Phi

Phi popularised small language models trained on carefully curated / synthetic “textbook-quality” data — strong capability per parameter, historically strong on code exercises in early Phi releases.

Adapted for this playbook from the 🤗 Transformers documentation by Hugging Face. Official page: https://huggingface.co/docs/transformers/model_doc/phi. Images © Hugging Face (documentation-images) unless noted. This is not a substitute for the upstream docs — verify against the current version.

Covers Hugging Face pages: model_doc/phi

Executive summary

Reach for Phi-class SLMs when you need on-device or low-cost assistants. Track which Phi generation you deploy (Phi-1/1.5/2/3/…) — APIs and licences differ.

Quick usage

from transformers import pipeline

pipe = pipeline(task="text-generation", model="microsoft/phi-1.5")
print(pipe("def print_prime(n):\n    \"\"\"Print all primes between 1 and n\"\"\"\n", max_new_tokens=80))

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("microsoft/phi-1")
model = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-1", device_map="auto", attn_implementation="sdpa"
)

Collection example: Phi-1.

Official API

Phi model doc.

What to do next

Return to Transformers overview or browse the API map.