File size: 1,156 Bytes
78210ab
63c4df3
78210ab
6703bdf
63c4df3
 
 
 
e4d8995
 
426e1d8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import spaces

@spaces.GPU
def get_unsloth():
    from unsloth import FastLanguageModel
    return FastLanguageModel

FastLanguageModel = get_unsloth()


class InferencePipeline:
    def __init__(self, conf, api_key):
        self.conf = conf
        self.token = api_key
        self.model, self.tokenizer = self.get_model()

    def get_model(self):
        model, tokenizer = FastLanguageModel.from_pretrained(
            model_name = self.conf["model"]["model_name"],
            max_seq_length = self.conf["model"]["max_seq_length"],
            dtype = self.conf["model"]["dtype"],
            load_in_4bit = self.conf["model"]["load_in_4bit"],
            token = self.token
        )

        FastLanguageModel.for_inference(model) # Enable native 2x faster inference
        return model, tokenizer

    def infer(self, prompt):
        inputs = self.tokenizer([prompt], return_tensors = "pt").to("cuda")
        outputs = model.generate(**inputs, 
                         max_new_tokens = self.conf["model"]["max_new_tokens"], 
                         use_cache = True)
        outputs = tokenizer.batch_decode(outputs)
        return outputs