@jarkkojs/readseek 0.4.33 → 0.4.34

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +14 -10
  2. package/package.json +4 -4
package/README.md CHANGED
@@ -50,19 +50,20 @@ readseek identify src/main.rs:42 --column 8 | readseek def --from-identify src -
50
50
  ## Images
51
51
 
52
52
  `detect` reports format, dimensions, and animation status for images. Add a vision
53
- flag to analyze image contents with the Qwen3-VL vision model:
53
+ flag to analyze image contents with the BLIP, YOLOv8-nano, and TrOCR models:
54
54
 
55
55
  ```sh
56
- readseek detect screenshot.png --transcribe # text + per-region bounding quads
57
56
  readseek detect photo.jpg --caption # detailed natural-language caption
58
57
  readseek detect photo.jpg --objects # object labels + bounding boxes
58
+ readseek detect photo.jpg --ocr # extracted text
59
59
  ```
60
60
 
61
- The flags can be combined; the model loads once per invocation. The model files
62
- (~1.1 GB GGUF + ~0.8 GB multimodal projection) are downloaded lazily into the
63
- user cache directory on first vision use and reused on subsequent runs; a
64
- progress bar is shown while downloading when stdout is an interactive TTY.
65
- Inference is CPU-only and takes a few to tens of seconds per image.
61
+ The flags can be combined; each model loads once per invocation. The model files
62
+ (~258 MB BLIP GGUF + ~6 MB YOLOv8-nano + ~1.24 GB TrOCR) are downloaded lazily
63
+ into the user cache directory on first vision use and reused on subsequent runs;
64
+ a progress bar is shown while downloading when stdout is an interactive TTY.
65
+ Inference is CPU-only; object detection and OCR take seconds and captioning up to
66
+ a couple of minutes per image.
66
67
 
67
68
  ## Cache
68
69
 
@@ -88,6 +89,9 @@ native binaries are licensed under `LGPL-2.1-or-later`. Corresponding source for
88
89
  each published native binary is available from the GitHub repository tag that
89
90
  matches the package version.
90
91
 
91
- readseek downloads the Qwen3-VL vision model (`Qwen/Qwen3-VL-2B-Instruct-GGUF`,
92
- quantized from `Qwen/Qwen3-VL-2B-Instruct`) into the user cache directory on
93
- first use. It is licensed under `Apache-2.0`.
92
+ readseek downloads the BLIP caption model (a quantized GGUF from
93
+ `lmz/candle-blip`, with the tokenizer from `Salesforce/blip-image-captioning-large`),
94
+ the YOLOv8-nano object-detection model (`lmz/candle-yolo-v8`), and the TrOCR
95
+ printed-text model (`microsoft/trocr-base-printed`) into the `models/` subdirectory
96
+ of the user cache on first use. BLIP is licensed under `BSD-3-Clause`; YOLOv8-nano
97
+ is derived from Ultralytics YOLOv8 (`AGPL-3.0`); TrOCR is released by Microsoft.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@jarkkojs/readseek",
3
- "version": "0.4.33",
3
+ "version": "0.4.34",
4
4
  "description": "A structural read command",
5
5
  "license": "Apache-2.0 AND LGPL-2.1-or-later",
6
6
  "homepage": "https://github.com/jarkkojs/readseek#readme",
@@ -21,9 +21,9 @@
21
21
  "LICENSE-LGPL-2.1"
22
22
  ],
23
23
  "optionalDependencies": {
24
- "@jarkkojs/readseek-darwin-arm64": "0.4.33",
25
- "@jarkkojs/readseek-linux-x64": "0.4.33",
26
- "@jarkkojs/readseek-win32-x64": "0.4.33"
24
+ "@jarkkojs/readseek-darwin-arm64": "0.4.34",
25
+ "@jarkkojs/readseek-linux-x64": "0.4.34",
26
+ "@jarkkojs/readseek-win32-x64": "0.4.34"
27
27
  },
28
28
  "publishConfig": {
29
29
  "access": "public"