Views
No views yet
| Property | Value |
|---|---|
| Base Model | jinaai/jina-embeddings-v2-base-code |
| Quantization | INT8 (dynamic) |
| Size | 154 MB (vs 612 MB fp32) |
| Dimensions | 768 |
| Max Tokens | 8192 |
| Languages | English + 30 programming languages |
1import onnxruntime as ort
2from huggingface_hub import hf_hub_download
3from tokenizers import Tokenizer
4import numpy as np
5
6# Load
7tokenizer = Tokenizer.from_file(hf_hub_download("nijaru/jina-code-int8", "tokenizer.json"))
8tokenizer.enable_padding(pad_id=0, pad_token="[PAD]")
9tokenizer.enable_truncation(max_length=512)
10session = ort.InferenceSession(hf_hub_download("nijaru/jina-code-int8", "model_int8.onnx"))
11
12def embed(texts):
13 encoded = tokenizer.encode_batch(texts)
14 input_ids = np.array([e.ids for e in encoded], dtype=np.int64)
15 attention_mask = np.array([e.attention_mask for e in encoded], dtype=np.int64)
16 outputs = session.run(None, {"input_ids": input_ids, "attention_mask": attention_mask})
17 embeddings = outputs[0]
18 mask = attention_mask[:, :, np.newaxis]
19 return (embeddings * mask).sum(axis=1) / mask.sum(axis=1)
20
21embeddings = embed(["def hello(): pass", "authentication flow"])