Rate limit handling (#11)

* Added a rate limit exception
* Reduce request/response timeouts because looks like there are major issues
* Add rate limit exception catch to all consumers
* Version to 0.6.3

trustgraph/model/text_completion/azure/llm.py

@@ -12,6 +12,7 @@ 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
 
 module = ".".join(__name__.split(".")[1:-1])
 
@@ -76,6 +77,10 @@ class Processor(ConsumerProducer):
         }
 
         resp = requests.post(url, data=body, headers=headers)
+
+        if resp.status_code == 429:
+            raise TooManyRequests()
+
         result = resp.json()
 
         message_content = result['choices'][0]['message']['content']

trustgraph/model/text_completion/bedrock/llm.py

@@ -13,6 +13,7 @@ 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
 
 module = ".".join(__name__.split(".")[1:-1])
 
@@ -121,6 +122,8 @@ class Processor(ConsumerProducer):
         accept = 'application/json'
         contentType = 'application/json'
 
+        # FIXME: Consider catching request limits and raise TooManyRequests
+        # See https://boto3.amazonaws.com/v1/documentation/api/latest/guide/retries.html
         response = self.bedrock.invoke_model(body=promptbody, modelId=self.model, accept=accept, contentType=contentType)
         
         # Mistral Response Structure

trustgraph/model/text_completion/claude/llm.py

@@ -57,6 +57,8 @@ class Processor(ConsumerProducer):
         print(f"Handling prompt {id}...", flush=True)
 
         prompt = v.prompt
+
+        # FIXME: Rate limits?
         response = message = self.claude.messages.create(
             model=self.model,
             max_tokens=1000,

trustgraph/model/text_completion/cohere/llm.py

@@ -59,15 +59,16 @@ class Processor(ConsumerProducer):
 
         prompt = v.prompt
 
+        # FIXME: Deal with rate limits?
         output = self.cohere.chat( 
-        model=self.model,
-        message=prompt,
-        preamble = "You are a helpful AI-assistant.",
-        temperature=0.0,
-        chat_history=[],
-        prompt_truncation='auto',
-        connectors=[]
-        ) 
+            model=self.model,
+            message=prompt,
+            preamble = "You are a helpful AI-assistant.",
+            temperature=0.0,
+            chat_history=[],
+            prompt_truncation='auto',
+            connectors=[]
+        )
 
         resp = output.text
         print(resp, flush=True)

trustgraph/model/text_completion/ollama/llm.py

@@ -65,6 +65,8 @@ class Processor(ConsumerProducer):
         print(f"Handling prompt {id}...", flush=True)
 
         prompt = v.prompt
+
+        # FIXME: Rate limits?
         response = self.llm.invoke(prompt)
 
         print("Send response...", flush=True)

trustgraph/model/text_completion/openai/llm.py

@@ -57,6 +57,8 @@ class Processor(ConsumerProducer):
         print(f"Handling prompt {id}...", flush=True)
 
         prompt = v.prompt
+
+        # FIXME: Rate limits
         resp = self.openai.chat.completions.create(
             model=self.model,
             messages=[

trustgraph/model/text_completion/vertexai/llm.py

@@ -26,6 +26,7 @@ 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
 
 module = ".".join(__name__.split(".")[1:-1])
 
@@ -139,9 +140,8 @@ class Processor(ConsumerProducer):
 
         except google.api_core.exceptions.ResourceExhausted:
 
-            print("429, resource busy, sleeping", flush=True)
-            time.sleep(15)
-            self.consumer.negative_acknowledge(msg)
+            # 429 / rate limits case
+            raise TooManyRequests
 
         # Let other exceptions fall through