feat(nemo): add Nemo (only asr for now) backend

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

backend/python/nemo/Makefile

@@ -0,0 +1,23 @@
+.PHONY: nemo-asr
+nemo-asr:
+	bash install.sh
+
+.PHONY: run
+run: nemo-asr
+	@echo "Running nemo-asr..."
+	bash run.sh
+	@echo "nemo-asr run."
+
+.PHONY: test
+test: nemo-asr
+	@echo "Testing nemo-asr..."
+	bash test.sh
+	@echo "nemo-asr tested."
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: clean
+clean: protogen-clean
+	rm -rf venv __pycache__

backend/python/nemo/backend.py

@@ -0,0 +1,147 @@
+#!/usr/bin/env python3
+"""
+gRPC server of LocalAI for NVIDIA NEMO Toolkit ASR.
+"""
+from concurrent import futures
+import time
+import argparse
+import signal
+import sys
+import os
+import backend_pb2
+import backend_pb2_grpc
+import torch
+import nemo.collections.asr as nemo_asr
+
+import grpc
+
+
+def is_float(s):
+    try:
+        float(s)
+        return True
+    except ValueError:
+        return False
+
+
+def is_int(s):
+    try:
+        int(s)
+        return True
+    except ValueError:
+        return False
+
+
+_ONE_DAY_IN_SECONDS = 60 * 60 * 24
+MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
+
+
+class BackendServicer(backend_pb2_grpc.BackendServicer):
+    def Health(self, request, context):
+        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
+
+    def LoadModel(self, request, context):
+        if torch.cuda.is_available():
+            device = "cuda"
+        else:
+            device = "cpu"
+        mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
+        if mps_available:
+            device = "mps"
+        if not torch.cuda.is_available() and request.CUDA:
+            return backend_pb2.Result(success=False, message="CUDA is not available")
+
+        self.device = device
+        self.options = {}
+
+        for opt in request.Options:
+            if ":" not in opt:
+                continue
+            key, value = opt.split(":", 1)
+            if is_float(value):
+                value = float(value)
+            elif is_int(value):
+                value = int(value)
+            elif value.lower() in ["true", "false"]:
+                value = value.lower() == "true"
+            self.options[key] = value
+
+        model_name = request.Model or "nvidia/parakeet-tdt-0.6b-v3"
+
+        try:
+            print(f"Loading NEMO ASR model from {model_name}", file=sys.stderr)
+            self.model = nemo_asr.models.ASRModel.from_pretrained(model_name=model_name)
+            print("NEMO ASR model loaded successfully", file=sys.stderr)
+        except Exception as err:
+            print(f"[ERROR] LoadModel failed: {err}", file=sys.stderr)
+            import traceback
+            traceback.print_exc(file=sys.stderr)
+            return backend_pb2.Result(success=False, message=str(err))
+
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def AudioTranscription(self, request, context):
+        result_segments = []
+        text = ""
+        try:
+            audio_path = request.dst
+            if not audio_path or not os.path.exists(audio_path):
+                print(f"Error: Audio file not found: {audio_path}", file=sys.stderr)
+                return backend_pb2.TranscriptResult(segments=[], text="")
+
+            # NEMO's transcribe method accepts a list of audio paths and returns a list of transcripts
+            results = self.model.transcribe([audio_path])
+
+            if not results or len(results) == 0:
+                return backend_pb2.TranscriptResult(segments=[], text="")
+
+            # Get the transcript text from the first result
+            text = results[0]
+            if text:
+                # Create a single segment with the full transcription
+                result_segments.append(backend_pb2.TranscriptSegment(
+                    id=0, start=0, end=0, text=text
+                ))
+
+        except Exception as err:
+            print(f"Error in AudioTranscription: {err}", file=sys.stderr)
+            import traceback
+            traceback.print_exc(file=sys.stderr)
+            return backend_pb2.TranscriptResult(segments=[], text="")
+
+        return backend_pb2.TranscriptResult(segments=result_segments, text=text)
+
+
+def serve(address):
+    server = grpc.server(
+        futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),
+        ])
+    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
+    server.add_insecure_port(address)
+    server.start()
+    print("Server started. Listening on: " + address, file=sys.stderr)
+
+    def signal_handler(sig, frame):
+        print("Received termination signal. Shutting down...")
+        server.stop(0)
+        sys.exit(0)
+
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    try:
+        while True:
+            time.sleep(_ONE_DAY_IN_SECONDS)
+    except KeyboardInterrupt:
+        server.stop(0)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Run the gRPC server.")
+    parser.add_argument("--addr", default="localhost:50051", help="The address to bind the server to.")
+    args = parser.parse_args()
+    serve(args.addr)

backend/python/nemo/install.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+set -e
+
+EXTRA_PIP_INSTALL_FLAGS="--no-build-isolation"
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+if [ "x${BUILD_PROFILE}" == "xintel" ]; then
+    EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
+fi
+
+PYTHON_VERSION="3.12"
+PYTHON_PATCH="12"
+PY_STANDALONE_TAG="20251120"
+
+installRequirements

backend/python/nemo/protogen.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto

backend/python/nemo/requirements-cpu.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cpu
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-cublas12.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cu128
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-cublas13.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-hipblas.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/rocm6.3
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-intel.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-l4t12.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu129/
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-l4t13.txt

@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements-mps.txt

@@ -0,0 +1,2 @@
+torch
+nemo_toolkit[asr]

backend/python/nemo/requirements.txt

@@ -0,0 +1,5 @@
+grpcio==1.71.0
+protobuf
+certifi
+packaging==24.1
+setuptools

backend/python/nemo/run.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+startBackend $@

backend/python/nemo/test.py

@@ -0,0 +1,99 @@
+"""
+Tests for the NEMO Toolkit ASR gRPC backend.
+"""
+import unittest
+import subprocess
+import time
+import os
+import tempfile
+import shutil
+import backend_pb2
+import backend_pb2_grpc
+
+import grpc
+
+# Skip heavy transcription test in CI (model download + inference)
+SKIP_ASR_TESTS = os.environ.get("SKIP_ASR_TESTS", "false").lower() == "true"
+
+
+class TestBackendServicer(unittest.TestCase):
+    def setUp(self):
+        self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
+        time.sleep(15)
+
+    def tearDown(self):
+        self.service.terminate()
+        self.service.wait()
+
+    def test_server_startup(self):
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.Health(backend_pb2.HealthMessage())
+                self.assertEqual(response.message, b'OK')
+        except Exception as err:
+            print(err)
+            self.fail("Server failed to start")
+        finally:
+            self.tearDown()
+
+    def test_load_model(self):
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.LoadModel(backend_pb2.ModelOptions(Model="nvidia/parakeet-tdt-0.6b-v3"))
+                self.assertTrue(response.success, response.message)
+                self.assertEqual(response.message, "Model loaded successfully")
+        except Exception as err:
+            print(err)
+            self.fail("LoadModel service failed")
+        finally:
+            self.tearDown()
+
+    @unittest.skipIf(SKIP_ASR_TESTS, "ASR transcription test skipped (SKIP_ASR_TESTS=true)")
+    def test_audio_transcription(self):
+        temp_dir = tempfile.mkdtemp()
+        audio_file = os.path.join(temp_dir, 'audio.wav')
+        try:
+            # Download a sample audio file for testing
+            url = "https://audio-samples.github.io/samples/mp3/crowd-cheering-and-applause-sound-effect.mp3"
+            result = subprocess.run(
+                ["wget", "-q", url, "-O", audio_file],
+                capture_output=True,
+                text=True,
+                timeout=30,
+            )
+            if result.returncode != 0:
+                self.skipTest(f"Could not download sample audio: {result.stderr}")
+            if not os.path.exists(audio_file):
+                self.skipTest("Sample audio file not found after download")
+
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                load_response = stub.LoadModel(backend_pb2.ModelOptions(Model="nvidia/parakeet-tdt-0.6b-v3"))
+                self.assertTrue(load_response.success, load_response.message)
+
+                transcript_response = stub.AudioTranscription(
+                    backend_pb2.TranscriptRequest(dst=audio_file)
+                )
+                self.assertIsNotNone(transcript_response)
+                self.assertIsNotNone(transcript_response.text)
+                self.assertGreaterEqual(len(transcript_response.segments), 0)
+                all_text = ""
+                for segment in transcript_response.segments:
+                    all_text += segment.text
+                print(f"Transcription result: {all_text}")
+                self.assertIn("big", all_text)
+                if transcript_response.segments:
+                    self.assertIsNotNone(transcript_response.segments[0].text)
+        finally:
+            self.tearDown()
+            if os.path.exists(temp_dir):
+                shutil.rmtree(temp_dir)
+
+
+if __name__ == '__main__':
+    unittest.main()

backend/python/nemo/test.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+runUnittests