Qwen3.5 extraction architecture
The captured configuration identifies Qwen3_5ForConditionalGeneration with model type qwen3_5 and Transformers support.
Open Source Model Profile · datalab-to
lift
lift is a 9.65B-parameter Qwen3.5-family extraction model from Datalab. According to the model card, it returns schema-matching JSON from PDFs and images.
- Publisher
- datalab-to
- Task
- image-text-to-text
- Model type
- qwen3_5
- License
- openrail
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
lift is published by datalab-to as an image-text-to-text structured-extraction model. The captured configuration identifies Qwen3_5ForConditionalGeneration, and Safetensors metadata reports about 9.65B parameters under openrail. According to the model card, it converts PDFs and images into schema-matching JSON.
Recorded capabilities
Schema-constrained extraction
According to the model card, any JSON schema can be supplied and lift returns a matching JSON object with schema-constrained decoding.
Multi-page document handling
According to the model card, multi-page documents are handled in a single pass, including values spanning pages.
Local and remote inference modes
According to the model card, two inference modes are offered: local Hugging Face and remote vLLM server, with a CLI and Schema Studio app.
Published comparison table
According to the model card, lift is compared against the Datalab API, Gemini Flash 3.5, Azure Content Understanding, NuExtract3, and Qwen3.5-9B on field and document accuracy plus latency.
Use cases in the source record
- Structured JSON extraction from PDFs and images against user-supplied JSON schemas.
- Batch document processing through the documented CLI and schema-testing workflows.
Limitations and unknowns
- Accuracy and latency figures are publisher-reported comparisons and were not independently measured by Ethen.
- According to the model card, weight use is limited for competitors and larger commercial use, so commercial terms need direct license review.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: datalab-to/lift
Captured: Unknown. Processed: 2026-09-07T19:34:43.070720+00:00.
lift lift is a structured extraction model from Datalab that pulls structured JSON out of PDFs and images. Pass any JSON schema and lift returns a JSON object matching it, using schema-constrained decoding to guarantee valid, well-typed output. Try lift in the free playground , or use the hosted API for higher accuracy, per-field verification, and citations. Features Extract structured data from documents Pass any JSON schema Handles multi-page documents in a single pass, including values that span pages Two inference modes: local (HuggingFace) and remote (vLLM server) CLI for single files, inline schemas, or whole directories Schem…
F001F002F003F004F005F006F007F008F009F010F011F013F015F016