Dolphin-v2-f32-GGUF

ByteDance Dolphin-v2 is a 3B-parameter vision-language model built on Qwen2.5-VL-3B with Native Resolution Vision Transformer (NaViT) encoder and autoregressive decoder, designed as a universal document parsing solution via a document-type-aware two-stage architecture that classifies digital-born vs. photographed documents before applying hybrid strategiesโ€”element-wise parallel parsing for clean PDFs and holistic parsing for distorted scans. It supports 21 element categories (headings sec_0-5, paragraphs, formulas in LaTeX, HTML tables, indented code blocks, figures, lists, etc.) with absolute pixel coordinates for precise localization, achieving state-of-the-art OmniDocBench v1.5 scores of 89.45 overall (+14.78 over original Dolphin), 0.054 edit distance for text/reading order, 86.72% CDM for formulas, and 87.02/90.48 TEDS/TEDS-S for tables at 0.1729 FPS on 8-12GB VRAM GPUs. Specialized modules (P_formula, P_code, P_table, P_paragraph) enable structured JSON/Markdown/HTML outputs for privacy-focused local inference in healthcare/legal/finance, outperforming general VLMs in speed (2x faster) and accuracy across distortions, skews, and perspectives.

Dolphin-v2 [GGUF]

File Name Quant Type File Size File Link
Dolphin-v2.BF16.gguf BF16 6.18 GB Download
Dolphin-v2.F32.gguf F32 12.3 GB Download
Dolphin-v2.IQ4_XS.gguf IQ4_XS 1.75 GB Download
Dolphin-v2.Q2_K.gguf Q2_K 1.27 GB Download
Dolphin-v2.Q3_K_L.gguf Q3_K_L 1.71 GB Download
Dolphin-v2.Q3_K_M.gguf Q3_K_M 1.59 GB Download
Dolphin-v2.Q3_K_S.gguf Q3_K_S 1.45 GB Download
Dolphin-v2.Q4_K_M.gguf Q4_K_M 1.93 GB Download
Dolphin-v2.Q4_K_S.gguf Q4_K_S 1.83 GB Download
Dolphin-v2.Q5_K_M.gguf Q5_K_M 2.22 GB Download
Dolphin-v2.Q5_K_S.gguf Q5_K_S 2.17 GB Download
Dolphin-v2.Q6_K.gguf Q6_K 2.54 GB Download
Dolphin-v2.Q8_0.gguf Q8_0 3.29 GB Download
Dolphin-v2.f16.gguf F16 6.18 GB Download
Dolphin-v2.i1-IQ1_M.gguf i1-IQ1_M 850 MB Download
Dolphin-v2.i1-IQ1_S.gguf i1-IQ1_S 791 MB Download
Dolphin-v2.i1-IQ2_M.gguf i1-IQ2_M 1.14 GB Download
Dolphin-v2.i1-IQ2_S.gguf i1-IQ2_S 1.06 GB Download
Dolphin-v2.i1-IQ2_XS.gguf i1-IQ2_XS 1.03 GB Download
Dolphin-v2.i1-IQ2_XXS.gguf i1-IQ2_XXS 948 MB Download
Dolphin-v2.i1-IQ3_M.gguf i1-IQ3_M 1.49 GB Download
Dolphin-v2.i1-IQ3_S.gguf i1-IQ3_S 1.46 GB Download
Dolphin-v2.i1-IQ3_XS.gguf i1-IQ3_XS 1.39 GB Download
Dolphin-v2.i1-IQ3_XXS.gguf i1-IQ3_XXS 1.28 GB Download
Dolphin-v2.i1-IQ4_NL.gguf i1-IQ4_NL 1.83 GB Download
Dolphin-v2.i1-IQ4_XS.gguf i1-IQ4_XS 1.74 GB Download
Dolphin-v2.i1-Q2_K.gguf i1-Q2_K 1.27 GB Download
Dolphin-v2.i1-Q2_K_S.gguf i1-Q2_K_S 1.2 GB Download
Dolphin-v2.i1-Q3_K_L.gguf i1-Q3_K_L 1.71 GB Download
Dolphin-v2.i1-Q3_K_M.gguf i1-Q3_K_M 1.59 GB Download
Dolphin-v2.i1-Q3_K_S.gguf i1-Q3_K_S 1.45 GB Download
Dolphin-v2.i1-Q4_0.gguf i1-Q4_0 1.83 GB Download
Dolphin-v2.i1-Q4_1.gguf i1-Q4_1 2 GB Download
Dolphin-v2.i1-Q4_K_M.gguf i1-Q4_K_M 1.93 GB Download
Dolphin-v2.i1-Q4_K_S.gguf i1-Q4_K_S 1.83 GB Download
Dolphin-v2.i1-Q5_K_M.gguf i1-Q5_K_M 2.22 GB Download
Dolphin-v2.i1-Q5_K_S.gguf i1-Q5_K_S 2.17 GB Download
Dolphin-v2.i1-Q6_K.gguf i1-Q6_K 2.54 GB Download
Dolphin-v2.imatrix.gguf imatrix 3.39 MB Download
Dolphin-v2.mmproj-Q8_0.gguf mmproj-Q8_0 848 MB Download
Dolphin-v2.mmproj-bf16.gguf mmproj-bf16 1.34 GB Download
Dolphin-v2.mmproj-f16.gguf mmproj-f16 1.34 GB Download
Dolphin-v2.mmproj-f32.gguf mmproj-f32 2.67 GB Download

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):

image.png

Downloads last month
-
GGUF
Model size
3B params
Architecture
qwen2vl
Hardware compatibility
Log In to view the estimation

1-bit

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

32-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for prithivMLmods/Dolphin-v2-f32-GGUF

Quantized
(4)
this model