davanstrien HF Staff commited on
Commit
d6f278f
·
verified ·
1 Parent(s): e35bb06

Sync from GitHub via hub-sync

Browse files
unlimited-ocr2-live.py CHANGED
@@ -84,11 +84,12 @@ import time
84
  from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
85
 
86
  # The model card's launch settings. The image is the SGLang release whose pins
87
- # match Baidu's wheel (torch 2.13.0, transformers 5.12.1); move both together.
 
88
  SERVING = {
89
  "model": "baidu/Unlimited-OCR-2",
90
  "served_name": "Unlimited-OCR-2",
91
- "wheel": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/main/wheel/"
92
  "sglang-0.5.20.dev10%2Bg8d62f5fb9c-py3-none-any.whl",
93
  "extra_pip": ["fastokens==0.3.2"],
94
  "env": {
 
84
  from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
85
 
86
  # The model card's launch settings. The image is the SGLang release whose pins
87
+ # match Baidu's wheel (torch 2.13.0, transformers 5.12.1); move both together. The wheel URL
88
+ # names a model-repo commit: Baidu replaces the wheel on main (dev10 was removed 2026-10-29).
89
  SERVING = {
90
  "model": "baidu/Unlimited-OCR-2",
91
  "served_name": "Unlimited-OCR-2",
92
+ "wheel": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/91da6546d8ed91e02525f59c012a1d243266d7dd/wheel/"
93
  "sglang-0.5.20.dev10%2Bg8d62f5fb9c-py3-none-any.whl",
94
  "extra_pip": ["fastokens==0.3.2"],
95
  "env": {
unlimited-ocr2-saturate.py CHANGED
@@ -106,11 +106,13 @@ import zlib
106
  # - crop_mode False (base, one global view per page): the card's multi-page
107
  # default (infer_multi). On a 32-page 1861 scan gundam (crop) mode read the
108
  # same text with 6.7x the image tokens (58k vs 8.7k).
 
 
109
  SERVING = {
110
  "model": "baidu/Unlimited-OCR-2",
111
  "served_name": "Unlimited-OCR-2",
112
  "image": "lmsysorg/sglang:v0.5.20-cu130",
113
- "wheel": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/main/wheel/"
114
  "sglang-0.5.20.dev10%2Bg8d62f5fb9c-py3-none-any.whl",
115
  "extra_pip": ["fastokens==0.3.2"],
116
  "env": {
 
106
  # - crop_mode False (base, one global view per page): the card's multi-page
107
  # default (infer_multi). On a 32-page 1861 scan gundam (crop) mode read the
108
  # same text with 6.7x the image tokens (58k vs 8.7k).
109
+ # - wheel: the URL names a model-repo commit, because Baidu replaces the wheel on
110
+ # main (dev10 was removed 2026-10-29).
111
  SERVING = {
112
  "model": "baidu/Unlimited-OCR-2",
113
  "served_name": "Unlimited-OCR-2",
114
  "image": "lmsysorg/sglang:v0.5.20-cu130",
115
+ "wheel": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/91da6546d8ed91e02525f59c012a1d243266d7dd/wheel/"
116
  "sglang-0.5.20.dev10%2Bg8d62f5fb9c-py3-none-any.whl",
117
  "extra_pip": ["fastokens==0.3.2"],
118
  "env": {
unlimited-ocr2.py CHANGED
@@ -62,11 +62,12 @@ import subprocess
62
  import sys
63
 
64
  # The model card's launch settings. The image is the SGLang release whose pins
65
- # match Baidu's wheel (torch 2.13.0, transformers 5.12.1); move both together.
 
66
  SERVING = {
67
  "model": "baidu/Unlimited-OCR-2",
68
  "served_name": "Unlimited-OCR-2",
69
- "wheel": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/main/wheel/"
70
  "sglang-0.5.20.dev10%2Bg8d62f5fb9c-py3-none-any.whl",
71
  "extra_pip": ["fastokens==0.3.2"],
72
  "env": {
 
62
  import sys
63
 
64
  # The model card's launch settings. The image is the SGLang release whose pins
65
+ # match Baidu's wheel (torch 2.13.0, transformers 5.12.1); move both together. The wheel URL
66
+ # names a model-repo commit: Baidu replaces the wheel on main (dev10 was removed 2026-10-29).
67
  SERVING = {
68
  "model": "baidu/Unlimited-OCR-2",
69
  "served_name": "Unlimited-OCR-2",
70
+ "wheel": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/91da6546d8ed91e02525f59c012a1d243266d7dd/wheel/"
71
  "sglang-0.5.20.dev10%2Bg8d62f5fb9c-py3-none-any.whl",
72
  "extra_pip": ["fastokens==0.3.2"],
73
  "env": {