feat: Add gRPC protocol for Unsloth fine-tuning backend (Phase 1)

.github/workflows/backend.yml

@@ -79,6 +79,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: ''
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-cpu-unsloth'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'true'
+            backend: "unsloth"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: ''
             cuda-major-version: ""
             cuda-minor-version: ""
@@ -444,6 +457,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "8"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-cuda-12-unsloth'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "unsloth"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "8"
@@ -965,6 +991,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: 'cublas'
+            cuda-major-version: "13"
+            cuda-minor-version: "0"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-cuda-13-unsloth'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "unsloth"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: 'cublas'
             cuda-major-version: "13"
             cuda-minor-version: "0"

Makefile

@@ -1,5 +1,5 @@
 # Disable parallel execution for backend builds
-.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/outetts backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/mlx-distributed backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/moonshine backends/pocket-tts backends/qwen-tts backends/faster-qwen3-tts backends/qwen-asr backends/nemo backends/voxcpm backends/whisperx backends/ace-step backends/acestep-cpp backends/fish-speech backends/voxtral backends/opus
+.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/outetts backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/mlx-distributed backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/moonshine backends/pocket-tts backends/qwen-tts backends/faster-qwen3-tts backends/qwen-asr backends/nemo backends/voxcpm backends/whisperx backends/ace-step backends/acestep-cpp backends/fish-speech backends/voxtral backends/opus backends/unsloth
 
 GOCMD=go
 GOTEST=$(GOCMD) test
@@ -421,6 +421,7 @@ prepare-test-extra: protogen-python
 	$(MAKE) -C backend/python/voxcpm
 	$(MAKE) -C backend/python/whisperx
 	$(MAKE) -C backend/python/ace-step
+	$(MAKE) -C backend/python/unsloth
 
 test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/transformers test
@@ -440,6 +441,7 @@ test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/voxcpm test
 	$(MAKE) -C backend/python/whisperx test
 	$(MAKE) -C backend/python/ace-step test
+	$(MAKE) -C backend/python/unsloth test
 
 DOCKER_IMAGE?=local-ai
 IMAGE_TYPE?=core
@@ -571,6 +573,7 @@ BACKEND_NEMO = nemo|python|.|false|true
 BACKEND_VOXCPM = voxcpm|python|.|false|true
 BACKEND_WHISPERX = whisperx|python|.|false|true
 BACKEND_ACE_STEP = ace-step|python|.|false|true
+BACKEND_UNSLOTH = unsloth|python|.|false|true
 BACKEND_MLX_DISTRIBUTED = mlx-distributed|python|./|false|true
 
 # Helper function to build docker image for a backend
@@ -627,14 +630,15 @@ $(eval $(call generate-docker-build-target,$(BACKEND_NEMO)))
 $(eval $(call generate-docker-build-target,$(BACKEND_VOXCPM)))
 $(eval $(call generate-docker-build-target,$(BACKEND_WHISPERX)))
 $(eval $(call generate-docker-build-target,$(BACKEND_ACE_STEP)))
+$(eval $(call generate-docker-build-target,$(BACKEND_UNSLOTH)))
 $(eval $(call generate-docker-build-target,$(BACKEND_ACESTEP_CPP)))
 $(eval $(call generate-docker-build-target,$(BACKEND_MLX_DISTRIBUTED)))
 
 # Pattern rule for docker-save targets
 docker-save-%: backend-images
 	docker save local-ai-backend:$* -o backend-images/$*.tar
 
-docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-transformers docker-build-outetts docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-fish-speech docker-build-faster-qwen3-tts docker-build-qwen-asr docker-build-nemo docker-build-voxcpm docker-build-whisperx docker-build-ace-step docker-build-acestep-cpp docker-build-voxtral docker-build-mlx-distributed
+docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-transformers docker-build-outetts docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-fish-speech docker-build-faster-qwen3-tts docker-build-qwen-asr docker-build-nemo docker-build-voxcpm docker-build-whisperx docker-build-ace-step docker-build-acestep-cpp docker-build-voxtral docker-build-mlx-distributed docker-build-unsloth
 
 ########################################################
 ### Mock Backend for E2E Tests

backend/backend.proto

@@ -39,6 +39,8 @@ service Backend {
   rpc AudioDecode(AudioDecodeRequest) returns (AudioDecodeResult) {}
 
   rpc ModelMetadata(ModelOptions) returns (ModelMetadataResponse) {}
+
+  rpc TrainStream(TrainRequest) returns (stream TrainResponse) {}
 }
 
 // Define the empty request
@@ -528,3 +530,49 @@ message ModelMetadataResponse {
   string rendered_template = 2;  // The rendered chat template with enable_thinking=true (empty if not applicable)
   ToolFormatMarkers tool_format = 3;  // Auto-detected tool format markers from differential template analysis
 }
+
+message TrainRequest {
+  string model = 1;                   // Base model name or HuggingFace model ID
+  string dataset = 2;                 // Path or HuggingFace dataset ID
+  string output_dir = 3;              // Output directory for the fine-tuned model
+
+  // Hyperparameters
+  int32 epochs = 4;                   // Number of training epochs
+  int32 batch_size = 5;               // Training batch size
+  float learning_rate = 6;            // Learning rate
+  int32 max_seq_length = 7;           // Maximum sequence length
+
+  // LoRA parameters
+  int32 lora_rank = 8;                // LoRA rank (r)
+  int32 lora_alpha = 9;               // LoRA alpha scaling factor
+  float lora_dropout = 10;            // LoRA dropout rate
+  repeated string target_modules = 11; // LoRA target modules (e.g., "q_proj", "v_proj")
+
+  // Quantization
+  string quantization = 12;           // Quantization method (e.g., "4bit", "8bit", "none")
+
+  // Additional options
+  map<string, string> options = 13;   // Generic key-value options for backend-specific settings
+}
+
+message TrainResponse {
+  enum TrainStatus {
+    UNKNOWN = 0;
+    STARTING = 1;
+    RUNNING = 2;
+    COMPLETED = 3;
+    FAILED = 4;
+  }
+
+  TrainStatus status = 1;             // Current training status
+  float progress = 2;                 // Overall progress percentage (0-100)
+  int32 current_epoch = 3;            // Current epoch number
+  int32 total_epochs = 4;             // Total number of epochs
+  int32 current_step = 5;             // Current training step
+  int32 total_steps = 6;              // Total number of training steps
+  float loss = 7;                     // Current training loss
+  float learning_rate_current = 8;    // Current learning rate (may change with schedulers)
+  string message = 9;                 // Human-readable status message
+  string error = 10;                  // Error message if status is FAILED
+  string output_path = 11;            // Path to the output model (set on completion)
+}

backend/index.yaml

@@ -118,7 +118,7 @@
   capabilities:
     nvidia: "cuda12-rfdetr"
     intel: "intel-rfdetr"
     #amd: "rocm-rfdetr"
     nvidia-l4t: "nvidia-l4t-arm64-rfdetr"
     metal: "metal-rfdetr"
     default: "cpu-rfdetr"
@@ -530,6 +530,25 @@
     nvidia-cuda-12: "cuda12-chatterbox"
     nvidia-l4t-cuda-12: "nvidia-l4t-arm64-chatterbox"
     nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-chatterbox"
+- &unsloth
+  urls:
+    - https://github.com/unslothai/unsloth
+  description: |
+    Fine-tuning backend powered by Unsloth. Supports QLoRA, LoRA, and full fine-tuning
+    of LLMs with 2x faster training and 70% less memory. Includes GGUF export support.
+  tags:
+    - fine-tuning
+    - training
+    - lora
+    - qlora
+  license: apache-2.0
+  name: "unsloth"
+  alias: "unsloth"
+  capabilities:
+    nvidia: "cuda12-unsloth"
+    default: "cpu-unsloth"
+    nvidia-cuda-13: "cuda13-unsloth"
+    nvidia-cuda-12: "cuda12-unsloth"
 - &vibevoice
   urls:
     - https://github.com/microsoft/VibeVoice
@@ -1521,7 +1540,7 @@
   capabilities:
     nvidia: "cuda12-rfdetr-development"
     intel: "intel-rfdetr-development"
     #amd: "rocm-rfdetr-development"
     nvidia-l4t: "nvidia-l4t-arm64-rfdetr-development"
     metal: "metal-rfdetr-development"
     default: "cpu-rfdetr-development"
@@ -3029,3 +3048,41 @@
   uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-voxtral"
   mirrors:
     - localai/localai-backends:master-metal-darwin-arm64-voxtral
+## unsloth
+- !!merge <<: *unsloth
+  name: "unsloth-development"
+  capabilities:
+    nvidia: "cuda12-unsloth-development"
+    default: "cpu-unsloth-development"
+    nvidia-cuda-13: "cuda13-unsloth-development"
+    nvidia-cuda-12: "cuda12-unsloth-development"
+- !!merge <<: *unsloth
+  name: "cpu-unsloth"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-unsloth"
+  mirrors:
+    - localai/localai-backends:latest-cpu-unsloth
+- !!merge <<: *unsloth
+  name: "cpu-unsloth-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-unsloth"
+  mirrors:
+    - localai/localai-backends:master-cpu-unsloth
+- !!merge <<: *unsloth
+  name: "cuda12-unsloth"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-unsloth"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-unsloth
+- !!merge <<: *unsloth
+  name: "cuda12-unsloth-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-unsloth"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-unsloth
+- !!merge <<: *unsloth
+  name: "cuda13-unsloth"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-unsloth"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-unsloth
+- !!merge <<: *unsloth
+  name: "cuda13-unsloth-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-unsloth"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-unsloth

backend/python/unsloth/Makefile

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