v2.104.0

dream-num/univerv2.104.0Jun 19, 2026by docling-ops[bot]

AI Summary

This release adds vLLM support for OCR models and introduces confidence scores to service response models. It also includes fixes for device selection, Excel page range handling, and installation instructions.

Key Highlights

vLLM support for chandra/glm-ocr
Dots bbox fix
ConfidenceScores added to service response models
Align DoclingServiceClient convert_all arg names
Respect page_range in MsExcelDocumentBackend

New Features

vLLM support for chandra/glm-ocr
ConfidenceScores to service response models

Full Release Notes

### Feature

* **service:** Add ConfidenceScores to service response models ([#3651](https://github.com/docling-project/docling/issues/3651)) ([`166003b`](https://github.com/docling-project/docling/commit/166003b64aefd18447fed1513445facb33d732ce))
* Add vLLM support for chandra/glm-ocr + dots bbox fix ([#3404](https://github.com/docling-project/docling/issues/3404)) ([`2c875ca`](https://github.com/docling-project/docling/commit/2c875ca0be3520f73e84acee1c5d7175206eb9a8))

### Fix

* **service:** Align DoclingServiceClient.convert_all arg names ([#3650](https://github.com/docling-project/docling/issues/3650)) ([`f38e222`](https://github.com/docling-project/docling/commit/f38e222964fdfa659b5d10cc399e83750b9ffa48))
* Make explicit device selection fail hard when device unavailable ([#3606](https://github.com/docling-project/docling/issues/3606)) ([`3117bd1`](https://github.com/docling-project/docling/commit/3117bd1c76f94ed75bd3754a97b6fd8bb0609aa0))
* #1649 Update RTX installation instructions for PyTorch ([#3644](https://github.com/docling-project/docling/issues/3644)) ([`f847b6c`](https://github.com/docling-project/docling/commit/f847b6ccc4859aeabe07a125fffac8d70b135826))
* **xlsx:** Respect page_range in MsExcelDocumentBackend ([#3637](https://github.com/docling-project/docling/issues/3637)) ([`dc61cdd`](https://github.com/docling-project/docling/commit/dc61cdd505e4dcd511cac40f016cf39d4609889e))