@aj-archipelago/cortex 1.0.22 → 1.0.24

package/config.js

@@ -164,6 +164,11 @@ var config = convict({
         default: 'null',
         env: 'WHISPER_TS_API_URL'
     },
+    subscriptionKeepAlive: {
+        format: Number,
+        default: 0,
+        env: 'SUBSCRIPTION_KEEP_ALIVE'
+    },
 });
 
 // Read in environment variables and set up service configuration

package/helper_apps/WhisperX/Dockerfile

@@ -28,4 +28,5 @@ RUN adduser -u 5678 --disabled-password --gecos "" appuser && chown -R appuser /
 USER appuser
 
 # During debugging, this entry point will be overridden. For more information, please refer to https://aka.ms/vscode-docker-python-debug
+# CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]
 CMD ["gunicorn", "--bind", "0.0.0.0:8000", "--timeout", "0", "-k", "uvicorn.workers.UvicornWorker", "app:app"]

package/helper_apps/WhisperX/app.py

@@ -2,11 +2,11 @@ import uvicorn
 from fastapi import FastAPI, HTTPException, Request
 from uuid import uuid4
 import os
-import requests
 import asyncio
 import whisper
 from whisper.utils import get_writer
 from fastapi.encoders import jsonable_encoder
+import time
 
 model_download_root = './models'
 model = whisper.load_model("large", download_root=model_download_root) #large, tiny
@@ -20,96 +20,85 @@ save_directory = "./tmp"  # folder for downloaded files
 os.makedirs(save_directory, exist_ok=True)
 
 
-def download_remote_file(url, save_directory):
-    # Generate a unique file name with a UUID
-    unique_name = str(uuid4()) + os.path.splitext(url)[-1]
-    save_path = os.path.join(save_directory, unique_name)
-
-    # Download the remote file
-    response = requests.get(url, stream=True)
-    response.raise_for_status()
-
-    # Save the downloaded file with the unique name
-    with open(save_path, 'wb') as file:
-        for chunk in response.iter_content(chunk_size=8192):
-            file.write(chunk)
-
-    return [unique_name, save_path]
-
 def delete_tmp_file(file_path):
     try:
         os.remove(file_path)
         print(f"Temporary file '{file_path}' has been deleted.")
     except OSError as e:
         print(f"Error: {e.strerror}")
-
-def modify_segments(result):
-    modified_segments = []
     
-    id = 0
-    for segment in result["segments"]:
-        for word_info in segment['words']:
-            word = word_info['word']
-            start = word_info['start']
-            end = word_info['end']
-            
-            modified_segment = {} #segment.copy()
-            modified_segment['id'] = id
-            modified_segment['text'] = word
-            modified_segment['start'] = start
-            modified_segment['end'] = end
-            modified_segments.append(modified_segment)
-            id+=1
+def transcribe(params):
+    if 'fileurl' not in params:
+        raise HTTPException(status_code=400, detail="fileurl parameter is required")
     
-    result["segments"] = modified_segments
-    
-def transcribe(fileurl):
-    print(f"Downloading file from: {fileurl}")
-    [unique_file_name, save_path] = download_remote_file(
-        fileurl, save_directory)
-    print(f"Downloaded file saved as: {unique_file_name}")
+    fileurl = params["fileurl"]
 
-    print(f"Transcribing file")
-    result = model.transcribe(save_path, word_timestamps=True)
+    #word_timestamps bool, default True
+    word_timestamps = True
+    if 'word_timestamps' in params: #parse as bool
+        word_timestamps = False if params['word_timestamps'] == 'False' else True
 
-    modify_segments(result)
+    print(f"Transcribing file {fileurl} with word_timestamps={word_timestamps}")
+    start_time = time.time()
+    result = model.transcribe(fileurl, word_timestamps=word_timestamps)
+    end_time = time.time()
+    execution_time = end_time - start_time
+    print("Transcribe execution time:", execution_time, "seconds")
 
     srtpath = os.path.join(save_directory, str(uuid4()) + ".srt")
 
     print(f"Saving transcription as : {srtpath}")
     writer = get_writer("srt", save_directory)
-    with open(srtpath, 'w', encoding='utf-8') as file_obj :
-        writer.write_result(result, file_obj)
+
+
+    writer_args = {'highlight_words': False, 'max_line_count': None, 'max_line_width': None, 'max_words_per_line': None}
+    if 'highlight_words' in params: #parse as bool
+        writer_args['highlight_words'] = params['highlight_words'] == 'True'
+    if 'max_line_count' in params: #parse as int
+        writer_args['max_line_count'] = int(params['max_line_count'])
+    if 'max_line_width' in params: #parse as int
+        writer_args['max_line_width'] = int(params['max_line_width'])
+    if 'max_words_per_line' in params: #parse as int
+        writer_args['max_words_per_line'] = int(params['max_words_per_line'])
+
+    # if and only if fileurl and word_timestamps=True, max_words_per_line=1
+    if fileurl and word_timestamps and len(params) <= 2:
+        writer_args['max_words_per_line'] = 1
+
+    # writer_args = {arg: args.pop(arg) for arg in word_options if arg in args}
+    writer(result, srtpath, **writer_args)
+
 
     with open(srtpath, "r") as f:
         srtstr = f.read()
 
-    # clean up tmp files
-    delete_tmp_file(save_path)
+    # clean up tmp out files
     delete_tmp_file(srtpath)
 
-    print(f"Transcription done.")
+    print(f"Transcription of file {fileurl} completed")
     return srtstr
 
 
-@app.get("/")
-@app.post("/")
-async def root(request: Request):
+async def get_params(request: Request):
+    params = {}
     if request.method == "POST":
         body = jsonable_encoder(await request.json())
-        fileurl = body.get("fileurl")
+        params = body
     else:
-        fileurl = request.query_params.get("fileurl")
-    if not fileurl:
-        return "No fileurl given!"
+        params = dict(request.query_params)
+    return params
 
+@app.get("/")
+@app.post("/")
+async def root(request: Request):
     if semaphore.locked():
         raise HTTPException(status_code=429, detail="Too Many Requests")
     
+    params = await get_params(request)
     async with semaphore:
-        result = await asyncio.to_thread(transcribe, fileurl)
+        result = await asyncio.to_thread(transcribe, params)
         return result
 
 if __name__ == "__main__":
-    print("Starting APPWhisper server", flush=True)
+    print("Starting APP Whisper server", flush=True)
     uvicorn.run(app, host="0.0.0.0", port=8000)

package/helper_apps/WhisperX/docker-compose.debug.yml

@@ -2,7 +2,7 @@ version: '3.4'
 
 services:
   cortex:
-    image: cortex
+    image: arc/whisper
     build:
       context: .
       dockerfile: ./Dockerfile

package/helper_apps/WhisperX/docker-compose.yml

@@ -2,9 +2,9 @@ version: '3.4'
 
 services:
   cortex:
-    image: cortex
+    image: arc/whisper
     build:
       context: .
       dockerfile: ./Dockerfile
     ports:
-      - 8000:8000
+      - 8000:8000  

package/helper_apps/WhisperX/requirements.txt

@@ -2,5 +2,4 @@
 fastapi[all]==0.89.0
 uvicorn[standard]==0.20.0
 gunicorn==20.1.0
-whisper
-stable-ts
+openai-whisper

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aj-archipelago/cortex",
-  "version": "1.0.22",
+  "version": "1.0.24",
   "description": "Cortex is a GraphQL API for AI. It provides a simple, extensible interface for using AI services from OpenAI, Azure and others.",
   "private": false,
   "repository": {

package/server/pathwayPrompter.js

@@ -13,7 +13,7 @@ import CohereSummarizePlugin from './plugins/cohereSummarizePlugin.js';
 import AzureCognitivePlugin from './plugins/azureCognitivePlugin.js';
 import OpenAiEmbeddingsPlugin from './plugins/openAiEmbeddingsPlugin.js';
 import OpenAIImagePlugin from './plugins/openAiImagePlugin.js';
-import OpenAIDallE3Plugin from './plugins/openAiDalle3Plugin.js';
+import OpenAIDallE3Plugin from './plugins/openAiDallE3Plugin.js';
 
 class PathwayPrompter {
     constructor(config, pathway, modelName, model) {

package/server/plugins/modelPlugin.js

@@ -241,7 +241,7 @@ class ModelPlugin {
             console.log(`\x1b[36m${modelInput}\x1b[0m`);
         }
     
-        console.log(`\x1b[34m> ${this.parseResponse(responseData)}\x1b[0m`);
+        console.log(`\x1b[34m> ${JSON.stringify(this.parseResponse(responseData))}\x1b[0m`);
     
         prompt && prompt.debugInfo && (prompt.debugInfo += `\n${JSON.stringify(data)}`);
     }

package/server/plugins/openAiDallE3Plugin.js

@@ -1,6 +1,7 @@
 import RequestDurationEstimator from '../../lib/requestDurationEstimator.js';
 import pubsub from '../pubsub.js';
 import ModelPlugin from './modelPlugin.js';
+import { request } from '../../lib/request.js';
 
 const requestDurationEstimator = new RequestDurationEstimator(10);
 
@@ -17,6 +18,22 @@ class OpenAIDallE3Plugin extends ModelPlugin {
      * we keep the request open and send progress updates to the client 
      * over a websocket.
      */
+
+    async executeRequest(url, data, params, headers, prompt, requestId, pathway) {
+        try {
+            this.aiRequestStartTime = new Date();
+            this.requestId = requestId;
+            this.logRequestStart(url, data);
+            const responseData = await request({ url, data, params, headers, cache: this.shouldCache }, this.modelName, this.requestId, pathway);
+               
+            this.logRequestData(data, responseData, prompt);
+            return this.parseResponse(responseData);
+        } catch (error) {
+            // Log the error and continue
+            console.error(error);
+        }
+    }
+
     async execute(text, parameters, _, pathwayResolver) {
         const url = this.requestUrl(text);
         const data = JSON.stringify({ prompt: text });
@@ -49,29 +66,25 @@ class OpenAIDallE3Plugin extends ModelPlugin {
         let attemptCount = 0;
         let data = null;
 
-        requestPromise.then((response) => {
-            state.status = "succeeded";
-            requestDurationEstimator.endRequest();
-            pubsub.publish('REQUEST_PROGRESS', {
-                requestProgress: {
-                    requestId,
-                    status: "succeeded",
-                    progress: 1,
-                    data: JSON.stringify(response),
-                }
-            });
-        }).catch((error) => {
-            state.status = "failed";
+        requestPromise
+        .then((response) => handleResponse(response))
+        .catch((error) => handleResponse(error));
+
+        function handleResponse(response) {
+            const status = response?.error ? "failed" : "succeeded";
+            const data = JSON.stringify(response?.error ? response : response);
+
+            const requestProgress = {
+                requestId,
+                status,
+                progress: 1,
+                data,
+            };
+
+            state.status = status;
             requestDurationEstimator.endRequest();
-            pubsub.publish('REQUEST_PROGRESS', {
-                requestProgress: {
-                    requestId,
-                    status: "failed",
-                    progress: 1,
-                    data: JSON.stringify(error),
-                }
-            });
-        });
+            pubsub.publish("REQUEST_PROGRESS", { requestProgress });
+        }
 
         // publish an update every 2 seconds, using the request duration estimator to calculate
         // the percent complete
@@ -96,7 +109,7 @@ class OpenAIDallE3Plugin extends ModelPlugin {
             await new Promise(resolve => setTimeout(resolve, 2000));
         }
         while (state.status !== "succeeded" && attemptCount++ < 30);
-
+        
         return data;
     }
 }

package/helper_apps/WhisperX/app-ts.py

@@ -1,76 +0,0 @@
-import uvicorn
-from fastapi import FastAPI
-import stable_whisper
-from uuid import uuid4
-import requests
-import os
-
-model_download_root = './models'
-model = stable_whisper.load_model('large', download_root=model_download_root) #large, tiny
-
-app = FastAPI()
-
-save_directory = "./tmp"  # folder for downloaded files
-os.makedirs(save_directory, exist_ok=True)
-
-
-def download_remote_file(url, save_directory):
-    # Generate a unique file name with a UUID
-    unique_name = str(uuid4()) + os.path.splitext(url)[-1]
-    save_path = os.path.join(save_directory, unique_name)
-
-    # Download the remote file
-    response = requests.get(url, stream=True)
-    response.raise_for_status()
-
-    # Save the downloaded file with the unique name
-    with open(save_path, 'wb') as file:
-        for chunk in response.iter_content(chunk_size=8192):
-            file.write(chunk)
-
-    return [unique_name, save_path]
-
-
-def delete_tmp_file(file_path):
-    try:
-        os.remove(file_path)
-        print(f"Temporary file '{file_path}' has been deleted.")
-    except OSError as e:
-        print(f"Error: {e.strerror}")
-
-
-async def transcribe(fileurl):
-    print(f"Downloading file from: {fileurl}")
-    [unique_file_name, save_path] = download_remote_file(
-        fileurl, save_directory)
-    print(f"Downloaded file saved as: {unique_file_name}")
-
-    print(f"Transcribing file")
-    result = model.transcribe(save_path)
-
-    srtpath = os.path.join(save_directory, str(uuid4()) + ".srt")
-
-    print(f"Saving transcription as : {srtpath}")
-    result.to_srt_vtt(srtpath, segment_level=False)
-
-    with open(srtpath, "r") as f:
-        srtstr = f.read()
-
-    # clean up tmp files
-    delete_tmp_file(save_path)
-    delete_tmp_file(srtpath)
-
-    print(f"Transcription done.")
-    return srtstr
-
-
-@app.get("/")
-async def root(fileurl: str):
-    if not fileurl:
-        return "No fileurl given!"
-    result = await transcribe(fileurl)
-    return result
-
-if __name__ == "__main__":
-    print("Starting server", flush=True)
-    uvicorn.run(app, host="0.0.0.0", port=8000)