cstr/internvl2-1b-crispembed-GGUF

InternVL2-1B — CrispEmbed GGUF

GGUF conversions of OpenGVLab/InternVL2-1B for use with CrispEmbed.

Smallest competitive VLM for OCR — ideal for edge, mobile, and WASM deployment.

Model Details

Property	Value
Architecture	InternVL2 (InternViT-300M + Qwen2-0.5B)
Total Parameters	~0.9B
Vision Encoder	InternViT-300M-448px (24L, 1024d, identical to InternVL2.5-2B)
Projector	Pixel unshuffle (4:1) + LayerNorm + Linear + GELU + Linear
LLM Decoder	Qwen2-0.5B-Instruct (24L, 896d, GQA 14/2, SwiGLU, RMSNorm)
Input Resolution	448x448 per tile, dynamic tiling (1-12 tiles)
License	MIT
OCRBench	779

Available Quantizations

File	Size	Compression	Notes
internvl2-1b-f16.gguf	2.3 GB	1x	Full precision
internvl2-1b-q8_0.gguf	955 MB	2.4x	Good quality
internvl2-1b-q4_k.gguf	~600 MB	~4x	Smallest, vision Q8_0 floor

Parity Verification

All components verified against Python reference (cos=1.000000):

• Vision encoder: 4/4 layers PASS
• Projector: PASS
• LLM decoder (Qwen2): 2/2 layers PASS

Credits

• Original model: OpenGVLab/InternVL2-1B (MIT)
• GGUF conversion: CrispEmbed

Provenance and EU AI Act Art. 53 note

• Upstream model: OpenGVLab/InternVL2-1B — published by OpenGVLab.
• Upstream licence: mit. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.