MACOS 26.2 M5芯片，32G内存 安装失败

[GithubFeed]

MACOS 26.2 M5芯片，32G内存

环境检测：

    (deepseek-ocr-mlx) jat@JacoTdeMacBook-Pro DeepSeek-OCR-WebUI % ./verify_mac_env.sh          
    ======================================
    🔍 Verifying Mac Environment
    ======================================
    
    ✅ Conda environment: deepseek-ocr-mlx
    
    🐍 Python 3.11.14
    
    📦 Checking required packages...
    
      ✅ torch (2.9.1)
      ✅ torchvision (0.24.1)
      ✅ transformers (4.46.3)
      ✅ tokenizers (0.20.3)
      ✅ fastapi (0.104.1)
      ✅ uvicorn (0.24.0)
      ✅ fitz (1.26.7)
      ✅ PIL (12.1.0)
      ✅ einops (0.8.1)
      ✅ addict (2.4.0)
      ✅ easydict (N/A)
      ✅ matplotlib (3.10.8)
      ✅ numpy (1.26.4)
    
    🍎 Checking MPS (Metal Performance Shaders)...
      ✅ MPS is available
    
    ======================================
    ✅ Environment verification passed!
       You can now run: ./start.sh
    ======================================

1、运行自动模式

没有自动下载Model，卡在sing conda environment: deepseek-ocr-mlx，内存占用13G、15G，即使开TUN、全局、退出代理，也是一样的

日志如下：

      Last login: Thu Jan 15 10:55:57 on ttys002
      (base) jat@JacoTdeMacBook-Pro DeepSeek-OCR-WebUI % conda activate deepseek-ocr-mlx
      (deepseek-ocr-mlx) jat@JacoTdeMacBook-Pro DeepSeek-OCR-WebUI % ./start.sh                     
      ==========================================
      🚀 DeepSeek-OCR-WebUI Startup
      ==========================================
      🍎 Running on macOS
         Apple Silicon detected
         Selected backend: mps
      ==========================================
      ✅ Using conda environment: deepseek-ocr-mlx

2、加载本地模型运行： 还是卡在上一步

在Model Scope和Hf-mirror下载的模型都试过。

export LOCAL_MODEL_PATH=/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model
./start.sh

3、手动运行：

没有自动下载Model。

日志如下：

  (deepseek-ocr-mlx) jat@JacoTdeMacBook-Pro DeepSeek-OCR-WebUI % python web_service_unified.py
  
  ==================================================
  🚀 DeepSeek-OCR Unified Service
  ==================================================
  📍 URL: http://0.0.0.0:8001
  📚 Docs: http://0.0.0.0:8001/docs
  ==================================================
  
  INFO:     Started server process [19594]
  INFO:     Waiting for application startup.
  ==================================================
  🚀 DeepSeek-OCR Unified Service Starting...
  ==================================================
  📁 Using local model: /Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model
  ✅ Detected Apple Silicon with MPS support
  📦 Loading DeepSeek-OCR with MPS
  You are using a model of type deepseek_vl_v2 to instantiate a model of type DeepseekOCR. This is not supported for all configurations of models and can yield errors.
  Some weights of DeepseekOCRForCausalLM were not initialized from the model checkpoint at /Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model and are newly initialized: ['model.vision_model.embeddings.position_ids']
  You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
  ✅ Model loaded on mps
  ✅ Backend loaded: MPS
  ==================================================
  INFO:     Application startup complete.
  INFO:     Uvicorn running on http://0.0.0.0:8001 (Press CTRL+C to quit)
  INFO:     127.0.0.1:50883 - "GET / HTTP/1.1" 200 OK
  /Users/jat/miniforge3/envs/deepseek-ocr-mlx/lib/python3.11/site-packages/transformers/generation/configuration_utils.py:590: UserWarning: `do_sample` is set to `False`. However, `temperature` is set to `0.0` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `temperature`.
    warnings.warn(
  The `seen_tokens` attribute is deprecated and will be removed in v4.41. Use the `cache_position` model input instead.
  `get_max_cache()` is deprecated for all Cache classes. Use `get_max_cache_shape()` instead. Calling `get_max_cache()` will raise error from v4.48
  =====================
  BASE:  torch.Size([1, 256, 1280])
  PATCHES:  torch.Size([6, 100, 1280])
  =====================
  The attention layers in this model are transitioning from computing the RoPE embeddings internally through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed `position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be removed and `position_embeddings` will be mandatory.

4、手动运行：提前修改本地Model位置再python web_service_unified.py

Model 是在Model Scope下载的：

 modelscope download --model deepseek-ai/DeepSeek-OCR

内存溢出：

修改：web_service.py

    # 全局变量
    model = None
    tokenizer = None
    MODEL_PATH = '/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model'
    MODEL_SOURCE = None  # 记录实际使用的模型源

修改：web_service_unified.py

  # Check for local model path
  local_model_path = os.environ.get("/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model", "")
  model_path = local_model_path if local_model_path else "deepseek-ai/DeepSeek-OCR"

日志如下：
(deepseek-ocr-mlx) jat@JacoTdeMacBook-Pro DeepSeek-OCR-WebUI % python web_service_unified.py

==================================================
🚀 DeepSeek-OCR Unified Service
==================================================
📍 URL: http://0.0.0.0:8001
📚 Docs: http://0.0.0.0:8001/docs
==================================================

INFO:     Started server process [9432]
INFO:     Waiting for application startup.
==================================================
🚀 DeepSeek-OCR Unified Service Starting...
==================================================
✅ Detected Apple Silicon with MPS support
📦 Loading DeepSeek-OCR with MPS
You are using a model of type deepseek_vl_v2 to instantiate a model of type DeepseekOCR. This is not supported for all configurations of models and can yield errors.
Some weights of DeepseekOCRForCausalLM were not initialized from the model checkpoint at deepseek-ai/DeepSeek-OCR and are newly initialized: ['model.vision_model.embeddings.position_ids']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
✅ Model loaded on mps
✅ Backend loaded: MPS
==================================================
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8001 (Press CTRL+C to quit)
INFO:     127.0.0.1:50245 - "GET / HTTP/1.1" 200 OK
INFO:     127.0.0.1:50245 - "GET /favicon.ico HTTP/1.1" 404 Not Found
/Users/jat/miniforge3/envs/deepseek-ocr-mlx/lib/python3.11/site-packages/transformers/generation/configuration_utils.py:590: UserWarning: `do_sample` is set to `False`. However, `temperature` is set to `0.0` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `temperature`.
  warnings.warn(
The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:None for open-end generation.
The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
The `seen_tokens` attribute is deprecated and will be removed in v4.41. Use the `cache_position` model input instead.
`get_max_cache()` is deprecated for all Cache classes. Use `get_max_cache_shape()` instead. Calling `get_max_cache()` will raise error from v4.48
=====================
BASE:  torch.Size([1, 256, 1280])
PATCHES:  torch.Size([6, 100, 1280])
=====================
The attention layers in this model are transitioning from computing the RoPE embeddings internally through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed `position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be removed and `position_embeddings` will be mandatory.

5、多次尝试，翻译失败：未调整任何代码

      export LOCAL_MODEL_PATH=/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model
  python web_service_unified.py

日志如下：

    (deepseek-ocr-mlx) jat@JacoTdeMacBook-Pro DeepSeek-OCR-WebUI % export LOCAL_MODEL_PATH=/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model
    python web_service_unified.py
    
    ==================================================
    🚀 DeepSeek-OCR Unified Service
    ==================================================
    📍 URL: http://0.0.0.0:8001
    📚 Docs: http://0.0.0.0:8001/docs
    ==================================================
    
    INFO:     Started server process [67116]
    INFO:     Waiting for application startup.
    ==================================================
    🚀 DeepSeek-OCR Unified Service Starting...
    ==================================================
    📁 Using local model: /Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model
    ✅ Detected Apple Silicon with MPS support
    📦 Loading DeepSeek-OCR with MPS
    You are using a model of type deepseek_vl_v2 to instantiate a model of type DeepseekOCR. This is not supported for all configurations of models and can yield errors.
    Some weights of DeepseekOCRForCausalLM were not initialized from the model checkpoint at /Volumes/MacD/ai/Deepseek/DeepSeek-OCR-Model and are newly initialized: ['model.vision_model.embeddings.position_ids']
    You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
    ✅ Model loaded on mps
    ✅ Backend loaded: MPS
    ==================================================
    INFO:     Application startup complete.
    INFO:     Uvicorn running on http://0.0.0.0:8001 (Press CTRL+C to quit)
    INFO:     127.0.0.1:50888 - "GET / HTTP/1.1" 200 OK
    /Users/jat/miniforge3/envs/deepseek-ocr-mlx/lib/python3.11/site-packages/torch/amp/autocast_mode.py:270: UserWarning: User provided device_type of 'cuda', but CUDA is not available. Disabling
      warnings.warn(
    ❌ Inference failed: Torch not compiled with CUDA enabled
    ❌ Error:
    Traceback (most recent call last):
      File "/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-WebUI/web_service_unified.py", line 235, in ocr_endpoint
        text = backend.infer(prompt=prompt, image_path=tmp_file)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/Volumes/MacD/ai/Deepseek/DeepSeek-OCR-WebUI/backends/mps_backend.py", line 47, in infer
        result = self.model.infer(
                 ^^^^^^^^^^^^^^^^^
      File "/Users/jat/.cache/huggingface/modules/transformers_modules/DeepSeek-OCR-Model/modeling_deepseekocr.py", line 935, in infer
        input_ids.unsqueeze(0).cuda(),
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/Users/jat/miniforge3/envs/deepseek-ocr-mlx/lib/python3.11/site-packages/torch/cuda/__init__.py", line 403, in _lazy_init
        raise AssertionError("Torch not compiled with CUDA enabled")
    AssertionError: Torch not compiled with CUDA enabled
    
    INFO:     127.0.0.1:50889 - "POST /ocr HTTP/1.1" 500 Internal Server Error

[niccoreyes]

Related to issue #42

I had the same issue, I had to install pytorch for Apple Silicon:

from

~/Github/DeepSeek-OCR-WebUI main ❯ ./start.sh                                                          1m 17s  deepseek-ocr
==========================================
🚀 DeepSeek-OCR-WebUI Startup
==========================================
🍎 Running on macOS
   Apple Silicon detected
   Selected backend: mps
==========================================
⚠️  Conda environment 'deepseek-ocr-mlx' not found
   Running with system Python

==================================================
🚀 DeepSeek-OCR Unified Service
==================================================
📍 URL: http://0.0.0.0:8001
📚 Docs: http://0.0.0.0:8001/docs
==================================================

INFO:     Started server process [14683]
INFO:     Waiting for application startup.
==================================================
🚀 DeepSeek-OCR Unified Service Starting...
==================================================
🔧 Forced backend: MPS
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
ERROR:    Traceback (most recent call last):
  File "/Users/niccoreyes/miniconda3/envs/deepseek-ocr/lib/python3.11/site-packages/starlette/routing.py", line 677, in lifespan
    async with self.lifespan_context(app) as maybe_state:
  File "/Users/niccoreyes/miniconda3/envs/deepseek-ocr/lib/python3.11/contextlib.py", line 210, in __aenter__
    return await anext(self.gen)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/niccoreyes/Github/DeepSeek-OCR-WebUI/web_service_unified.py", line 209, in lifespan
    from backends.mps_backend import MPSBackend
  File "/Users/niccoreyes/Github/DeepSeek-OCR-WebUI/backends/mps_backend.py", line 4, in <module>
    import torch
ModuleNotFoundError: No module named 'torch'

ERROR:    Application startup failed. Exiting.

then

conda install pytorch torchvision torchaudio -c pytorch-nightly

---

After installing pytorch, I encountered:

~/Github/DeepSeek-OCR-WebUI main ❯ ./start.sh                                                                  deepseek-ocr
==========================================
🚀 DeepSeek-OCR-WebUI Startup
==========================================
🍎 Running on macOS
   Apple Silicon detected
   Selected backend: mps
==========================================
⚠️  Conda environment 'deepseek-ocr-mlx' not found
   Running with system Python

==================================================
🚀 DeepSeek-OCR Unified Service
==================================================
📍 URL: http://0.0.0.0:8001
📚 Docs: http://0.0.0.0:8001/docs
==================================================

INFO:     Started server process [15727]
INFO:     Waiting for application startup.
==================================================
🚀 DeepSeek-OCR Unified Service Starting...
==================================================
🔧 Forced backend: MPS
📦 Loading DeepSeek-OCR with MPS
processor_config.json: 100%|█████████████████████████████████████████████████████████████████| 460/460 [00:00<00:00, 317kB/s]
tokenizer_config.json: 166kB [00:00, 178MB/s]
tokenizer.json: 9.98MB [00:03, 3.27MB/s]
special_tokens_map.json: 100%|██████████████████████████████████████████████████████████████| 801/801 [00:00<00:00, 3.80MB/s]
config.json: 2.67kB [00:00, 7.47MB/s]
modeling_deepseekocr.py: 42.0kB [00:00, 397kB/s]
conversation.py: 9.25kB [00:00, 9.18MB/s]
deepencoder.py: 38.0kB [00:00, 27.4MB/s]
configuration_deepseek_v2.py: 10.6kB [00:00, 22.3MB/s]
modeling_deepseekv2.py: 82.2kB [00:00, 60.1MB/s]
You are using a model of type deepseek_vl_v2 to instantiate a model of type DeepseekOCR. This is not supported for all configurations of models and can yield errors.
❌ Model loading failed: Using `low_cpu_mem_usage=True` or a `device_map` requires Accelerate: `pip install 'accelerate>=0.26.0'`
ERROR:    Traceback (most recent call last):
  File "/Users/niccoreyes/miniconda3/envs/deepseek-ocr/lib/python3.11/site-packages/starlette/routing.py", line 677, in lifespan
    async with self.lifespan_context(app) as maybe_state:
  File "/Users/niccoreyes/miniconda3/envs/deepseek-ocr/lib/python3.11/contextlib.py", line 210, in __aenter__
    return await anext(self.gen)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/niccoreyes/Github/DeepSeek-OCR-WebUI/web_service_unified.py", line 211, in lifespan
    backend.load_model()
  File "/Users/niccoreyes/Github/DeepSeek-OCR-WebUI/backends/mps_backend.py", line 26, in load_model
    self.model = AutoModel.from_pretrained(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/niccoreyes/miniconda3/envs/deepseek-ocr/lib/python3.11/site-packages/transformers/models/auto/auto_factory.py", line 559, in from_pretrained
    return model_class.from_pretrained(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/niccoreyes/miniconda3/envs/deepseek-ocr/lib/python3.11/site-packages/transformers/modeling_utils.py", line 3577, in from_pretrained
    raise ImportError(
ImportError: Using `low_cpu_mem_usage=True` or a `device_map` requires Accelerate: `pip install 'accelerate>=0.26.0'`

ERROR:    Application startup failed. Exiting.

Which then I had to install

~/Github/DeepSeek-OCR-WebUI main ❯ pip install "accelerate>=0.26.0"                                       39s  deepseek-ocr

now I am currently at:

~/Github/DeepSeek-OCR-WebUI main ❯ ./start.sh                                                          3m 58s  deepseek-ocr
==========================================
🚀 DeepSeek-OCR-WebUI Startup
==========================================
🍎 Running on macOS
   Apple Silicon detected
   Selected backend: mps
==========================================
⚠️  Conda environment 'deepseek-ocr-mlx' not found
   Running with system Python

==================================================
🚀 DeepSeek-OCR Unified Service
==================================================
📍 URL: http://0.0.0.0:8001
📚 Docs: http://0.0.0.0:8001/docs
==================================================

INFO:     Started server process [17312]
INFO:     Waiting for application startup.
==================================================
🚀 DeepSeek-OCR Unified Service Starting...
==================================================
🔧 Forced backend: MPS
📦 Loading DeepSeek-OCR with MPS
You are using a model of type deepseek_vl_v2 to instantiate a model of type DeepseekOCR. This is not supported for all configurations of models and can yield errors.
Downloading shards:   0%|                                                                              | 0/1 [00:00<?, ?it/s]
model-00001-of-000001.safetensors:  97%|██████████████████████████████████████████████▎ | 6.44G/6.67G [23:17<00:10, 22.5MB/s]

download is still ongoing

I will report the results