72B模型量化需要多大内存，192G的内存都会被kill掉
sweetcard opened this issue · 9 comments
请问有大佬知道吗？
also want to know
估计要超过200G的内存才行。llama.cpp已经支持了，可以通过temp文件的方式，而不是用内存的方式进行量化了。
我们什么时候可以用上呢？
我们什么时候可以用上呢？
去下载自己量化就可以了
去下载自己量化就可以了
哦？这个是可以兼容的吗？我马上试试
去下载自己量化就可以了
哦？这个是可以兼容的吗？我马上试试
直接用llama.cpp量化就OK了
去下载自己量化就可以了
哦？这个是可以兼容的吗？我马上试试
直接用llama.cpp量化就OK了
请教：用llama.cpp量化时报错：
Traceback (most recent call last):
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 1228, in
main()
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 1161, in main
model_plus = load_some_model(args.model)
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 1078, in load_some_model
model_plus = merge_multifile_models(models_plus)
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 593, in merge_multifile_models
model = merge_sharded([mp.model for mp in models_plus])
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 572, in merge_sharded
return {name: convert(name) for name in names}
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 572, in
return {name: convert(name) for name in names}
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 547, in convert
lazy_tensors: list[LazyTensor] = [model[name] for model in models]
File "/Users/xxx/AIproject/llama.cpp/convert.py", line 547, in
lazy_tensors: list[LazyTensor] = [model[name] for model in models]
KeyError: 'transformer.h.0.attn.c_attn.bias'
pip list :
Package Version
Brotli 1.1.0
certifi 2023.11.17
charset-normalizer 3.3.2
contourpy 1.2.0
cycler 0.12.1
einops 0.7.0
filelock 3.13.1
fonttools 4.46.0
fsspec 2023.10.0
gguf 0.5.1
gmpy2 2.1.2
huggingface-hub 0.19.4
idna 3.6
Jinja2 3.1.2
kiwisolver 1.4.5
MarkupSafe 2.1.3
matplotlib 3.8.2
mpmath 1.3.0
networkx 3.2.1
numpy 1.24.4
packaging 23.2
Pillow 10.1.0
pip 23.3.1
pyparsing 3.1.1
PySocks 1.7.1
python-dateutil 2.8.2
PyYAML 6.0.1
qwen-cpp 0.1.2
regex 2023.10.3
requests 2.31.0
safetensors 0.4.1
sentencepiece 0.1.98
setuptools 68.2.2
six 1.16.0
sympy 1.12
tabulate 0.9.0
tiktoken 0.5.1
tokenizers 0.15.0
torch 2.2.0.dev20231130
torchaudio 2.2.0.dev20231130
torchvision 0.17.0.dev20231130
tqdm 4.66.1
transformers 4.35.2
transformers-stream-generator 0.0.4
typing_extensions 4.8.0
urllib3 2.1.0
wheel 0.42.0
是我装错了什么库吗？？？没搞懂？？？
去下载自己量化就可以了
哦？这个是可以兼容的吗？我马上试试
直接用llama.cpp量化就OK了
请教：用llama.cpp量化时报错：
Traceback (most recent call last): File "/Users/xxx/AIproject/llama.cpp/convert.py", line 1228, in main() File "/Users/xxx/AIproject/llama.cpp/convert.py", line 1161, in main model_plus = load_some_model(args.model) File "/Users/xxx/AIproject/llama.cpp/convert.py", line 1078, in load_some_model model_plus = merge_multifile_models(models_plus) File "/Users/xxx/AIproject/llama.cpp/convert.py", line 593, in merge_multifile_models model = merge_sharded([mp.model for mp in models_plus]) File "/Users/xxx/AIproject/llama.cpp/convert.py", line 572, in merge_sharded return {name: convert(name) for name in names} File "/Users/xxx/AIproject/llama.cpp/convert.py", line 572, in return {name: convert(name) for name in names} File "/Users/xxx/AIproject/llama.cpp/convert.py", line 547, in convert lazy_tensors: list[LazyTensor] = [model[name] for model in models] File "/Users/xxx/AIproject/llama.cpp/convert.py", line 547, in lazy_tensors: list[LazyTensor] = [model[name] for model in models] KeyError: 'transformer.h.0.attn.c_attn.bias'
pip list :
Package Version
Brotli 1.1.0 certifi 2023.11.17 charset-normalizer 3.3.2 contourpy 1.2.0 cycler 0.12.1 einops 0.7.0 filelock 3.13.1 fonttools 4.46.0 fsspec 2023.10.0 gguf 0.5.1 gmpy2 2.1.2 huggingface-hub 0.19.4 idna 3.6 Jinja2 3.1.2 kiwisolver 1.4.5 MarkupSafe 2.1.3 matplotlib 3.8.2 mpmath 1.3.0 networkx 3.2.1 numpy 1.24.4 packaging 23.2 Pillow 10.1.0 pip 23.3.1 pyparsing 3.1.1 PySocks 1.7.1 python-dateutil 2.8.2 PyYAML 6.0.1 qwen-cpp 0.1.2 regex 2023.10.3 requests 2.31.0 safetensors 0.4.1 sentencepiece 0.1.98 setuptools 68.2.2 six 1.16.0 sympy 1.12 tabulate 0.9.0 tiktoken 0.5.1 tokenizers 0.15.0 torch 2.2.0.dev20231130 torchaudio 2.2.0.dev20231130 torchvision 0.17.0.dev20231130 tqdm 4.66.1 transformers 4.35.2 transformers-stream-generator 0.0.4 typing_extensions 4.8.0 urllib3 2.1.0 wheel 0.42.0
是我装错了什么库吗？？？没搞懂？？？
要用这个脚本： convert-hf-to-gguf.py
哈哈哈，果然可行，感谢。
72b的推理速度真的慢