feat: Add /v1/audio/transcriptions and /v1/audio/translations with whisper.cpp

.github/scripts/e2e-test-linux-and-mac.sh → .github/scripts/e2e-test-llama-linux-and-mac.sh

@@ -21,12 +21,12 @@ range=$((max - min + 1))
 PORT=$((RANDOM % range + min))
 
 # Start the binary file
-"$BINARY_PATH" 1 127.0.0.1 $PORT > /tmp/nitro.log 2>&1 &
+"$BINARY_PATH" 1 127.0.0.1 $PORT >/tmp/nitro.log 2>&1 &
 
 # Get the process id of the binary file
 pid=$!
 
-if ! ps -p $pid > /dev/null; then
+if ! ps -p $pid >/dev/null; then
     echo "nitro failed to start. Logs:"
     cat /tmp/nitro.log
     exit 1
@@ -35,26 +35,27 @@ fi
 # Wait for a few seconds to let the server start
 sleep 5
 
-# Check if /tmp/testmodel exists, if not, download it
-if [[ ! -f "/tmp/testmodel" ]]; then
-    wget $DOWNLOAD_URL -O /tmp/testmodel
+# Check if /tmp/testllm exists, if not, download it
+if [[ ! -f "/tmp/testllm" ]]; then
+    wget $DOWNLOAD_URL -O /tmp/testllm
 fi
 
 # Run the curl commands
 response1=$(curl -o /tmp/response1.log -s -w "%{http_code}" --location "http://127.0.0.1:$PORT/inferences/llamacpp/loadModel" \
---header 'Content-Type: application/json' \
---data '{
-    "llama_model_path": "/tmp/testmodel",
+    --header 'Content-Type: application/json' \
+    --data '{
+    "llama_model_path": "/tmp/testllm",
     "ctx_len": 50,
     "ngl": 32,
     "embedding": false
 }' 2>&1)
 
-response2=$(curl -o /tmp/response2.log -s -w "%{http_code}" --location "http://127.0.0.1:$PORT/inferences/llamacpp/chat_completion" \
---header 'Content-Type: application/json' \
---header 'Accept: text/event-stream' \
---header 'Access-Control-Allow-Origin: *' \
---data '{
+response2=$(
+    curl -o /tmp/response2.log -s -w "%{http_code}" --location "http://127.0.0.1:$PORT/v1/chat/completions" \
+        --header 'Content-Type: application/json' \
+        --header 'Accept: text/event-stream' \
+        --header 'Access-Control-Allow-Origin: *' \
+        --data '{
         "messages": [
             {"content": "Hello there", "role": "assistant"},
             {"content": "Write a long and sad story for me", "role": "user"}
@@ -98,7 +99,6 @@ echo "----------------------"
 echo "Log run test:"
 cat /tmp/response2.log
 
-
 echo "Nitro test run successfully!"
 
 # Kill the server process

.github/scripts/e2e-test-windows.bat → .github/scripts/e2e-test-llama-windows.bat

@@ -1,7 +1,7 @@
 @echo off
 
 set "TEMP=C:\Users\%UserName%\AppData\Local\Temp"
-set "MODEL_PATH=%TEMP%\testmodel"
+set "MODEL_PATH=%TEMP%\testllm"
 
 rem Check for required arguments
 if "%~2"=="" (

.github/scripts/e2e-test-whisper-linux-and-mac.sh

@@ -0,0 +1,93 @@
+#!/bin/bash
+
+## Example run command
+# ./linux-and-mac.sh './jan/plugins/@janhq/inference-plugin/dist/nitro/nitro_mac_arm64' https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v0.3-GGUF/resolve/main/tinyllama-1.1b-chat-v0.3.Q2_K.gguf
+
+# Check for required arguments
+if [[ $# -ne 2 ]]; then
+    echo "Usage: $0 <path_to_binary> <url_to_download>"
+    exit 1
+fi
+
+rm /tmp/response1.log /tmp/response2.log /tmp/nitro.log
+
+BINARY_PATH=$1
+DOWNLOAD_URL=$2
+
+# Random port to ensure it's not used
+min=10000
+max=11000
+range=$((max - min + 1))
+PORT=$((RANDOM % range + min))
+
+# Start the binary file
+"$BINARY_PATH" 1 127.0.0.1 $PORT >/tmp/nitro.log 2>&1 &
+
+# Get the process id of the binary file
+pid=$!
+
+if ! ps -p $pid >/dev/null; then
+    echo "nitro failed to start. Logs:"
+    cat /tmp/nitro.log
+    exit 1
+fi
+
+# Wait for a few seconds to let the server start
+sleep 5
+
+# Check if /tmp/testwhisper exists, if not, download it
+if [[ ! -f "/tmp/testwhisper" ]]; then
+    wget $DOWNLOAD_URL -O /tmp/testwhisper
+fi
+
+# Run the curl commands
+response1=$(curl -o /tmp/response1.log -s -w "%{http_code}" --location "http://127.0.0.1:$PORT/v1/audio/load_model" \
+    --header 'Content-Type: application/json' \
+    --data '{
+    "model_path": "/tmp/testwhisper",
+    "model_id": "whisper.cpp"
+}' 2>&1)
+
+response2=$(
+    curl -o /tmp/response2.log -s -w "%{http_code}" --location "http://127.0.0.1:$PORT/v1/audio/transcriptions" \
+        --header 'Access-Control-Allow-Origin: *' \
+        --form 'file=@"../whisper.cpp/samples/jfk.wav"' \
+        --form 'model_id="whisper.cpp"' \
+        --form 'temperature="0.0"' \
+        --form 'prompt="The transcript is about OpenAI which makes technology like DALL·E, GPT-3, and ChatGPT with the hope of one day building an AGI system that benefits all of humanity. The president is trying to raly people to support the cause."' \
+        2>&1
+)
+
+error_occurred=0
+if [[ "$response1" -ne 200 ]]; then
+    echo "The first curl command failed with status code: $response1"
+    cat /tmp/response1.log
+    error_occurred=1
+fi
+
+if [[ "$response2" -ne 200 ]]; then
+    echo "The second curl command failed with status code: $response2"
+    cat /tmp/response2.log
+    error_occurred=1
+fi
+
+if [[ "$error_occurred" -eq 1 ]]; then
+    echo "Nitro test run failed!!!!!!!!!!!!!!!!!!!!!!"
+    echo "Nitro Error Logs:"
+    cat /tmp/nitro.log
+    kill $pid
+    exit 1
+fi
+
+echo "----------------------"
+echo "Log load model:"
+cat /tmp/response1.log
+
+echo "----------------------"
+echo "Log run test:"
+cat /tmp/response2.log
+
+echo "Nitro test run successfully!"
+
+# Kill the server process
+kill $pid

.github/scripts/e2e-test-whisper-windows.bat

@@ -0,0 +1,109 @@
+@echo off
+
+set "TEMP=C:\Users\%UserName%\AppData\Local\Temp"
+set "MODEL_PATH=%TEMP%\testwhisper"
+
+rem Check for required arguments
+if "%~2"=="" (
+    echo Usage: %~0 ^<path_to_binary^> ^<url_to_download^>
+    exit /b 1
+)
+
+set "BINARY_PATH=%~1"
+set "DOWNLOAD_URL=%~2"
+
+for %%i in ("%BINARY_PATH%") do set "BINARY_NAME=%%~nxi"
+
+echo BINARY_NAME=%BINARY_NAME%
+
+del %TEMP%\response1.log 2>nul
+del %TEMP%\response2.log 2>nul
+del %TEMP%\nitro.log 2>nul
+
+set /a min=9999
+set /a max=11000
+set /a range=max-min+1
+set /a PORT=%min% + %RANDOM% %% %range%
+
+rem Start the binary file
+start /B "" "%BINARY_PATH%" 1 "127.0.0.1" %PORT% > %TEMP%\nitro.log 2>&1
+
+ping -n 6 127.0.0.1 %PORT% > nul
+
+rem Capture the PID of the started process with "nitro" in its name
+for /f "tokens=2" %%a in ('tasklist /fi "imagename eq %BINARY_NAME%" /fo list ^| findstr /B "PID:"') do (
+    set "pid=%%a"
+)
+
+echo pid=%pid%
+
+if not defined pid (
+    echo nitro failed to start. Logs:
+    type %TEMP%\nitro.log
+    exit /b 1
+)
+
+rem Wait for a few seconds to let the server start
+
+rem Check if %TEMP%\testmodel exists, if not, download it
+if not exist "%MODEL_PATH%" (
+    bitsadmin.exe /transfer "DownloadTestModel" %DOWNLOAD_URL% "%MODEL_PATH%"
+)
+
+rem Define JSON strings for curl data
+call set "MODEL_PATH_STRING=%%MODEL_PATH:\=\\%%"
+set "curl_data1={\"model_path\":\"%MODEL_PATH_STRING%\",\"model_id\":\"whisper.cpp\"}"
+
+rem Print the values of curl_data1 for debugging
+echo curl_data1=%curl_data1%
+
+rem Run the curl commands and capture the status code
+curl.exe -o %TEMP%\response1.log -s -w "%%{http_code}" --location "http://127.0.0.1:%PORT%/v1/audio/load_model" --header "Content-Type: application/json" --data "%curl_data1%" > %TEMP%\response1_code.log 2>&1
+
+curl.exe -o %TEMP%\response2.log -s -w "%%{http_code}" --location "http://127.0.0.1:%PORT%/v1/audio/transcriptions" ^
+--header "Access-Control-Allow-Origin: *" ^
+--form 'model_id="whisper.cpp"' ^
+--form 'file=@"..\whisper.cpp\samples\jfk.wav"' ^
+--form 'temperature="0.0"' ^
+--form 'prompt="The transcript is about OpenAI which makes technology like DALL·E, GPT-3, and ChatGPT with the hope of one day building an AGI system that benefits all of humanity. The president is trying to raly people to support the cause."' ^
+> %TEMP%\response2_code.log 2>&1
+
+set "error_occurred=0"
+
+rem Read the status codes from the log files
+for /f %%a in (%TEMP%\response1_code.log) do set "response1=%%a"
+for /f %%a in (%TEMP%\response2_code.log) do set "response2=%%a"
+
+if "%response1%" neq "200" (
+    echo The first curl command failed with status code: %response1%
+    type %TEMP%\response1.log
+    set "error_occurred=1"
+)
+
+if "%response2%" neq "200" (
+    echo The second curl command failed with status code: %response2%
+    type %TEMP%\response2.log
+    set "error_occurred=1"
+)
+
+if "%error_occurred%"=="1" (
+    echo Nitro test run failed!!!!!!!!!!!!!!!!!!!!!!
+    echo Nitro Error Logs:
+    type %TEMP%\nitro.log
+    taskkill /f /pid %pid%
+    exit /b 1
+)
+
+
+echo ----------------------
+echo Log load model:
+type %TEMP%\response1.log
+
+echo ----------------------
+echo "Log run test:"
+type %TEMP%\response2.log
+
+echo Nitro test run successfully!
+
+rem Kill the server process
+taskkill /f /pid %pid%