Closed HughPH closed 3 years ago
The Jupyter Notebook is written for TPU and takes a little fiddling to work for GPU. We keep meaning to make a second GPU colab, but haven’t gotten around to it.
One of our users wrote up a guide to using it on GPU here
I was able to work around this issue by changing line 20 to read
mesh_shape = [("all_processors", 1)]
I imagine - not being wholly familiar with python - that 1 could be replaced with len(params["gpu_ids"])
- though obviously this would only work when there is a gpu_ids parameter.
Thanks for the information Stella, I had seen that before and turned back to GPTNeo "vanilla" because DeepSpeed requires at least a GTX2080. I have run into some further issues, so maybe I need to set this aside until I've upgraded to a more powerful machine.
I’m glad you were able to work around it! I look forward to hearing about what you get up to with our models :)
I've been experimenting with OpenAI's GPT-3, but I've immediately hit limitations and items that - for my use case - will have a very high ticket price.
What I want to do is use a GPT with additional training on my own novels to assist me with discerning likely traits from characters, and likely extensions to situations and relationships.
I was able to work around this issue by changing line 20 to read
mesh_shape = [("all_processors", 1)]
I imagine - not being wholly familiar with python - that 1 could be replaced with
len(params["gpu_ids"])
- though obviously this would only work when there is a gpu_ids parameter.
Thanks for solution, worked for me. Just to note: it is in file 'models_fns.py'
Describe the bug model_fns.py line 112 makes a call to TensorFlow which results in the following error:
ValueError: Argument not a list with same length as devices arg=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255] devices=['device:GPU:0']
Obviously it is correct, the argument has a list much longer than the list of devices.
To Reproduce Steps to reproduce the behavior:
python main.py --predict --prompt testprompt.txt --gpu_ids device:GPU:0 --model ~/GPT3_2-7B/config.json
Expected behavior Not to get an error 🤷
Proposed solution I haven't the faintest clue. I imagine TensorFlow 2.5.0 has a breaking change in a call where the arguments are flipped, or an argument has been removed or added.
Screenshots If applicable, add screenshots to help explain your problem.
Environment (please complete the following information):
Additional context Full log: