apunkt/trustgraph
1
0
Fork 0
mirror of https://github.com/trustgraph-ai/trustgraph.git synced 2026-07-22 03:31:02 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

Refactoring some LLM handlers

Browse source
This commit is contained in:
Cyber MacGeddon 2025-04-25 16:21:46 +01:00
parent 099018e103
commit 8ab0a9ace6
5 changed files with 91 additions and 259 deletions
Download patch file Download diff file Expand all files Collapse all files

5
trustgraph-base/trustgraph/base/llm_service.py
View file

@ -13,6 +13,11 @@ from .. base import FlowProcessor, ConsumerSpec, ProducerSpec
default_ident = "text-completion" default_ident = "text-completion"
class LlmResult: class LlmResult:
def __init__(self, text=None, in_token=None, out_token=None, model=None):
self.text = text
self.in_token = in_token
self.out_token = out_token
self.model = model
__slots__ = ["text", "in_token", "out_token", "model"] __slots__ = ["text", "in_token", "out_token", "model"]
class LlmService(FlowProcessor): class LlmService(FlowProcessor):

87
trustgraph-flow/trustgraph/model/text_completion/azure/llm.py
View file

@ -9,31 +9,21 @@ import json
from prometheus_client import Histogram from prometheus_client import Histogram
import os import os
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 .... 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_temperature = 0.0 default_temperature = 0.0
default_max_output = 4192 default_max_output = 4192
default_model = "AzureAI" default_model = "AzureAI"
default_endpoint = os.getenv("AZURE_ENDPOINT") default_endpoint = os.getenv("AZURE_ENDPOINT")
default_token = os.getenv("AZURE_TOKEN") default_token = os.getenv("AZURE_TOKEN")
class Processor(ConsumerProducer): class Processor(LlmService):
def __init__(self, **params): def __init__(self, **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)
endpoint = params.get("endpoint", default_endpoint) endpoint = params.get("endpoint", default_endpoint)
token = params.get("token", default_token) token = params.get("token", default_token)
temperature = params.get("temperature", default_temperature) temperature = params.get("temperature", default_temperature)
@ -48,30 +38,13 @@ class Processor(ConsumerProducer):
super(Processor, self).__init__( super(Processor, self).__init__(
**params | { **params | {
"input_queue": input_queue, "endpoint": endpoint,
"output_queue": output_queue,
"subscriber": subscriber,
"input_schema": TextCompletionRequest,
"output_schema": TextCompletionResponse,
"temperature": temperature, "temperature": temperature,
"max_output": max_output, "max_output": max_output,
"model": model, "model": model,
} }
) )
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.endpoint = endpoint self.endpoint = endpoint
self.token = token self.token = token
self.temperature = temperature self.temperature = temperature
@ -123,25 +96,16 @@ class Processor(ConsumerProducer):
return result return result
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: try:
prompt = self.build_prompt( prompt = self.build_prompt(
v.system, system,
v.prompt prompt
) )
with __class__.text_completion_metric.time(): response = self.call_llm(prompt)
response = self.call_llm(prompt)
resp = response['choices'][0]['message']['content'] resp = response['choices'][0]['message']['content']
inputtokens = response['usage']['prompt_tokens'] inputtokens = response['usage']['prompt_tokens']
@ -153,8 +117,14 @@ class Processor(ConsumerProducer):
print("Send response...", flush=True) print("Send response...", flush=True)
r = TextCompletionResponse(response=resp, error=None, in_token=inputtokens, out_token=outputtokens, model=self.model) resp = LlmResult(
await self.send(r, properties={"id": id}) text = resp,
in_token = inputtokens,
out_token = outputtokens,
model = self.model
)
return resp
except TooManyRequests: except TooManyRequests:
@ -168,33 +138,14 @@ class Processor(ConsumerProducer):
# Apart from rate limits, treat all exceptions as unrecoverable # Apart from rate limits, treat all exceptions as unrecoverable
print(f"Exception: {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)
print("Done.", flush=True) print("Done.", flush=True)
@staticmethod @staticmethod
def add_args(parser): def add_args(parser):
ConsumerProducer.add_args( LlmService.add_args(parser)
parser, default_input_queue, default_subscriber,
default_output_queue,
)
parser.add_argument( parser.add_argument(
'-e', '--endpoint', '-e', '--endpoint',
@ -224,4 +175,4 @@ class Processor(ConsumerProducer):
def run(): def run():
Processor.launch(module, __doc__) Processor.launch(default_ident, __doc__)

123
trustgraph-flow/trustgraph/model/text_completion/azure_openai/llm.py
View file

@ -9,18 +9,11 @@ from prometheus_client import Histogram
from openai import AzureOpenAI, RateLimitError from openai import AzureOpenAI, RateLimitError
import os import os
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 .... 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_temperature = 0.0 default_temperature = 0.0
default_max_output = 4192 default_max_output = 4192
default_api = "2024-12-01-preview" default_api = "2024-12-01-preview"
@ -28,13 +21,10 @@ default_endpoint = os.getenv("AZURE_ENDPOINT", None)
default_token = os.getenv("AZURE_TOKEN", None) default_token = os.getenv("AZURE_TOKEN", None)
default_model = os.getenv("AZURE_MODEL", None) default_model = os.getenv("AZURE_MODEL", None)
class Processor(ConsumerProducer): class Processor(LlmService):
def __init__(self, **params): def __init__(self, **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)
temperature = params.get("temperature", default_temperature) temperature = params.get("temperature", default_temperature)
max_output = params.get("max_output", default_max_output) max_output = params.get("max_output", default_max_output)
@ -51,11 +41,6 @@ class Processor(ConsumerProducer):
super(Processor, self).__init__( super(Processor, self).__init__(
**params | { **params | {
"input_queue": input_queue,
"output_queue": output_queue,
"subscriber": subscriber,
"input_schema": TextCompletionRequest,
"output_schema": TextCompletionResponse,
"temperature": temperature, "temperature": temperature,
"max_output": max_output, "max_output": max_output,
"model": model, "model": model,
@ -63,19 +48,6 @@ class Processor(ConsumerProducer):
} }
) )
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.temperature = temperature self.temperature = temperature
self.max_output = max_output self.max_output = max_output
self.model = model self.model = model
@ -84,41 +56,31 @@ class Processor(ConsumerProducer):
api_key=token, api_key=token,
api_version=api, api_version=api,
azure_endpoint = endpoint, azure_endpoint = endpoint,
) )
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)
prompt = v.system + "\n\n" + v.prompt
prompt = system + "\n\n" + prompt
try: try:
with __class__.text_completion_metric.time(): resp = self.openai.chat.completions.create(
resp = self.openai.chat.completions.create( model=self.model,
model=self.model, messages=[
messages=[ {
{ "role": "user",
"role": "user", "content": [
"content": [ {
{ "type": "text",
"type": "text", "text": prompt
"text": prompt }
} ]
] }
} ],
], temperature=self.temperature,
temperature=self.temperature, max_tokens=self.max_output,
max_tokens=self.max_output, top_p=1,
top_p=1, )
)
inputtokens = resp.usage.prompt_tokens inputtokens = resp.usage.prompt_tokens
outputtokens = resp.usage.completion_tokens outputtokens = resp.usage.completion_tokens
@ -127,15 +89,14 @@ class Processor(ConsumerProducer):
print(f"Output Tokens: {outputtokens}", flush=True) print(f"Output Tokens: {outputtokens}", flush=True)
print("Send response...", flush=True) print("Send response...", flush=True)
r = TextCompletionResponse( r = LlmResult(
response=resp.choices[0].message.content, text = resp.choices[0].message.content,
error=None, in_token = inputtokens,
in_token=inputtokens, out_token = outputtokens,
out_token=outputtokens, model = self.model
model=self.model
) )
await self.send(r, properties={"id": id}) return r
except RateLimitError: except RateLimitError:
@ -147,35 +108,15 @@ class Processor(ConsumerProducer):
except Exception as e: except Exception as e:
# Apart from rate limits, treat all exceptions as unrecoverable # Apart from rate limits, treat all exceptions as unrecoverable
print(f"Exception: {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)
print("Done.", flush=True) print("Done.", flush=True)
@staticmethod @staticmethod
def add_args(parser): def add_args(parser):
ConsumerProducer.add_args( LlmService.add_args(parser)
parser, default_input_queue, default_subscriber,
default_output_queue,
)
parser.add_argument( parser.add_argument(
'-e', '--endpoint', '-e', '--endpoint',
@ -217,4 +158,4 @@ class Processor(ConsumerProducer):
def run(): def run():
Processor.launch(module, __doc__) Processor.launch(default_ident, __doc__)

124
trustgraph-flow/trustgraph/model/text_completion/claude/llm.py
View file

@ -8,30 +8,20 @@ import anthropic
from prometheus_client import Histogram from prometheus_client import Histogram
import os import os
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 .... 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 = 'claude-3-5-sonnet-20240620' default_model = 'claude-3-5-sonnet-20240620'
default_temperature = 0.0 default_temperature = 0.0
default_max_output = 8192 default_max_output = 8192
default_api_key = os.getenv("CLAUDE_KEY") default_api_key = os.getenv("CLAUDE_KEY")
class Processor(ConsumerProducer): class Processor(LlmService):
def __init__(self, **params): def __init__(self, **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) model = params.get("model", default_model)
api_key = params.get("api_key", default_api_key) api_key = params.get("api_key", default_api_key)
temperature = params.get("temperature", default_temperature) temperature = params.get("temperature", default_temperature)
@ -42,30 +32,12 @@ class Processor(ConsumerProducer):
super(Processor, self).__init__( super(Processor, self).__init__(
**params | { **params | {
"input_queue": input_queue,
"output_queue": output_queue,
"subscriber": subscriber,
"input_schema": TextCompletionRequest,
"output_schema": TextCompletionResponse,
"model": model, "model": model,
"temperature": temperature, "temperature": temperature,
"max_output": max_output, "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.model = model
self.claude = anthropic.Anthropic(api_key=api_key) self.claude = anthropic.Anthropic(api_key=api_key)
self.temperature = temperature self.temperature = temperature
@ -73,39 +45,27 @@ class Processor(ConsumerProducer):
print("Initialised", flush=True) print("Initialised", flush=True)
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)
prompt = v.prompt
try: try:
with __class__.text_completion_metric.time(): response = message = self.claude.messages.create(
model=self.model,
response = message = self.claude.messages.create( max_tokens=self.max_output,
model=self.model, temperature=self.temperature,
max_tokens=self.max_output, system = system,
temperature=self.temperature, messages=[
system = v.system, {
messages=[ "role": "user",
{ "content": [
"role": "user", {
"content": [ "type": "text",
{ "text": prompt
"type": "text", }
"text": prompt ]
} }
] ]
} )
]
)
resp = response.content[0].text resp = response.content[0].text
inputtokens = response.usage.input_tokens inputtokens = response.usage.input_tokens
@ -114,17 +74,12 @@ class Processor(ConsumerProducer):
print(f"Input Tokens: {inputtokens}", flush=True) print(f"Input Tokens: {inputtokens}", flush=True)
print(f"Output Tokens: {outputtokens}", flush=True) print(f"Output Tokens: {outputtokens}", flush=True)
print("Send response...", flush=True) resp = LlmResult(
r = TextCompletionResponse( text = resp,
response=resp, in_token = inputtokens,
error=None, out_token = outputtokens,
in_token=inputtokens, model = self.model
out_token=outputtokens,
model=self.model
) )
self.send(r, properties={"id": id})
print("Done.", flush=True)
except anthropic.RateLimitError: except anthropic.RateLimitError:
@ -136,31 +91,12 @@ class Processor(ConsumerProducer):
# Apart from rate limits, treat all exceptions as unrecoverable # Apart from rate limits, treat all exceptions as unrecoverable
print(f"Exception: {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 @staticmethod
def add_args(parser): def add_args(parser):
ConsumerProducer.add_args( LlmService.add_args(parser)
parser, default_input_queue, default_subscriber,
default_output_queue,
)
parser.add_argument( parser.add_argument(
'-m', '--model', '-m', '--model',
@ -189,7 +125,5 @@ class Processor(ConsumerProducer):
) )
def run(): def run():
Processor.launch(module, __doc__)
Processor.launch(default_ident, __doc__)

11
trustgraph-vertexai/trustgraph/model/text_completion/vertexai/llm.py
View file

@ -105,11 +105,12 @@ class Processor(LlmService):
safety_settings=self.safety_settings safety_settings=self.safety_settings
) )
resp = LlmResult() resp = LlmResult(
resp.text = response.text text = response.text,
resp.in_token = response.usage_metadata.prompt_token_count in_token = response.usage_metadata.prompt_token_count,
resp.out_token = response.usage_metadata.candidates_token_count out_token = response.usage_metadata.candidates_token_count,
resp.model = self.model model = self.model
)
print(f"Input Tokens: {resp.in_token}", flush=True) print(f"Input Tokens: {resp.in_token}", flush=True)
print(f"Output Tokens: {resp.out_token}", flush=True) print(f"Output Tokens: {resp.out_token}", flush=True)