Commits · LLDDWW/MedCard

Use PaddleOCR predict API and normalize inputs

a97e706

Running

LLDDWW commited on Oct 13

Handle new PaddleOCR output format

2879fbc

LLDDWW commited on Oct 13

Specify PaddleOCR recognition model name

ea3177c

LLDDWW commited on Oct 13

Add paddlepaddle dependency

310be32

LLDDWW commited on Oct 13

Adjust PaddleOCR args for new API

ae810a8

LLDDWW commited on Oct 13

Configure PaddleOCR to use Korean PP-OCRv5

26e52a1

LLDDWW commited on Oct 13

Update requirements for PaddleOCR

d6bf509

LLDDWW commited on Oct 13

Switch OCR engine to PaddleOCR

7a2cdb5

LLDDWW commited on Oct 13

perf: switch to Gemma-2-2B for faster inference

63c2769

LLDDWW Claude commited on Oct 13

perf: replace VLM with EasyOCR for ultra-fast Korean OCR

dcb7540

LLDDWW Claude commited on Oct 13

fix: upgrade to Qwen2.5-VL-3B with 8bit quantization

7fabc42

LLDDWW Claude commited on Oct 13

perf: switch to faster Qwen2-VL-2B for OCR

e96841e

LLDDWW Claude commited on Oct 10

perf: preload models at startup for faster inference

ab48ca2

LLDDWW Claude commited on Oct 10

fix: strip whitespace from HF_TOKEN

115fae5

LLDDWW Claude commited on Oct 1

chore: upgrade to MedGemma-4B-IT for medical analysis

f89c606

LLDDWW Claude commited on Oct 1

chore: upgrade to MedGemma-4B-IT for medical analysis

7d833fa

LLDDWW Claude commited on Oct 1

feat: use Gemma-2-2B for medical analysis

d5aff0d

LLDDWW Claude commited on Oct 1

feat: add medication analysis with easy explanations

0e6a905

LLDDWW Claude commited on Oct 1

chore: upgrade to Qwen2.5-VL-7B-Instruct

8a7e01c

LLDDWW Claude commited on Oct 1

fix: use correct Qwen2_5_VL class for Qwen2.5-VL model

c5fa8ab

LLDDWW Claude commited on Oct 1

chore: switch to Qwen2.5-VL-3B-Instruct for OCR

664b899

LLDDWW Claude commited on Oct 1

feat: upgrade to Qwen2-VL-2B for state-of-the-art OCR

28a7334

LLDDWW Claude commited on Oct 1

fix: move EasyOCR initialization inside GPU decorator

0ffc67e

LLDDWW Claude commited on Oct 1

feat: switch from PaddleOCR to EasyOCR

33831f7

LLDDWW Claude commited on Oct 1

fix: update PaddleOCR initialization parameters

70f28f9

LLDDWW Claude commited on Oct 1

feat: switch to PaddleOCR for better Korean text recognition

f16cb1a

LLDDWW Claude commited on Oct 1

refactor: simplify to OCR-only output

39446f7

LLDDWW Claude commited on Oct 1

chore: switch to Korean-optimized TrOCR model

7138a91

LLDDWW Claude commited on Oct 1

fix: match input dtype to model dtype for TrOCR

c74ab95

LLDDWW Claude commited on Oct 1

fix: remove torch_dtype to fix TrOCR dtype mismatch

e4a5da6

LLDDWW Claude commited on Oct 1

chore: switch OCR model to TrOCR

6a2327c

LLDDWW Claude commited on Oct 1

sdfdsfads23333

e94e117

LLDDWW commited on Oct 1

sdfdsfads23333

92bb45b

LLDDWW commited on Oct 1

sdfdsfads2333

64c57fb

LLDDWW commited on Oct 1

sdfdsfads233

52bda02

LLDDWW commited on Oct 1

refactor: simplify to medication name extractor only

7ce0e26

LLDDWW Claude commited on Oct 1

sdfdsfads23

ea6e0a5

LLDDWW commited on Oct 1

sdfdsfadsㄴ

e5ef627

LLDDWW commited on Oct 1

sdfdsfads

bfe2214

LLDDWW commited on Sep 30

fix: resolve 401 error by using public Qwen2-VL-7B model with ultra quality optimizations

24e39c0

LLDDWW Claude commited on Sep 30

sdf

736677f

LLDDWW commited on Sep 30

feat: upgrade to 72B model with 8-bit quantization for maximum quality

e5f80e3

LLDDWW Claude commited on Sep 30

fix: change to VL-7B for ZeroGPU compatibility

84e5d9f

LLDDWW commited on Sep 30

feat: complete redesign with single VL-32B model and modern UI

98b216f

LLDDWW Claude commited on Sep 30

feat: add real-time web search for drug verification

3cec557

LLDDWW commited on Sep 30

feat: ultra-quality improvement for production-ready app

8020bc3

LLDDWW Claude commited on Sep 30

feat: add professional drug information system

ad98685

LLDDWW Claude commited on Sep 30

fix: change image model to SDXL (FLUX access restricted)

88cbfb6

LLDDWW commited on Sep 30

feat: optimize performance and improve UX (Phase 1)

5e231af

LLDDWW Claude commited on Sep 30

feat: upgrade models and improve quality

8a13800

LLDDWW Claude commited on Sep 30

Commit History

Use PaddleOCR predict API and normalize inputs a97e706 Running

Handle new PaddleOCR output format 2879fbc

Specify PaddleOCR recognition model name ea3177c

Add paddlepaddle dependency 310be32

Adjust PaddleOCR args for new API ae810a8

Configure PaddleOCR to use Korean PP-OCRv5 26e52a1

Update requirements for PaddleOCR d6bf509

Switch OCR engine to PaddleOCR 7a2cdb5

perf: switch to Gemma-2-2B for faster inference 63c2769

perf: replace VLM with EasyOCR for ultra-fast Korean OCR dcb7540

fix: upgrade to Qwen2.5-VL-3B with 8bit quantization 7fabc42

perf: switch to faster Qwen2-VL-2B for OCR e96841e

perf: preload models at startup for faster inference ab48ca2

fix: strip whitespace from HF_TOKEN 115fae5

chore: upgrade to MedGemma-4B-IT for medical analysis f89c606

chore: upgrade to MedGemma-4B-IT for medical analysis 7d833fa

feat: use Gemma-2-2B for medical analysis d5aff0d

feat: add medication analysis with easy explanations 0e6a905

chore: upgrade to Qwen2.5-VL-7B-Instruct 8a7e01c

fix: use correct Qwen2_5_VL class for Qwen2.5-VL model c5fa8ab

chore: switch to Qwen2.5-VL-3B-Instruct for OCR 664b899

feat: upgrade to Qwen2-VL-2B for state-of-the-art OCR 28a7334

fix: move EasyOCR initialization inside GPU decorator 0ffc67e

feat: switch from PaddleOCR to EasyOCR 33831f7

fix: update PaddleOCR initialization parameters 70f28f9

feat: switch to PaddleOCR for better Korean text recognition f16cb1a

refactor: simplify to OCR-only output 39446f7

chore: switch to Korean-optimized TrOCR model 7138a91

fix: match input dtype to model dtype for TrOCR c74ab95

fix: remove torch_dtype to fix TrOCR dtype mismatch e4a5da6

chore: switch OCR model to TrOCR 6a2327c

sdfdsfads23333 e94e117

sdfdsfads23333 92bb45b

sdfdsfads2333 64c57fb

sdfdsfads233 52bda02

refactor: simplify to medication name extractor only 7ce0e26

sdfdsfads23 ea6e0a5

sdfdsfadsㄴ e5ef627

sdfdsfads bfe2214

fix: resolve 401 error by using public Qwen2-VL-7B model with ultra quality optimizations 24e39c0

sdf 736677f

feat: upgrade to 72B model with 8-bit quantization for maximum quality e5f80e3

fix: change to VL-7B for ZeroGPU compatibility 84e5d9f

feat: complete redesign with single VL-32B model and modern UI 98b216f

feat: add real-time web search for drug verification 3cec557

feat: ultra-quality improvement for production-ready app 8020bc3

feat: add professional drug information system ad98685

fix: change image model to SDXL (FLUX access restricted) 88cbfb6

feat: optimize performance and improve UX (Phase 1) 5e231af

feat: upgrade models and improve quality 8a13800

Use PaddleOCR predict API and normalize inputs

a97e706

Running

Handle new PaddleOCR output format

2879fbc

Specify PaddleOCR recognition model name

ea3177c

Add paddlepaddle dependency

310be32

Adjust PaddleOCR args for new API

ae810a8

Configure PaddleOCR to use Korean PP-OCRv5

26e52a1

Update requirements for PaddleOCR

d6bf509

Switch OCR engine to PaddleOCR

7a2cdb5

perf: switch to Gemma-2-2B for faster inference

63c2769

perf: replace VLM with EasyOCR for ultra-fast Korean OCR

dcb7540

fix: upgrade to Qwen2.5-VL-3B with 8bit quantization

7fabc42

perf: switch to faster Qwen2-VL-2B for OCR

e96841e

perf: preload models at startup for faster inference

ab48ca2

fix: strip whitespace from HF_TOKEN

115fae5

chore: upgrade to MedGemma-4B-IT for medical analysis

f89c606

chore: upgrade to MedGemma-4B-IT for medical analysis

7d833fa

feat: use Gemma-2-2B for medical analysis

d5aff0d

feat: add medication analysis with easy explanations

0e6a905

chore: upgrade to Qwen2.5-VL-7B-Instruct

8a7e01c

fix: use correct Qwen2_5_VL class for Qwen2.5-VL model

c5fa8ab

chore: switch to Qwen2.5-VL-3B-Instruct for OCR

664b899

feat: upgrade to Qwen2-VL-2B for state-of-the-art OCR

28a7334

fix: move EasyOCR initialization inside GPU decorator

0ffc67e

feat: switch from PaddleOCR to EasyOCR

33831f7

fix: update PaddleOCR initialization parameters

70f28f9

feat: switch to PaddleOCR for better Korean text recognition

f16cb1a

refactor: simplify to OCR-only output

39446f7

chore: switch to Korean-optimized TrOCR model

7138a91

fix: match input dtype to model dtype for TrOCR

c74ab95

fix: remove torch_dtype to fix TrOCR dtype mismatch

e4a5da6

chore: switch OCR model to TrOCR

6a2327c

sdfdsfads23333

e94e117

sdfdsfads23333

92bb45b

sdfdsfads2333

64c57fb

sdfdsfads233

52bda02

refactor: simplify to medication name extractor only

7ce0e26

sdfdsfads23

ea6e0a5

sdfdsfadsㄴ

e5ef627

sdfdsfads

bfe2214

fix: resolve 401 error by using public Qwen2-VL-7B model with ultra quality optimizations

24e39c0

sdf

736677f

feat: upgrade to 72B model with 8-bit quantization for maximum quality

e5f80e3

fix: change to VL-7B for ZeroGPU compatibility

84e5d9f

feat: complete redesign with single VL-32B model and modern UI

98b216f

feat: add real-time web search for drug verification

3cec557

feat: ultra-quality improvement for production-ready app

8020bc3

feat: add professional drug information system

ad98685

fix: change image model to SDXL (FLUX access restricted)

88cbfb6

feat: optimize performance and improve UX (Phase 1)

5e231af

feat: upgrade models and improve quality

8a13800