Phi-3 / Phi-4#

Phi-3 and Phi-4 are Microsoft’s high-capability small language models using a shared transformer decoder architecture (Phi3ForCausalLM). Phi-4-mini and Phi-4 achieve strong benchmark results at relatively small parameter counts.

Task

Text Generation

Architecture

Phi3ForCausalLM

Parameters

3.8B – 14B

HF Org

microsoft

Available Models#

  • Phi-4: 14B

  • Phi-4-mini-instruct: 3.8B

  • Phi-3.5-mini-instruct: 3.8B

  • Phi-3-medium-128k-instruct: 14B

  • Phi-3-mini-128k-instruct: 3.8B

  • Phi-3-mini-4k-instruct: 3.8B

Architecture#

  • Phi3ForCausalLM

Example HF Models#

Model

HF ID

Phi-4

microsoft/Phi-4

Phi-4-mini-instruct

microsoft/Phi-4-mini-instruct

Phi-3-mini-4k-instruct

microsoft/Phi-3-mini-4k-instruct

Phi-3-mini-128k-instruct

microsoft/Phi-3-mini-128k-instruct

Phi-3-medium-128k-instruct

microsoft/Phi-3-medium-128k-instruct

Example Recipes#

Recipe

Description

phi_4_squad.yaml

SFT — Phi-4 on SQuAD

phi_4_squad_peft.yaml

LoRA — Phi-4 on SQuAD

phi_3_mini_it_squad.yaml

SFT — Phi-3-mini Instruct on SQuAD

phi_3_mini_it_squad_peft.yaml

LoRA — Phi-3-mini Instruct on SQuAD

Try with NeMo AutoModel#

1. Install (full instructions):

pip install nemo-automodel

2. Clone the repo to get the example recipes:

git clone https://github.com/NVIDIA-NeMo/Automodel.git
cd Automodel

3. Run the recipe from inside the repo:

automodel --nproc-per-node=8 examples/llm_finetune/phi/phi_4_squad.yaml
Run with Docker

1. Pull the container and mount a checkpoint directory:

docker run --gpus all -it --rm \
  --shm-size=8g \
  -v $(pwd)/checkpoints:/opt/Automodel/checkpoints \
  nvcr.io/nvidia/nemo-automodel:26.02.00

2. Navigate to the AutoModel directory (where the recipes are):

cd /opt/Automodel

3. Run the recipe:

automodel --nproc-per-node=8 examples/llm_finetune/phi/phi_4_squad.yaml

See the Installation Guide and LLM Fine-Tuning Guide.

Fine-Tuning#

See the LLM Fine-Tuning Guide.

Hugging Face Model Cards#