Qwen2.5-1.5B-Instruct — ONNX (FP16)

ONNX export of Qwen2.5-1.5B-Instruct (1.5B parameters, FP16 weights) with KV cache support for efficient autoregressive generation.

Converted for use with inference4j, an inference-only AI library for Java.

Original Source

• Repository: Qwen/Qwen2.5-1.5B-Instruct
• License: Apache 2.0

Usage with inference4j

try (var gen = OnnxTextGenerator.qwen2().build()) {
    GenerationResult result = gen.generate("What is Java?");
    System.out.println(result.text());
}

Model Details

Property	Value
Architecture	Qwen2ForCausalLM (1.5B parameters, 28 layers, 1536 hidden, 12 heads, 2 KV heads)
Task	Text generation (instruction-tuned)
Precision	FP16
Context length	32,768 tokens
Vocabulary	151,936 tokens (BPE)
Chat template	ChatML (`<
Original framework	PyTorch (transformers)
Export method	Hugging Face Optimum (with KV cache, FP16)

License

This model is licensed under the Apache License 2.0. Original model by Qwen Team, Alibaba Cloud.