File size: 4,352 Bytes
5eb40a3
 
4f97961
5eb40a3
 
 
 
 
1303c4e
98cad23
 
1303c4e
5eb40a3
 
 
 
 
1303c4e
98cad23
64d30b7
98cad23
 
5eb40a3
1303c4e
5eb40a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
64d30b7
 
 
 
 
 
5eb40a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4f97961
 
 
 
 
 
 
 
5eb40a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1303c4e
 
 
 
 
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
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel

from torch import cuda
from transformers import AutoModelForCausalLM, AutoTokenizer

from huggingface_hub import login
from dotenv import load_dotenv
import os

import logging
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Required for access to a gated model
load_dotenv()
hf_token = os.getenv("HF_TOKEN", None)
if hf_token is not None:
    login(token=hf_token)

# Keep data in session
model = None
tokenizer = None

class TextInput(BaseModel):
    text: str
    min_length: int = 3
    # Apertus by default supports a context length up to 65,536 tokens.
    max_length: int = 65536

class ModelResponse(BaseModel):
    text: str
    confidence: float
    processing_time: float


@asynccontextmanager
async def lifespan(app: FastAPI):
    """Load the transformer model on startup"""
    global model, tokenizer
    try:
        logger.info("Loading sentiment analysis model...")
        # TODO: make this configurable
        model_name = "swiss-ai/Apertus-8B-Instruct-2509"

        # Automatically select device based on availability
        device = "cuda" if cuda.is_available() else "cpu"

        # load the tokenizer and the model
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForCausalLM.from_pretrained(
            model_name,
            device_map="auto",        # Automatically splits model across CPU/GPU
            low_cpu_mem_usage=True,   # Avoids unnecessary CPU memory duplication
            offload_folder="offload", # Temporary offload to disk
        )
        #.to(device)
        logger.info(f"Model loaded successfully! ({device})")
    except Exception as e:
        logger.error(f"Failed to load model: {e}")
        raise e
    # Release resources when the app is stopped
    yield
    model.clear()
    tokenizer.clear()


# Setup our app
app = FastAPI(
    title="Apertus API",
    description="REST API for serving Apertus models via Hugging Face transformers",
    version="0.1.0",
    lifespan=lifespan
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

@app.get("/predict", response_model=ModelResponse)
async def predict(q: str):
    """Generate a model response for input text"""
    if model is None or tokenizer is None:
        raise HTTPException(status_code=503, detail="Model not loaded")

    try:
        import time
        start_time = time.time()

        input_data = TextInput(text=q)

        # Truncate text if too long
        text = input_data.text[:input_data.max_length]
        if len(text) == input_data.max_length:
            logger.warning("Warning: text truncated")
        if len(text) < input_data.min_length:
            logger.warning("Warning: empty text, aborting")
            return None

        # Prepare the model input
        messages_think = [
            {"role": "user", "content": text}
        ]
        text = tokenizer.apply_chat_template(
            messages_think,
            tokenize=False,
            add_generation_prompt=True,
        )
        model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

        # Generate the output
        generated_ids = model.generate(**model_inputs, max_new_tokens=32768)

        # Get and decode the output
        output_ids = generated_ids[0][len(model_inputs.input_ids[0]) :]
        result = tokenizer.decode(output_ids, skip_special_tokens=True)

        # Checkpoint
        processing_time = time.time() - start_time

        return ModelResponse(
            text=result['label'],
            confidence=result['score'],
            processing_time=processing_time
        )

    except Exception as e:
        logger.error(f"Evaluation error: {e}")
        raise HTTPException(status_code=500, detail="Evaluation failed")

@app.get("/health")
async def health_check():
    """Health check and basic configuration"""
    return {
        "status": "healthy",
        "model_loaded": model is not None,
        "gpu_available": cuda.is_available()
    }

@app.get("/")
def read_root():
    return PlainTextResponse('Habemus Apertus')