Merge pull request #179 from Gregory-Pereira/fix-llamacpp-builds

llamacpp ms fix: builds and allow variable model

.gitignore

@@ -4,6 +4,7 @@
 port_check.lock
 *build
 models/*
+model_servers/llamacpp_python/model.gguf
 !models/convert_models/*
 !models/Containerfile
 !models/README.md

model_servers/llamacpp_python/Makefile

@@ -10,22 +10,28 @@ VULKAN_IMAGE := quay.io/ai-lab/model_servers/$(APP)_vulkan:latest
 LLAMA_MODEL_NAME := llama-2-7b-chat.Q5_K_S.gguf
 LLAMA_MODEL_URL := https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q5_K_S.gguf
 
+TINY_LLAMA_MODEL_NAME := tinyllama-1.1b-chat-v1.0.Q4_K_S.gguf
+TINY_LLAMA_MODEL_URL := https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/blob/main/tinyllama-1.1b-chat-v1.0.Q4_K_S.gguf
+
 MISTRAL_MODEL_NAME := mistral-7b-instruct-v0.1.Q4_K_M.gguf
-MISTRAL_MODEL_URL := https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-GGUF/resolve/main/mistral-7b-instruct-v0.1.Q4_K_M.gguf 
+MISTRAL_MODEL_URL := https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-GGUF/resolve/main/mistral-7b-instruct-v0.1.Q4_K_M.gguf
 
 # --- END MODEL OPTIONS ---
 
-MODEL_PATH_FROM_ROOT := locallm/models/$(MISTRAL_MODEL_NAME) # CHOOSE MODEL HERE BY NAME
-RELATIVE_MODEL_PATH := ../../models
+SELECTED_MODEL_NAME := $(or $(SELECTED_MODEL),$(MISTRAL_MODEL_NAME))
+SELECTED_MODEL_URL := $(or $(SELECTED_MODEL_LINK),$(MISTRAL_MODEL_URL))
+
+RELATIVE_MODELS_PATH := ../../models
+MODELS_PATH := /locallm/models
 
-BIND_MOUNT_OPTIONS :=  ro
+BIND_MOUNT_OPTIONS := ro
 OS := $(shell uname -s)
 ifeq ($(OS),Linux)
     BIND_MOUNT_OPTIONS := ro,Z
 endif
 
 .Phony: all
-all: build  download-mistral-7b-instruct-v0.1.Q4_K_M.gguf run
+all: build download-model-mistral run
 
 .PHONY: build
 build:
@@ -39,26 +45,29 @@ build-cuda:
 build-vulkan:
 	podman build -t $(VULKAN_IMAGE) . -f cuda/Containerfile
 
+.PHONY: download-model-tiny-llama
+download-model-tiny-llama:
+	curl -H "Cache-Control: no-cache" -s -S -L -f $(TINY_LLAMA_MODEL_URL) -z $(RELATIVE_MODELS_PATH)/$(LLAMA_MODEL_NAME) -o $(RELATIVE_MODELS_PATH)/$(LLAMA_MODEL_NAME).tmp && mv -f $(RELATIVE_MODELS_PATH)/$(LLAMA_MODEL_NAME).tmp $(RELATIVE_MODELS_PATH)/$(LLAMA_MODEL_NAME) 2>/dev/null || rm -f $(RELATIVE_MODELS_PATH)/$(LLAMA_MODEL_NAME).tmp $(RELATIVE_MODELS_PATH)/$(LLAMA_MODEL_NAME)
+
 .PHONY: download-model-llama
 download-model-llama:
-	cd ../../models && \
-	curl -s -S -L -f $(LLAMA_MODEL_URL) -z $@ -o $@.tmp && mv -f $@.tmp $@ 2>/dev/null || rm -f $@.tmp $@
+	curl -H "Cache-Control: no-cache" -s -S -L -f $(LLAMA_MODEL_URL) -z $(RELATIVE_MODELS_PATH)/$(TINY_LLAMA_MODEL_NAME) -o $(RELATIVE_MODELS_PATH)/$(TINY_LLAMA_MODEL_NAME).tmp && mv -f $(RELATIVE_MODELS_PATH)/$(TINY_LLAMA_MODEL_NAME).tmp $(RELATIVE_MODELS_PATH)/$(TINY_LLAMA_MODEL_NAME) 2>/dev/null || rm -f $(RELATIVE_MODELS_PATH)/$(TINY_LLAMA_MODEL_NAME).tmp $(RELATIVE_MODELS_PATH)/$(TINY_LLAMA_MODEL_NAME)
+
 
 .PHONY: download-model-mistral
 download-model-mistral:
-	cd ../../models && \
-	curl -s -S -L -f $(MISTRAL_MODEL_URL)  -z $@ -o $@.tmp && mv -f $@.tmp $@ 2>/dev/null || rm -f $@.tmp $@
+	curl -H "Cache-Control: no-cache" -s -S -L -f $(MISTRAL_MODEL_URL) -z $(RELATIVE_MODELS_PATH)/$(MISTRAL_MODEL_NAME) -o $(RELATIVE_MODELS_PATH)/$(MISTRAL_MODEL_NAME).tmp && mv -f $(RELATIVE_MODELS_PATH)/$(MISTRAL_MODEL_NAME).tmp $(RELATIVE_MODELS_PATH)/$(MISTRAL_MODEL_NAME) 2>/dev/null || rm -f $(RELATIVE_MODELS_PATH)/$(MISTRAL_MODEL_NAME).tmp $(RELATIVE_MODELS_PATH)/$(MISTRAL_MODEL_NAME)
 
 .PHONY: install
 install:
 	pip install -r tests/requirements.txt
 
 .PHONY: run
 run:
-	cd ../.. && \
-	podman run -it -d -p $(PORT):$(PORT) -v ./models:$(MODEL_PATH_FROM_ROOT):$(BIND_MOUNT_OPTIONS) -e MODEL_PATH=$(MODEL_PATH_FROM_ROOT) -e HOST=0.0.0.0 -e PORT=$(PORT) --net=host $(IMAGE);
+	cd ../../models && \
+	podman run -it -d -p $(PORT):$(PORT) -v ./$(SELECTED_MODEL_NAME):$(MODELS_PATH)/model.gguf:$(BIND_MOUNT_OPTIONS) -e MODEL_PATH=$(MODELS_PATH)/model.gguf -e HOST=0.0.0.0 -e PORT=$(PORT) --net=host $(IMAGE)
 
 .PHONY: test
 test:
-	$(MAKE) all
-	pytest --log-cli-level NOTSET
+	curl -H "Cache-Control: no-cache" -s -S -L -f $(SELECTED_MODEL_URL) -z ./model.gguf -o ./model.gguf.tmp && mv -f ./model.gguf.tmp ./model.gguf 2>/dev/null || rm -f ./model.gguf.tmp ./model.gguf
+	pytest --log-cli-level NOTSET

model_servers/llamacpp_python/tests/conftest.py

@@ -1,18 +1,17 @@
 import pytest_container
 import os
 
-
 MS = pytest_container.Container(
         url=f"containers-storage:{os.environ['REGISTRY']}/{os.environ['IMAGE_NAME']}",
         volume_mounts=[
             pytest_container.container.BindMount(
-                container_path="/locallm/models",
-                host_path="./",
+                container_path="/locallm/models/model.gguf",
+                host_path=f"./model.gguf",
                 flags=["ro"]
             )
         ],
         extra_environment_variables={
-            "MODEL_PATH": "models/mistral-7b-instruct-v0.1.Q4_K_M.gguf",
+            "MODEL_PATH": "/locallm/models/model.gguf",
             "HOST": "0.0.0.0",
             "PORT": "8001"
         },
@@ -22,7 +21,6 @@
                 host_port=8001
             )
         ],
-        extra_launch_args=["--net=host"]
     )
 
 def pytest_generate_tests(metafunc):