phi:latest

Upstream reference: Phi-2

Publisher: Microsoft

Phi-2 is a small Transformer text model trained on synthetic and filtered web material for language-model research.

Publisher-described context, separate from Sigma5C observations and recorded catalog statistics.

Published upstream specifications

Parameters

2.7 billion Source

Upstream license label

MIT Source

Publisher-described uses

• Research using question/answer, chat-style and code-format prompts. Source

Reference limitations

• The card says this model was not post-trained with human-feedback reinforcement learning and documents limitations including factual errors, bias and FP16 attention overflow. The original Phi tag must not inherit Phi-4 specifications. Source

A recorded tag and an upstream card association do not verify the deployed checkpoint, quantization, configured context or current hosted availability.

Publisher sources

• Phi-2 official documentation — Microsoft. Checked 2026-11-02.

Links

Recorded references do not establish downloadable weights, licensing or current availability.

• Microsoft — official site
• Ollama
• Hugging Face
• Hugging Face
• Official model announcement / documentation

Where phi:latest comes from

Named in 1 observation page

• Xeon vs AMD workstation — CPU inference head-to-head · V1