Revert nightly index, add note: vLLM nightly x86_64 builds broken as of 2026-02-13
Browse files- glm-ocr.py +7 -2
glm-ocr.py
CHANGED
|
@@ -5,11 +5,14 @@
|
|
| 5 |
# "pyarrow>=17.0.0,<18.0.0",
|
| 6 |
# "huggingface-hub",
|
| 7 |
# "pillow",
|
| 8 |
-
# "vllm
|
| 9 |
# "toolz",
|
| 10 |
# "torch",
|
| 11 |
# ]
|
| 12 |
#
|
|
|
|
|
|
|
|
|
|
| 13 |
# [tool.uv]
|
| 14 |
# prerelease = "allow"
|
| 15 |
# override-dependencies = ["transformers>=5.1.0"]
|
|
@@ -22,8 +25,10 @@ GLM-OCR is a compact 0.9B parameter OCR model achieving 94.62% on OmniDocBench V
|
|
| 22 |
Uses CogViT visual encoder with GLM-0.5B language decoder and Multi-Token Prediction
|
| 23 |
(MTP) loss for fast, accurate document parsing.
|
| 24 |
|
| 25 |
-
NOTE: Requires vLLM
|
| 26 |
transformers>=5.1.0 (GLM-OCR support landed in stable release).
|
|
|
|
|
|
|
| 27 |
First run may take a few minutes to download and install dependencies.
|
| 28 |
|
| 29 |
Features:
|
|
|
|
| 5 |
# "pyarrow>=17.0.0,<18.0.0",
|
| 6 |
# "huggingface-hub",
|
| 7 |
# "pillow",
|
| 8 |
+
# "vllm",
|
| 9 |
# "toolz",
|
| 10 |
# "torch",
|
| 11 |
# ]
|
| 12 |
#
|
| 13 |
+
# [[tool.uv.index]]
|
| 14 |
+
# url = "https://wheels.vllm.ai/nightly"
|
| 15 |
+
#
|
| 16 |
# [tool.uv]
|
| 17 |
# prerelease = "allow"
|
| 18 |
# override-dependencies = ["transformers>=5.1.0"]
|
|
|
|
| 25 |
Uses CogViT visual encoder with GLM-0.5B language decoder and Multi-Token Prediction
|
| 26 |
(MTP) loss for fast, accurate document parsing.
|
| 27 |
|
| 28 |
+
NOTE: Requires vLLM nightly wheels (GLM-OCR added in v0.16.0, PR #33005) and
|
| 29 |
transformers>=5.1.0 (GLM-OCR support landed in stable release).
|
| 30 |
+
As of 2026-02-13, vLLM nightly only has ARM wheels — x86_64 builds broken.
|
| 31 |
+
Check https://wheels.vllm.ai/nightly for updated builds before running.
|
| 32 |
First run may take a few minutes to download and install dependencies.
|
| 33 |
|
| 34 |
Features:
|