encode_request(v.system, v.prompt)
+            promptbody = self.variant.encode_request(system, prompt)
 
             accept = 'application/json'
             contentType = 'application/json'
 
-            with __class__.text_completion_metric.time():
+            response = self.bedrock.invoke_model(
-                response = self.bedrock.invoke_model(
+                body=promptbody,
-                    body=promptbody,
+                modelId=self.model,
-                    modelId=self.model,
+                accept=accept,
-                    accept=accept,
+                contentType=contentType
-                    contentType=contentType
+            )
-                )
 
             # Response structure decode
             outputtext = self.variant.decode_response(response)
@@ -293,18 +254,14 @@ class Processor(ConsumerProducer):
             print(f"Input Tokens: {inputtokens}", flush=True)
             print(f"Output Tokens: {outputtokens}", flush=True)
 
-            print("Send response...", flush=True)
+            resp = LlmResult(
-            r = TextCompletionResponse(
+                text = outputtext,
-                error=None,
+                in_token = inputtokens,
-                response=outputtext,
+                out_token = outputtokens,
-                in_token=inputtokens,
+                model = self.model
-                out_token=outputtokens,
-                model=str(self.model),
             )
 
-            await self.send(r, properties={"id": id})
+            return resp
-
-            print("Done.", flush=True)
 
         except self.bedrock.exceptions.ThrottlingException as e:
 
@@ -319,31 +276,12 @@ class Processor(ConsumerProducer):
 
             print(type(e))
             print(f"Exception: {e}")
-
+            raise e
-            print("Send error response...", flush=True)
-
-            r = TextCompletionResponse(
-                error=Error(
-                    type = "llm-error",
-                    message = str(e),
-                ),
-                response=None,
-                in_token=None,
-                out_token=None,
-                model=None,
-            )
-
-            await self.send(r, properties={"id": id})
-
-            self.consumer.acknowledge(msg)
 
     @staticmethod
     def add_args(parser):
 
-        ConsumerProducer.add_args(
+        LlmService.add_args(parser)
-            parser, default_input_queue, default_subscriber,
-            default_output_queue,
-        )
 
         parser.add_argument(
             '-m', '--model',
@@ -391,5 +329,4 @@ class Processor(ConsumerProducer):
 
 def run():
 
-    Processor.launch(module, __doc__)
+    Processor.launch(default_ident, __doc__)
-