BAAI
/

3v324v23 commited on
Commit
b61e215
·
1 Parent(s): 9018197

update readme

Browse files
Files changed (1) hide show
  1. README.md +15 -1
README.md CHANGED
@@ -27,6 +27,20 @@ TODO
27
 
28
  **Tips: Our inference code still under updating, you could update it by assign "--include '\*.py'" in huggingface-cli to only update the inference code, avoid downloading the whole model.*
29
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
  ---
31
  ### 1. Inference w/o. Efficiency Optimization
32
  ```python
@@ -88,7 +102,7 @@ import argparse
88
 
89
  torch.cuda.reset_peak_memory_stats()
90
  # load model
91
- model_path = '/share/minghao/Models2/Video-XL-2'
92
  tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
93
  device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
94
  model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, device_map=device,quantization_config=None, attn_implementation="sdpa", torch_dtype=torch.float16, low_cpu_mem_usage=True) # sdpa
 
27
 
28
  **Tips: Our inference code still under updating, you could update it by assign "--include '\*.py'" in huggingface-cli to only update the inference code, avoid downloading the whole model.*
29
 
30
+ ---
31
+ ### 0. Installing Required Packages
32
+ ```bash
33
+ pip install transformers==4.43.0
34
+ pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu121
35
+ pip install decord
36
+ pip install einops
37
+ pip install opencv-python
38
+ pip install accelerate==0.30.0
39
+ pip install numpy==1.26.4
40
+ # optional
41
+ pip install flash-attn --no-build-isolation
42
+ ```
43
+
44
  ---
45
  ### 1. Inference w/o. Efficiency Optimization
46
  ```python
 
102
 
103
  torch.cuda.reset_peak_memory_stats()
104
  # load model
105
+ model_path = '/root/Models/Video-XL-2'
106
  tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
107
  device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
108
  model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, device_map=device,quantization_config=None, attn_implementation="sdpa", torch_dtype=torch.float16, low_cpu_mem_usage=True) # sdpa