From c7cd97b6d3acee8432132aab010afa55ffd082a2 Mon Sep 17 00:00:00 2001 From: JackColquitt Date: Sun, 15 Sep 2024 16:14:09 -0700 Subject: [PATCH] Added basic Llamafile integration --- Makefile | 2 +- scripts/text-completion-llamafile | 6 + setup.py | 3 +- .../text_completion/llamafile/__init__.py | 3 + .../text_completion/llamafile/__main__.py | 7 + .../model/text_completion/llamafile/llm.py | 195 ++++++++++++++++++ 6 files changed, 214 insertions(+), 2 deletions(-) create mode 100755 scripts/text-completion-llamafile create mode 100644 trustgraph/model/text_completion/llamafile/__init__.py create mode 100755 trustgraph/model/text_completion/llamafile/__main__.py create mode 100755 trustgraph/model/text_completion/llamafile/llm.py diff --git a/Makefile b/Makefile index 63c971ba..b82117a2 100644 --- a/Makefile +++ b/Makefile @@ -1,6 +1,6 @@ # VERSION=$(shell git describe | sed 's/^v//') -VERSION=0.9.4 +VERSION=0.9.5 DOCKER=podman diff --git a/scripts/text-completion-llamafile b/scripts/text-completion-llamafile new file mode 100755 index 00000000..38c48ac2 --- /dev/null +++ b/scripts/text-completion-llamafile @@ -0,0 +1,6 @@ +#!/usr/bin/env python3 + +from trustgraph.model.text_completion.llamafile import run + +run() + diff --git a/setup.py b/setup.py index e11b21bf..e1ac688f 100644 --- a/setup.py +++ b/setup.py @@ -4,7 +4,7 @@ import os with open("README.md", "r") as fh: long_description = fh.read() -version = "0.9.4" +version = "0.9.5" setuptools.setup( name="trustgraph", @@ -94,6 +94,7 @@ setuptools.setup( "scripts/text-completion-bedrock", "scripts/text-completion-claude", "scripts/text-completion-cohere", + "scripts/text-completion-llamafile", "scripts/text-completion-ollama", "scripts/text-completion-openai", "scripts/text-completion-vertexai", diff --git a/trustgraph/model/text_completion/llamafile/__init__.py b/trustgraph/model/text_completion/llamafile/__init__.py new file mode 100644 index 00000000..f2017af8 --- /dev/null +++ b/trustgraph/model/text_completion/llamafile/__init__.py @@ -0,0 +1,3 @@ + +from . llm import * + diff --git a/trustgraph/model/text_completion/llamafile/__main__.py b/trustgraph/model/text_completion/llamafile/__main__.py new file mode 100755 index 00000000..91342d2d --- /dev/null +++ b/trustgraph/model/text_completion/llamafile/__main__.py @@ -0,0 +1,7 @@ +#!/usr/bin/env python3 + +from . llm import run + +if __name__ == '__main__': + run() + diff --git a/trustgraph/model/text_completion/llamafile/llm.py b/trustgraph/model/text_completion/llamafile/llm.py new file mode 100755 index 00000000..c42ec472 --- /dev/null +++ b/trustgraph/model/text_completion/llamafile/llm.py @@ -0,0 +1,195 @@ + +""" +Simple LLM service, performs text prompt completion using OpenAI. +Input is prompt, output is response. +""" + +from openai import OpenAI +from prometheus_client import Histogram + +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 + +module = ".".join(__name__.split(".")[1:-1]) + +default_input_queue = text_completion_request_queue +default_output_queue = text_completion_response_queue +default_subscriber = module +default_model = 'LLaMA_CPP' +default_llamafile = 'http://localhost:8080/v1' +default_temperature = 0.0 +default_max_output = 4096 + +class Processor(ConsumerProducer): + + 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) + llamafile = params.get("llamafile", default_llamafile) + temperature = params.get("temperature", default_temperature) + max_output = params.get("max_output", default_max_output) + + 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, + "llamafile" : llamafile, + } + ) + + 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.llamafile=llamafile + self.temperature = temperature + self.max_output = max_output + self.openai = OpenAI( + base_url=self.llamafile, + api_key = "sk-no-key-required", + ) + + print("Initialised", flush=True) + + def handle(self, msg): + + v = msg.value() + + # Sender-produced ID + + id = msg.properties()["id"] + + print(f"Handling prompt {id}...", flush=True) + + prompt = v.prompt + + try: + + # FIXME: Rate limits + + with __class__.text_completion_metric.time(): + + resp = self.openai.chat.completions.create( + model=self.model, + messages=[ + {"role": "user", "content": prompt} + ] + #temperature=self.temperature, + #max_tokens=self.max_output, + #top_p=1, + #frequency_penalty=0, + #presence_penalty=0, + #response_format={ + # "type": "text" + #} + ) + + print(resp.choices[0].message.content, flush=True) + + print("Send response...", flush=True) + r = TextCompletionResponse( + response=resp.choices[0].message.content, + error=None, + ) + self.send(r, properties={"id": id}) + + print("Done.", flush=True) + + # FIXME: Wrong exception, don't know what this LLM throws + # for a rate limit + except TooManyRequests: + + print("Send rate limit response...", flush=True) + + r = TextCompletionResponse( + error=Error( + type = "rate-limit", + message = str(e), + ), + response=None, + ) + + self.producer.send(r, properties={"id": id}) + + self.consumer.acknowledge(msg) + + except Exception as e: + + print(f"Exception: {e}") + + print("Send error response...", flush=True) + + r = TextCompletionResponse( + error=Error( + type = "llm-error", + message = str(e), + ), + response=None, + ) + + self.producer.send(r, properties={"id": id}) + + self.consumer.acknowledge(msg) + + @staticmethod + def add_args(parser): + + ConsumerProducer.add_args( + parser, default_input_queue, default_subscriber, + default_output_queue, + ) + + parser.add_argument( + '-m', '--model', + default=default_model, + help=f'LLM model (default: LLaMA_CPP)' + ) + + parser.add_argument( + '-r', '--llamafile', + default=default_llamafile, + help=f'ollama (default: {default_llamafile})' + ) + + parser.add_argument( + '-t', '--temperature', + type=float, + default=default_temperature, + help=f'LLM temperature parameter (default: {default_temperature})' + ) + + parser.add_argument( + '-x', '--max-output', + type=int, + default=default_max_output, + help=f'LLM max output tokens (default: {default_max_output})' + ) + +def run(): + + Processor.start(module, __doc__) + +