TesseractJs.Models (tesseract_js v0.1.0)

Copy Markdown View Source

Model registry — single source of truth for the languages and core WASM files that tesseract_js knows about. Both CDN mode and local mode are driven from this module.

The curated list below covers ~20 common languages with checksums and approximate sizes for the :standard tessdata tier. Any language code (e.g. "hin", "chi_tra+chi_sim") that is not in the curated list still works at runtime — it just falls through to the URL template without checksum verification.

Use the helpers to resolve URLs and paths:

TesseractJs.Models.cdn_url("eng")
TesseractJs.Models.cdn_url("jpn_vert", :best)
TesseractJs.Models.local_path("eng")
TesseractJs.Models.filename("eng")

The tessdata-version pinned by this package is 4.0.0 (the tessdata bundle format used by tesseract.js 5.x). Bump it in lockstep with the JS core release.

Summary

Functions

cdn_url(lang, tier \\ :standard)

Builds the jsDelivr URL for a language's traineddata file.

core_cdn_url(variant \\ :simd_lstm)

jsDelivr URL for the tesseract.js-core WASM bundle.

core_filename(variant \\ :simd_lstm)

Filename for a WASM core variant.

core_local_path(variant \\ :simd_lstm, base \\ "/assets/vendor/tesseract")

Local path for the WASM core.

core_version()

Returns the tesseract.js-core version this package is pinned to.

filename(lang)

Filename for a language's traineddata file.

get(lang)

Returns the registry entry for a language, or nil if not curated.

list()

Returns the curated registry as a map of lang => %{name, size_mb, sha256}.

local_path(lang, base \\ "/assets/vendor/tesseract")

Local filesystem path (relative to a Phoenix app's priv/static/) where the Mix download task will write a language file.

split_langs(lang)

Splits a +-combined lang string into individual lang codes.

tessdata_version()

Returns the tessdata version (4.0.0) this package is pinned to.

Functions

cdn_url(lang, tier \\ :standard)

@spec cdn_url(String.t(), atom()) :: String.t()

Builds the jsDelivr URL for a language's traineddata file.

iex> TesseractJs.Models.cdn_url("eng")
"https://cdn.jsdelivr.net/npm/@tesseract.js-data/eng@1.0.0/4.0.0/eng.traineddata.gz"

Supports +-combined langs by returning the URL for the first lang — the consumer is expected to download each lang separately. (For local mode, all combined langs must be present in the same dir.)

Tiers

  • :standard — full LSTM+legacy combined model, ~11 MB/lang gzipped.
  • :best — smaller LSTM-only model (the _best_int variant on jsDelivr), ~3 MB/lang gzipped, similar accuracy to standard for most languages.

The :fast tier (tessdata_fast) requires uncompressed .traineddata files served from a different source and isn't supported in v0.1.

core_cdn_url(variant \\ :simd_lstm)

@spec core_cdn_url(atom()) :: String.t()

jsDelivr URL for the tesseract.js-core WASM bundle.

core_filename(variant \\ :simd_lstm)

@spec core_filename(atom()) :: String.t()

Filename for a WASM core variant.

core_local_path(variant \\ :simd_lstm, base \\ "/assets/vendor/tesseract")

@spec core_local_path(atom(), Path.t()) :: String.t()

Local path for the WASM core.

core_version()

Returns the tesseract.js-core version this package is pinned to.

filename(lang)

@spec filename(String.t()) :: String.t()

Filename for a language's traineddata file.

get(lang)

@spec get(String.t()) :: map() | nil

Returns the registry entry for a language, or nil if not curated.

list()

@spec list() :: %{required(String.t()) => map()}

Returns the curated registry as a map of lang => %{name, size_mb, sha256}.

local_path(lang, base \\ "/assets/vendor/tesseract")

@spec local_path(String.t(), Path.t()) :: String.t()

Local filesystem path (relative to a Phoenix app's priv/static/) where the Mix download task will write a language file.

iex> TesseractJs.Models.local_path("eng")
"/assets/vendor/tesseract/eng.traineddata.gz"

split_langs(lang)

@spec split_langs(String.t()) :: [String.t()]

Splits a +-combined lang string into individual lang codes.

tessdata_version()

Returns the tessdata version (4.0.0) this package is pinned to.