Hugging Face's logo

Spaces:
interfaze-ai
/
README
Configuration error

App Files Community
README / README.md
yoeven's picture
yoeven
Update README.md
abc01c2 verified
preview code
|
raw
history blame contribute delete
8.24 kB
metadata 
title: Interfaze
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/62fb53b572a7ab50b4b06fca/II4KdeJkepE_NOg9HNnq_.jpeg
short_description: The AI model built for deterministic developer tasks

Extraction

The AI model built for deterministic developer tasks

Interfaze is an AI model built on a new architecture that merges specialized DNN/CNN models with LLMs for developer tasks that require deterministic output and high consistency like OCR, scraping, classification, web search and more.

Try now or Read paper

  • OCR, web scraping, web search, classification and more
  • OpenAI chat completion API compatible
  • High accuracy structured output consistency
  • Built-in code execution and sandboxing
  • Custom web engine for scraping and web research capabilities
  • Auto reasoning when needed
  • Controllable guardrails
  • Fully managed and scalable
  • Globally distributed fallback system with high uptime

Model Comparison

Benchmark interfaze-beta GPT-4.1 Claude Sonnet 4 Gemini 2.5 Flash Claude Sonnet 4 (Thinking) Claude Opus 4 (Thinking) GPT-5-Minimal Gemini-2.5-Pro
MMLU-Pro 83.6 80.6 83.7 80.9 83.7 86 80.6 86.2
MMLU 91.38 90.2 - - 88.8 89 - 89.2
MMMU 77.33 74.8 - 79.7 74.4 76.5 - 82
AIME-2025 90 34.7 38 60.3 74.3 73.3 31.7 87.7
GPQA-Diamond 81.31 66.3 68.3 68.3 77.7 79.6 67.3 84.4
LiveCodeBench 57.77 45.7 44.9 49.5 65.5 63.6 55.8 75.9
ChartQA 90.88 - - - - - - -
AI2D 91.51 85.9 - - - - - 89.5
Common-Voice-v16 90.8 - - - - - - -

*Results for Non-Interfaze models are sourced from model providers, leaderboards, and evaluation providers such as Artificial Analysis.

Works like any other LLM

OpenAI API compatible, works with every AI SDK out of the box

import OpenAI from "openai";

const interfaze = new OpenAI({
    baseURL: "https://api.interfaze.ai/v1",
    apiKey: "<your-api-key>"
});

const completion = await interfaze.chat.completions.create({
    model: "interfaze-beta",
    messages: [\
        {\
            role: "user",\
            content: "Get the company description of JigsawStack from their linkedin page",\
        },\
    ],
});

console.log(completion.choices[0].message.content);

OCR & Document Extraction

vision docs ->

prompt = "Get the person information from the following ID."

schema = z.object({
  first_name: z.string(),
  last_name: z.string(),
  dob: z.string(),
  expiry: z.string(),
});

Extraction

Smart Web Scraping

web docs ->

prompt = "Extract the information from Yoeven D Khemlani's linkedin page based on the schema."

schema = z.object({
  first_name: z.string(),
  last_name: z.string(),
  about: z.string(),
  current_company: z.string(),
  current_position: z.string(),
});

scraping

Translation

translation docs ->

prompt = "The UK drinks about 100–160 million cups of tea every day, and 98% of tea drinkers add milk to their tea."

schema = z.object({
  zh: z.string(),
  hi: z.string(),
  es: z.string(),
  fr: z.string(),
  de: z.string(),
  it: z.string(),
  ja: z.string(),
  ko: z.string(),
});

zh: 英国每天饮用约100–160百万杯茶，有98%的茶饮者在茶中加入牛奶。
hi: यूके हर दिन लगभग 100–160 मिलियन कप चाय पीता है, और 98% चाय पीने वाले अपनी चाय में दूध मिलाते हैं।
es: El Reino Unido bebe alrededor de 100–160 millones de tazas de té cada día, y el 98 % de los consumidores de té añade leche a su té.
fr: Le Royaume-Uni boit environ 100–160 millions de tasses de thé chaque jour, et 98 % des buveurs de thé ajoutent du lait à leur thé.
de: Das Vereinigte Königreich trinkt etwa 100–160 Millionen Tassen Tee pro Tag, und 98 % der Teetrinker fügen ihrem Tee Milch hinzu.
it: Il Regno Unito beve circa 100–160 milioni di tazze di tè ogni giorno e il 98% degli amanti del tè aggiunge latte al proprio tè.
ja: イギリスでは毎日約100～160百万杯の紅茶が飲まれており、紅茶を飲む人の98%が紅茶に牛乳を加えます。
ko: 영국에서는 매일 약 1억 ~ 1억 6천만 잔의 차를 마시며, 차를 마시는 사람의 98%가 차에 우유를 넣습니다.

Speech-to-text (STT) and diarization

stt docs ->

prompt = "Transcribe https://jigsawstack.com/preview/stt-example.wav"

schema = z.object({
  text: z.string(),
  speakers: z.object({
    id: z.string(),
    start: z.number(),
    end: z.number()
  })
});

{
  "text": " The little tales they tell are false The door was barred, locked and bolted as well Ripe pears are fit for a queen's table A big wet stain was on the round carpet The kite dipped and swayed but stayed aloft The pleasant hours fly by much too soon The room was crowded with a mild wob The room was crowded with a wild mob This strong arm shall shield your honour She blushed when he gave her a white orchid The beetle droned in the hot June sun",
  "speakers": [\
    {\
      "start":0,\
      "end":4.78,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":4.78,\
      "end":9.48,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":9.48,\
      "end":13.06,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":13.06,\
      "end":17.24,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":17.24,\
      "end":21.78,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":21.78,\
      "end":26.3,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":26.3,\
      "end":30.76,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":30.76,\
      "end":35.08,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":35.08,\
      "end":39.24,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":39.24,\
      "end":43.94,\
      "id": "SPEAKER_00"\
    },\
    {\
      "start":43.94,\
      "end":48.5,\
      "id": "SPEAKER_00"\
    }\
  ]
}

Configurable guardrails and NSFW checks

guardrails docs ->

Fully configurable guardrails for text and images

S1: Violent Crimes
S2: Non-Violent Crimes
S3: Sex-Related Crimes
S4: Child Sexual Exploitation
S5: Defamation
S6: Specialized Advice
S7: Privacy
S8: Intellectual Property
S9: Indiscriminate Weapons
S10: Hate
S11: Suicide & Self-Harm
S12: Sexual Content
S12_IMAGE: Sexual Content (Image)
S13: Elections
S14: Code Interpreter Abuse

Architecture

read paper ->

This architecture combines a suite of small specialized models supported with custom tools and infrastructure while automatically routing to the best model for the task that prioritizes accuracy and speed.

How it works

Specs

  • Context window: 1m tokens
  • Max output tokens: 32k tokens
  • Input modalities: Text, Images, Audio, File, Video
  • Reasoning: Available

Research references

Who are we?

We are a small team of ML, Software and Infrastructure engineers engrossed in the fact that a small model can do a lot more when specialized. Allowing us to make AI available in every dev workflow.