Merge pull request #5 from mzarnecki/PHPRAG-3-deepseek-models-ollama-support

support DeepSeek models Coder-V2 and R1-7B

Author: mzarnecki

README.md

@@ -57,12 +57,12 @@ The application demonstrates an interesting use case of distinguishing between t
     - Copy `.env-sample` to `.env` in `app/src`
     - Choose your model in `.env`:
       ```env
-      MODEL=<model-option>  # Options: GPT-4o, Claude-3.5, Llama3.2, Mixtral, Bielik, Gemini2, DeepSeek
+      MODEL=<model-option>  # Options: GPT-4o, Claude-3.5, Llama3.2, Mixtral, Bielik, Gemini2, DeepSeek, , DeepSeek-R1-7B, DeepSeek-Coder-v2
       ```
 
 3. **API Configuration**
 
-   #### Local API Options (Mixtral, Llama3.2, Bielik)
+   #### Local API Options (Mixtral, Llama3.2, Bielik, DeepSeek-R1-7B, DeepSeek-Coder-v2)
     - No API key required (go directly to point 4.)
     - Requires more CPU/RAM
     - GPU recommended for better performance
@@ -234,7 +234,7 @@ docker-compose up
 
 ## 🎚 Customize
 - Use different LLMs. \
-You can pick from available LLMs: `GPT-4o, Claude-3.5, Llama3.2, Mixtral, Gemini2` \
+You can pick from available LLMs: `GPT-4o, Claude-3.5, Llama3.2, Mixtral, Bielik, DeepSeek, DeepSeek-R1-7B, DeepSeek-Coder-v2, Gemini2` \
 For using other ones you can just modify model name in LLM client class for model provider, for example `app/src/service/openai/GeneratedTextFromGPTProvider.php:13`
 ```php
     final class GeneratedTextFromGPTProvider extends AbstractGPTAPIClient

app/src/.env-sample

@@ -1,5 +1,5 @@
-MODEL=Llama3.2  #Available options: GPT-4o, Claude-3.5, Llama3.2, Mixtral, Bielik, Gemini2, DeepSeek
-#fill one API keys below only when using other model than Llama3.2 and Mixtral
+MODEL=Llama3.2  #Available options: GPT-4o, Claude-3.5, Llama3.2, Mixtral, Bielik, Gemini2, DeepSeek, DeepSeek-R1-7B, DeepSeek-Coder-v2
+#fill one API keys below only when using other model than Llama3.2, Mixtral, DeepSeek-R1-7B, DeepSeek-Coder-v2 (these are pulled locally and served with ollama)
 #OpenAI API
 OPENAI_API_KEY=your-open-ai-api-key
 #Google Gemini API

app/src/service/ServicesForSpecificModelFactory.php

@@ -6,6 +6,8 @@
 use service\gemini\GeminiTextEncoder;
 use service\gemini\GeneratedTextFromGeminiProvider;
 use service\ollama\GeneratedTextFromLocalBielikProvider;
+use service\ollama\GeneratedTextFromLocalDeepSeekCoderV2Provider;
+use service\ollama\GeneratedTextFromLocalDeepSeekR17BProvider;
 use service\ollama\GeneratedTextFromLocalLlama3Provider;
 use service\ollama\GeneratedTextFromMixtralProvider;
 use service\ollama\MxbaiTextEncoder;
@@ -24,6 +26,8 @@ public function getGeneratedTextProvider(string $model): GeneratedTextProviderIn
             'llama3.2' => GeneratedTextFromLocalLlama3Provider::class,
             'mixtral' => GeneratedTextFromMixtralProvider::class,
             'bielik' => GeneratedTextFromLocalBielikProvider::class,
+            'deepseek-r1-7b' => GeneratedTextFromLocalDeepSeekR17BProvider::class,
+            'deepseek-coder-v2' => GeneratedTextFromLocalDeepSeekCoderV2Provider::class,
             'gemini2' => GeneratedTextFromGeminiProvider::class,
         ];
 
@@ -43,6 +47,8 @@ public function getEmbeddingsService(string $model): TextEncoderInterface
             'llama3.2' => MxbaiTextEncoder::class,
             'mixtral' => MxbaiTextEncoder::class,
             'bielik' => MxbaiTextEncoder::class,
+            'deepseek-r1-7b' => MxbaiTextEncoder::class,
+            'deepseek-coder-v2' => MxbaiTextEncoder::class,
             'gemini2' => GeminiTextEncoder::class,
         ];
 

app/src/service/ollama/GeneratedTextFromLocalDeepSeekCoderV2Provider.php

@@ -0,0 +1,36 @@
+<?php
+
+namespace service\ollama;
+
+use League\Pipeline\StageInterface;
+use service\GeneratedTextProviderInterface;
+use service\pipeline\Payload;
+
+class GeneratedTextFromLocalDeepSeekCoderV2Provider extends AbstractOllamaAPIClient
+    implements StageInterface, GeneratedTextProviderInterface
+{
+    /**
+     * @param Payload $payload
+     * @return string
+     */
+    public function __invoke($payload)
+    {
+        return $this->generateText($payload->getPrompt(), $payload->getRagPrompt());
+    }
+
+    protected function getEndpoint(): string
+    {
+        return '/api/generate';
+    }
+
+    protected function getBodyParams(string $input): array
+    {
+        return [
+            "model" => "deepseek-coder-v2",
+            "prompt" => $input
+        ];
+    }
+}
+{
+
+}

app/src/service/ollama/GeneratedTextFromLocalDeepSeekR17BProvider.php

@@ -0,0 +1,36 @@
+<?php
+
+namespace service\ollama;
+
+use League\Pipeline\StageInterface;
+use service\GeneratedTextProviderInterface;
+use service\pipeline\Payload;
+
+class GeneratedTextFromLocalDeepSeekR17BProvider extends AbstractOllamaAPIClient
+    implements StageInterface, GeneratedTextProviderInterface
+{
+    /**
+     * @param Payload $payload
+     * @return string
+     */
+    public function __invoke($payload)
+    {
+        return $this->generateText($payload->getPrompt(), $payload->getRagPrompt());
+    }
+
+    protected function getEndpoint(): string
+    {
+        return '/api/generate';
+    }
+
+    protected function getBodyParams(string $input): array
+    {
+        return [
+            "model" => "deepseek-r1:7b",
+            "prompt" => $input
+        ];
+    }
+}
+{
+
+}

docker-compose.yaml

@@ -52,6 +52,7 @@ services:
       - .:/code
       - ./ollama/ollama:/root/.ollama
       - ./entrypoint.sh:/entrypoint.sh
+      - ./app/src/.env:/.env
     container_name: ollama-container
     tty: true
     restart: always

entrypoint.sh

@@ -1,28 +1,64 @@
 #!/bin/bash
 
-# Start Ollama in the background.
-/bin/ollama serve &
-# Record Process ID.
-pid=$!
+# Source the .env file to get the MODEL variable
+# Try multiple possible locations for the .env file
+if [ -f ".env" ]; then
+    echo "Found .env in current directory"
+    source ".env"
+else
+    echo "Error: .env file not found after trying multiple locations!"
+    echo "Please mount the .env file into the container or specify the correct path."
+    exit 1
+fi
 
-# Pause for Ollama to start.
-sleep 5
+# Check if MODEL is one of the locally hosted models
+if [[ "$MODEL" == "Llama3.2" || "$MODEL" == "Mixtral" || "$MODEL" == "Bielik" || "$MODEL" == "DeepSeek-R1-7B" || "$MODEL" == "DeepSeek-Coder-v2" ]]; then
+    # Start Ollama in the background
+    /bin/ollama serve &
+    # Record Process ID
+    pid=$!
 
-echo "🔴 Retrieve LLAMA3 model..."
-ollama pull llama3.2
-echo "🟢 Done!"
+    # Pause for Ollama to start
+    sleep 5
 
-echo "🔴 Retrieve Mixtral model..."
-ollama pull mistral
-echo "🟢 Done!"
+    # Pull only the selected model
+    case "$MODEL" in
+        "Llama3.2")
+            echo "🔴 Retrieving LLAMA3 model..."
+            ollama pull llama3.2
+            echo "🟢 Done!"
+            ;;
+        "Mixtral")
+            echo "🔴 Retrieving Mixtral model..."
+            ollama pull mistral
+            echo "🟢 Done!"
+            ;;
+        "Bielik")
+            echo "🔴 Retrieving Bielik model..."
+            ollama pull mwiewior/bielik
+            echo "🟢 Done!"
+            ;;
+        "DeepSeek-R1-7B")
+            echo "🔴 Retrieving DeepSeek-R1:7B model..."
+            ollama pull deepseek-r1:7b
+            echo "🟢 Done!"
+            ;;
+        "DeepSeek-Coder-v2")
+            echo "🔴 Retrieving DeepSeek-Coder-v2 model..."
+            ollama pull deepseek-coder-v2
+            echo "🟢 Done!"
+            ;;
+    esac
 
-echo "🔴 Retrieve mxbai embedding model..."
-ollama pull mxbai-embed-large
-echo "🟢 Done!"
+    # For embedding capabilities, always pull the mxbai model
+    echo "🔴 Retrieving mxbai embedding model..."
+    ollama pull mxbai-embed-large
+    echo "🟢 Done!"
 
-echo "🔴 Retrieve Bielik model..."
-ollama pull mwiewior/bielik
-echo "🟢 Done!"
-
-# Wait for Ollama process to finish.
-wait $pid
+    # Wait for Ollama process to finish
+    wait $pid
+else
+    echo "Using cloud-based model: $MODEL"
+    echo "No need to pull local models via Ollama."
+    exit 0
+fi