Qwen2.5-0.5B-Instruct OpenVINO (INT8)

OpenVINO IR conversion of Qwen/Qwen2.5-0.5B-Instruct with weight-only INT8 quantization, produced on CPU-only Kaggle with optimum-cli (fully offline, token-free).

• Base model: Qwen/Qwen2.5-0.5B-Instruct (Qwen2ForCausalLM, 0.5B params, Apache 2.0)
• Task: text-generation (instruction-tuned)
• Precision: weights INT8 (NNCF), activations FP16/FP32 as traced
• Converted with: optimum-intel 2.2.0, openvino 2026.4.1, transformers 5.16.1
• 32768 context
• 3 IR parts: openvino_model, openvino_tokenizer, openvino_detokenizer

Usage

optimum-intel

from optimum.intel import OVModelForCausalLM
from transformers import AutoTokenizer

model = OVModelForCausalLM.from_pretrained("Nekodeus/qwen2-5-05b-instruct-openvino")
tokenizer = AutoTokenizer.from_pretrained("Nekodeus/qwen2-5-05b-instruct-openvino")
inputs = tokenizer("Explain quantum computing in one sentence:", return_tensors="pt")
print(tokenizer.decode(model.generate(**inputs, max_new_tokens=64)[0]))

OpenVINO GenAI

ov::genai::LLMPipeline pipe("path/to/qwen2-5-05b-instruct-openvino", "CPU");

Reproduce

optimum-cli export openvino \
  --model Qwen/Qwen2.5-0.5B-Instruct \
  --task text-generation \
  --weight-format int8 \
  qwen2-5-05b-instruct-openvino

Note: built with transformers==5.16.1 pinned (Kaggle image ships 5.20.2, which removed VisionRotaryEmbedding that optimum-intel 2.2.0 imports).