Microsoft Phi-3.5-mini

As developers grow AI applications, Azure's model flexibility and vast toolset meet modern enterprises' unique, complex, and diversified needs

[{"selector":"#anim-82ab47c3-31e3-40f5-aa35-854103104335 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.249999904429615%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-e85c9f42-d16f-4810-b212-1b33eb546a29","keyframes":{"transform":["rotate(-540deg) scale(0.1)","none"],"opacity":[0,1]},"delay":0,"duration":1000,"fill":"both","iterations":1}]

Azure is excited to provide developers new options and freedom when building AI applications with Azure AI

[{"selector":"#anim-2691d9bd-7af8-48a4-b103-2af36aebd4ad [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.249999904429615%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-822e00a6-4d11-4b24-89ff-10c1cd280414","keyframes":{"transform":["translate3d(115.55555%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1a5357f3-2ecb-4122-ab21-7533c615f1c3","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-cd7ab3c4-89d7-4eca-b842-09318548aeee","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

The Phi-3.5-MoE mixed-experts model is novel. A new model with 16 tiny experts improves quality and latency

[{"selector":"#anim-1c4b9556-9529-48df-ad4d-831a4b806e50 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.249999904429615%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-53b4853e-7536-4df9-adf2-9923f8289375","keyframes":{"transform":["translate3d(-123.84342%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .8, 1)","fill":"both"}] [{"selector":"#anim-79628fbf-5707-4fde-bfb2-9406987514c1","keyframes":{"transform":["rotateZ(-180deg)","rotateZ(0deg)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .5, 1)","fill":"forwards"}]

A Phi-3.5-mini variant is also revealed. Over 20 languages are supported by the new MoE model and tiny model

[{"selector":"#anim-d8a9c5b1-13f8-4f1d-90c4-2cdeea0dad98 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.30232548614247%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-f480f8c6-22ad-4f3a-ab00-a523baed3734","keyframes":{"transform":["translate3d(122.55891%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-067d9ae9-67ed-4dd6-8897-bf955c86e735","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-6404135c-96b4-4e87-bc6d-9029675ebe5f","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Phi-3.5-mini serverless endpoint guidance improves output predictability and application structure

[{"selector":"#anim-67ef6fca-ed9a-4979-96b5-341f935681ba [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.249999904429615%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-cceae012-3fc9-44e2-9873-53d41e6f6f4c","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-035519d9-2904-442b-9a6a-f0be7770a07e","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Allows developers express the model's structured output programmatic limitations in JSON, Python, HTML, SQL, or whatever the use case demands in a single API call

[{"selector":"#anim-69e3bf72-f18c-4dce-b3e5-b30580409781 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.249999904429615%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-483dcd59-ce4f-40e8-b0cd-6c018700b15d","keyframes":{"transform":["translate3d(-116.38226%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .8, 1)","fill":"both"}] [{"selector":"#anim-38722bcc-adf6-4d6d-9d8a-6e2cdc49f970","keyframes":{"transform":["rotateZ(-180deg)","rotateZ(0deg)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .5, 1)","fill":"forwards"}]

Phi-3.5-vision (4.2B parameters) can reason over several input photos supporting novel image comparison situations

[{"selector":"#anim-63ee8291-9391-4c80-a6ac-2fc3fd8b90c4 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(34.83455873150611%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-14461c95-830b-4db9-a8ed-5eea9aa942ad","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-b47dc96b-ea72-40ac-ab45-6a7c75fc660b","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Phi model developers may leverage Azure AI built-in and custom metrics to check quality and safety and guide mitigations

[{"selector":"#anim-fe9d84fe-161f-43bc-b8c6-93b6bf6060e9 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.30232548614247%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-11a449d6-01ee-4f1b-9cce-f6cf814697c9","keyframes":{"transform":["rotate(-540deg) scale(0.1)","none"],"opacity":[0,1]},"delay":0,"duration":1000,"fill":"both","iterations":1}]

Azure is excited to offer Jamba 1.5 Large and Jamba 1.5 to the Azure AI model portfolio to provide developers more choices

[{"selector":"#anim-32ff3362-e934-4d95-bdea-a6000a9af9da [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-29.574619647922702%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-1f35f582-110f-4a70-b0a0-b7187e78c2e8","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-3dd2cb19-027a-4061-b821-c529a734a779","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

RAG adds private organization data to generative AI models without retraining For more details Govindhtech.com

[{"selector":"#anim-efc903bf-bcb6-432e-b9e9-a3c509a1306e [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-34.249999904429615%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}] [{"selector":"#anim-52ddc9f8-8c56-402a-870b-a6610aa8596c","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-cf983689-1eba-45fc-8eb0-b8be120c112e","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]