Nekodeus/phi-4-mini-instruct-openvino

Phi-4-mini-instruct OpenVINO (INT8)

OpenVINO IR conversion of microsoft/Phi-4-mini-instruct with weight-only INT8 quantization, produced on CPU-only Kaggle with optimum-cli (fully offline, token-free).

• Base model: microsoft/Phi-4-mini-instruct (Phi3ForCausalLM, 3.8B params, MIT)
• Task: text-generation
• Precision: weights INT8 (NNCF), activations FP16/FP32 as traced
• Converted with: optimum-intel 2.2.0, openvino 2026.4.1, transformers 5.16.1
• 128k context, function calling, improved reasoning over Phi-3.5-mini
• 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/phi-4-mini-instruct-openvino")
tokenizer = AutoTokenizer.from_pretrained("Nekodeus/phi-4-mini-instruct-openvino")
inputs = tokenizer("The capital of France is", return_tensors="pt")
print(tokenizer.decode(model.generate(**inputs, max_new_tokens=32)[0]))

OpenVINO GenAI

ov::genai::LLMPipeline pipe("path/to/Nekodeus/phi-4-mini-instruct-openvino", "CPU");

Reproduce

optimum-cli export openvino \
  --model microsoft/Phi-4-mini-instruct \
  --task text-generation \
  --weight-format int8 \
  phi-4-mini-instruct-openvino