File size: 6,242 Bytes
aab002d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
import os
import requests
from flask import Flask, request, jsonify
from transformers import pipeline
from PIL import Image
import io
import base64

# Import for image generation
from diffusers import AutoPipelineForText2Image

app = Flask(__name__)

# --- Configuration ---
GEMMA_MODEL_ID = "google/gemma-4-E2B-it"
IMAGE_GEN_MODEL_ID = "stabilityai/sd-turbo" # A fast, small Stable Diffusion model for demonstration
MAX_NEW_TOKENS = 200 # Adjust as needed for Gemma 4 response length
IMAGE_SIZE = (512, 512) # For generated images

# Determine device for models
# For a CPU-focused Dockerfile, this will default to CPU (-1 or "cpu")
if os.environ.get("USE_GPU", "false").lower() == "true" and os.getenv("CUDA_VISIBLE_DEVICES", "") != "":
    device = 0 # Use the first GPU
    torch_device_name = "cuda"
else:
    device = -1 # Use CPU
    torch_device_name = "cpu"

# --- Model Loading ---
gemma_pipeline = None
image_gen_pipeline = None

try:
    print(f"Loading Gemma 4 multimodal model: {GEMMA_MODEL_ID} on device {torch_device_name} (pipeline device {device})...")
    gemma_pipeline = pipeline("any-to-any", model=GEMMA_MODEL_ID, device=device)
    print("Gemma 4 model loaded successfully.")
except Exception as e:
    print(f"Error loading Gemma 4 model: {e}")

try:
    print(f"Loading Image Generation model: {IMAGE_GEN_MODEL_ID} on device {torch_device_name}...")
    image_gen_pipeline = AutoPipelineForText2Image.from_pretrained(IMAGE_GEN_MODEL_ID).to(torch_device_name)
    # Only enable xformers if on GPU
    if torch_device_name == "cuda":
        try:
            # Note: xformers might require a specific CUDA version or manual installation.
            # If this line causes issues, comment it out.
            image_gen_pipeline.enable_xformers_memory_efficient_attention() # Optional: for memory efficiency on GPU
            print("xFormers enabled for image generation.")
        except ImportError:
            print("xFormers not installed or not available, skipping memory efficient attention.")
    print("Image Generation model loaded successfully.")
except Exception as e:
    print(f"Error loading Image Generation model: {e}")

# --- Helper Functions ---
def encode_image_to_base64(image: Image.Image) -> str:
    buffered = io.BytesIO()
    image.save(buffered, format="PNG")
    return base64.b64encode(buffered.getvalue()).decode('utf-8')

# --- API Endpoints ---
@app.route('/')
def home():
    return "Multimodal AI (Gemma 4) and Image Generation API is running. Use /gemma-predict or /generate-image."

@app.route('/gemma-predict', methods=['POST'])
def gemma_predict():
    """
    Endpoint for Gemma 4 multimodal text generation (image + text -> text).
    """
    if gemma_pipeline is None:
        return jsonify({"error": "Gemma 4 model not loaded. Please check server logs."}), 503

    try:
        data = request.json
        if not data:
            return jsonify({"error": "No JSON data provided"}), 400

        image_base64 = data.get('image_base64')
        text_prompt = data.get('text_prompt', '')

        if not image_base64 and not text_prompt:
            return jsonify({"error": "At least 'image_base64' or 'text_prompt' must be provided"}), 400

        messages = []

        if image_base64:
            try:
                image_bytes = base64.b64decode(image_base64)
                image = Image.open(io.BytesIO(image_bytes))
                messages.append({
                    "type": "image",
                    "image": image,
                })
            except Exception as e:
                return jsonify({"error": f"Invalid image_base64 provided: {e}"}), 400

        if text_prompt:
            messages.append({
                "type": "text",
                "text": text_prompt,
            })

        if not messages:
            return jsonify({"error": "No valid input (image or text) provided for Gemma."}), 400

        full_messages = [
            {
                "role": "user",
                "content": messages,
            }
        ]

        output = gemma_pipeline(full_messages, max_new_tokens=MAX_NEW_TOKENS, return_full_text=False)

        if output and len(output) > 0 and "generated_text" in output[0]:
            return jsonify({"prediction": output[0]["generated_text"]})
        else:
            return jsonify({"error": "Gemma 4 model did not return generated text."}), 500

    except Exception as e:
        print(f"Error during Gemma 4 prediction: {e}")
        return jsonify({"error": f"An error occurred during Gemma 4 prediction: {str(e)}"}), 500

@app.route('/generate-image', methods=['POST'])
def generate_image():
    """
    Endpoint for text-to-image generation.
    """
    if image_gen_pipeline is None:
        return jsonify({"error": "Image generation model not loaded. Please check server logs."}), 503

    try:
        data = request.json
        if not data:
            return jsonify({"error": "No JSON data provided"}), 400

        prompt = data.get('prompt')
        if not prompt:
            return jsonify({"error": "Missing 'prompt' for image generation."}), 400

        # Generate image
        # You can add more parameters here like num_inference_steps, guidance_scale
        generated_image = image_gen_pipeline(prompt).images[0]

        # Encode the generated image to base64
        image_base64 = encode_image_to_base64(generated_image)

        return jsonify({"image_base64": image_base64, "prompt": prompt})

    except Exception as e:
        print(f"Error during image generation: {e}")
        return jsonify({"error": f"An error occurred during image generation: {str(e)}"}), 500

@app.route('/status', methods=['GET'])
def status():
    """
    Checks the status of both AI models.
    """
    gemma_status = "ready" if gemma_pipeline else "not_loaded"
    image_gen_status = "ready" if image_gen_pipeline else "not_loaded"
    return jsonify({
        "gemma_4_model_id": GEMMA_MODEL_ID,
        "gemma_4_status": gemma_status,
        "image_gen_model_id": IMAGE_GEN_MODEL_ID,
        "image_gen_status": image_gen_status,
        "device_used": torch_device_name
    })

# --- Main Execution ---
if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5000, debug=True)