model family
DeepSeek-OCR 2 logo

DeepSeek-OCR 2

Published by deepseek-ai. 1 member measured on real machines, across 1 build. Every number below was signed by the machine that produced it — none of it is from a model card.

what it takes, what it returns
image + text → text

Declared by the network’s catalog, not measured — a signature says what this family would be asked. Whether it answered is the signed capability flags on each member.

1 of 27 builds measured · 26 still open
DeepSeek-OCR 2 BASE1 build measured →
hf.co/sabafallah/DeepSeek-OCR-2-GGUF:Q4_K_Mbase · Q4_K_M · sabafallah · ollama · 3 machines · 2.93B
—recalled
—served
411.7best tok/s
411.7 tok/s · unnamed machine334.8 tok/s · unnamed machine15.4 tok/s · x86 · CPU · Hyderabad
15.4 → 411.7 tok/s across 3 machines · ×26.8 spreadunnamed machinex86 · CPU

Builds this network knows of

Declared, not measured — this is what the catalog says exists and what it would cost a machine to hold. Nothing here is signed, and none of it claims the build works.

how much memory does your machine have

27 of 27 builds fit a 16 GB machine — budgeting 70% of it, or 11.2 GiB, because the cache and the runtime want the rest.

  • optimum-intel-internal-testing/tiny-random-deepseek-ocr-2FP32vllmderived by optimum-intel-internal-testing
    0.1 GiB
  • hf.co/SandLogicTechnologies/DeepSeek-OCR-2-GGUF:IQ3_MIQ3_Mollamaquantized by SandLogicTechnologies
    1.3 GiB
  • hf.co/SandLogicTechnologies/DeepSeek-OCR-2-GGUF:IQ4_XSIQ4_XSollamaquantized by SandLogicTechnologies
    1.5 GiB
  • hf.co/sabafallah/DeepSeek-OCR-2-GGUF:Q4_K_MQ4_K_Mollamameasured herequantized by sabafallah
    1.8 GiB
  • hf.co/aditya00196/DeepSeek-OCR-2-Q4_K_M.gguf:Q4_K_MQ4_K_Mollamaderived by aditya00196
    1.8 GiB
  • hf.co/harishforaiandml/DeepSeek-OCR-2-Q4_K_M-GGUF:Q4_K_MQ4_K_Mollamaderived by harishforaiandml
    1.8 GiB
  • hf.co/aditya00196/DeepSeek-OCR-2-Q5_K_M.gguf:Q5_K_MQ5_K_Mollamaderived by aditya00196
    2.1 GiB
  • mlx-community/DeepSeek-OCR-2-4bit4BITmlxderived by mlx-community
    2.4 GiB
  • AutomatosX/AX-DeepSeek-OCR-2-CUDA-AXQ-NVFP4-W4A16W4A16vllmderived by AutomatosX
    2.8 GiB
  • AutomatosX/AX-DeepSeek-OCR-2-CUDA-AXQ-NVFP4-W4A4W4A4vllmderived by AutomatosX
    2.8 GiB
  • hf.co/sabafallah/DeepSeek-OCR-2-GGUF:Q8_0Q8_0ollamaquantized by sabafallah
    2.9 GiB
  • mlx-community/DeepSeek-OCR-2-6bit6BITmlxderived by mlx-community
    3.1 GiB
  • AutomatosX/AX-DeepSeek-OCR-2-MLX-AXQ-6bit6BITmlxderived by AutomatosX
    3.5 GiB
  • richarddavison/DeepSeek-OCR-2-FP8FP8vllmquantized by richarddavison
    3.5 GiB
  • mlx-community/DeepSeek-OCR-2-8bit8BITmlxderived by mlx-community
    3.8 GiB
  • AutomatosX/AX-DeepSeek-OCR-2-MLX-AXQ-MXFP88BITmlxderived by AutomatosX
    3.8 GiB
  • intel-ai/DeepSeek-OCR-2-MXFP8-CT-AutoRoundFP8vllmderived by intel-ai
    4.0 GiB
  • bartsolutions/DeepSeek-OCR-2-GPTQ-INT8GPTQvllmquantized by bartsolutions
    4.2 GiB
  • hf.co/sabafallah/DeepSeek-OCR-2-GGUF:BF16BF16ollamaquantized by sabafallah
    5.5 GiB
  • deepseek-ai/DeepSeek-OCR-2BF16vllmpublished by deepseek-ai
    6.3 GiB
  • AIArchiveInfo/DeepSeek-OCR-2BF16vllmderived by AIArchiveInfo
    6.3 GiB
  • deepseek-community/DeepSeek-OCR-2BF16vllmderived by deepseek-community
    6.3 GiB
  • Dogacel/Universal-DeepSeek-OCR-2BF16vllmderived by Dogacel
    6.3 GiB
  • loay/Arabic-OCR-DeepSeek-OCR-2FP16vllmderived by loay
    6.3 GiB
  • strangervisionhf/deepseek-ocr-2-transformers-v4.57.1BF16vllmderived by strangervisionhf
    6.3 GiB
  • TitanML/DeepSeek-OCR-2BF16vllmderived by TitanML
    6.3 GiB
  • unsloth/DeepSeek-OCR-2BF16vllmderived by unsloth
    6.3 GiB

Sizes are what the catalog declares, not what a machine measured — fitting on disk is necessary and not sufficient. Nobody here has run these on your machine, and this narrows the candidates rather than certifying one.

the same builds, by member
DeepSeek-OCR 2 BASE
hf.co/SandLogicTechnologies/DeepSeek-OCR-2-GGUF:IQ3_MIQ3_Mollama1.3 GiB on diskquantized by SandLogicTechnologies
hf.co/SandLogicTechnologies/DeepSeek-OCR-2-GGUF:IQ4_XSIQ4_XSollama1.5 GiB on diskquantized by SandLogicTechnologies
hf.co/sabafallah/DeepSeek-OCR-2-GGUF:Q4_K_MQ4_K_Mollama1.8 GiB on diskquantized by sabafallah
hf.co/sabafallah/DeepSeek-OCR-2-GGUF:Q8_0Q8_0ollama2.9 GiB on diskquantized by sabafallah
richarddavison/DeepSeek-OCR-2-FP8FP8vllm3.5 GiB on diskquantized by richarddavison
bartsolutions/DeepSeek-OCR-2-GPTQ-INT8GPTQvllm4.2 GiB on diskquantized by bartsolutions
hf.co/sabafallah/DeepSeek-OCR-2-GGUF:BF16BF16ollama5.5 GiB on diskquantized by sabafallah
deepseek-ai/DeepSeek-OCR-2BF16vllm6.3 GiB on diskpublished by deepseek-ai
DeepSeek-OCR-2-Q5_K_M.gguf
hf.co/aditya00196/DeepSeek-OCR-2-Q4_K_M.gguf:Q4_K_MQ4_K_Mollama1.8 GiB on diskderived by aditya00196
hf.co/aditya00196/DeepSeek-OCR-2-Q5_K_M.gguf:Q5_K_MQ5_K_Mollama2.1 GiB on diskderived by aditya00196
DeepSeek-OCR-2
AIArchiveInfo/DeepSeek-OCR-2BF16vllm6.3 GiB on diskderived by AIArchiveInfo
AX-DeepSeek-OCR-2-MLX-AXQ-6bit
AutomatosX/AX-DeepSeek-OCR-2-MLX-AXQ-6bit6BITmlx3.5 GiB on diskderived by AutomatosX
AutomatosX/AX-DeepSeek-OCR-2-MLX-AXQ-MXFP88BITmlx3.8 GiB on diskderived by AutomatosX
AX-DeepSeek-OCR-2-CUDA-AXQ-NVFP4-W4A16
AutomatosX/AX-DeepSeek-OCR-2-CUDA-AXQ-NVFP4-W4A16W4A16vllm2.8 GiB on diskderived by AutomatosX
AutomatosX/AX-DeepSeek-OCR-2-CUDA-AXQ-NVFP4-W4A4W4A4vllm2.8 GiB on diskderived by AutomatosX
DeepSeek-OCR-2
deepseek-community/DeepSeek-OCR-2BF16vllm6.3 GiB on diskderived by deepseek-community
Universal-DeepSeek-OCR-2
Dogacel/Universal-DeepSeek-OCR-2BF16vllm6.3 GiB on diskderived by Dogacel
DeepSeek-OCR-2-Q4_K_M-GGUF
hf.co/harishforaiandml/DeepSeek-OCR-2-Q4_K_M-GGUF:Q4_K_MQ4_K_Mollama1.8 GiB on diskderived by harishforaiandml
DeepSeek-OCR-2-MXFP8-CT-AutoRound
intel-ai/DeepSeek-OCR-2-MXFP8-CT-AutoRoundFP8vllm4.0 GiB on diskderived by intel-ai
Arabic-OCR-DeepSeek-OCR-2
loay/Arabic-OCR-DeepSeek-OCR-2FP16vllm6.3 GiB on diskderived by loay
DeepSeek-OCR-2-8bit
mlx-community/DeepSeek-OCR-2-4bit4BITmlx2.4 GiB on diskderived by mlx-community
mlx-community/DeepSeek-OCR-2-6bit6BITmlx3.1 GiB on diskderived by mlx-community
mlx-community/DeepSeek-OCR-2-8bit8BITmlx3.8 GiB on diskderived by mlx-community
deepseek-ocr-2-transformers-v4.57.1
strangervisionhf/deepseek-ocr-2-transformers-v4.57.1BF16vllm6.3 GiB on diskderived by strangervisionhf
DeepSeek-OCR-2
TitanML/DeepSeek-OCR-2BF16vllm6.3 GiB on diskderived by TitanML
DeepSeek-OCR-2
unsloth/DeepSeek-OCR-2BF16vllm6.3 GiB on diskderived by unsloth
tiny-random-deepseek-ocr-2
optimum-intel-internal-testing/tiny-random-deepseek-ocr-2FP32vllm131 MiB on diskderived by optimum-intel-internal-testing
ask your own gram

Your concierge can answer this from the network rather than from this page. The ask carries the operator id, so it resolves to a capability and an engine — not to a name somebody has to recognise.

What has this network measured about the DeepSeek-OCR 2 family (epn.family.deepseek-ocr-2), and can my gram run any of it? See /inferences/deepseek-ocr-2

open a gramx with this ask →epn.family.deepseek-ocr-2