anirudhacharya opened a new issue #13805: Not able to load float16 parameters with Block.load_parameters URL: https://github.com/apache/incubator-mxnet/issues/13805 ## Description Not able to load float16 parameters with Block.load_parameters. I get the error ``` dtype incompatible expected <type 'numpy.float32'> vs saved <type 'numpy.float16'> ``` ## Environment info (Required) ``` ----------Python Info---------- ('Version :', '2.7.12') ('Compiler :', 'GCC 5.4.0 20160609') ('Build :', ('default', 'Nov 12 2018 14:36:49')) ('Arch :', ('64bit', 'ELF')) ------------Pip Info----------- ('Version :', '10.0.1') ('Directory :', '/usr/local/lib/python2.7/dist-packages/pip') ----------MXNet Info----------- ('Version :', '1.5.0') ('Directory :', '/usr/local/lib/python2.7/dist-packages/mxnet-1.5.0-py2.7.egg/mxnet') Hashtag not found. Not installed from pre-built package. ----------System Info---------- ('Platform :', 'Linux-4.4.0-1074-aws-x86_64-with-Ubuntu-16.04-xenial') ('system :', 'Linux') ('node :', 'ip-172-31-85-46') ('release :', '4.4.0-1074-aws') ('version :', '#84-Ubuntu SMP Thu Dec 6 08:57:58 UTC 2018') ----------Hardware Info---------- ('machine :', 'x86_64') ('processor :', 'x86_64') Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 64 On-line CPU(s) list: 0-63 Thread(s) per core: 2 Core(s) per socket: 16 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 79 Model name: Intel(R) Xeon(R) CPU E5-2686 v4 @ 2.30GHz Stepping: 1 CPU MHz: 1200.312 CPU max MHz: 3000.0000 CPU min MHz: 1200.0000 BogoMIPS: 4600.12 Hypervisor vendor: Xen Virtualization type: full L1d cache: 32K L1i cache: 32K L2 cache: 256K L3 cache: 46080K NUMA node0 CPU(s): 0-15,32-47 NUMA node1 CPU(s): 16-31,48-63 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology nonstop_tsc aperfmperf pni pclmulqdq monitor est ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single kaiser fsgsbase bmi1 hle avx2 smep bmi2 erms invpcid rtm rdseed adx xsaveopt ida ----------Network Test---------- Setting timeout: 10 Timing for MXNet: https://github.com/apache/incubator-mxnet, DNS: 0.0026 sec, LOAD: 0.3206 sec. Timing for PYPI: https://pypi.python.org/pypi/pip, DNS: 0.0036 sec, LOAD: 0.0824 sec. Timing for FashionMNIST: https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/dataset/fashion-mnist/train-labels-idx1-ubyte.gz, DNS: 0.0089 sec, LOAD: 0.5292 sec. Timing for Conda: https://repo.continuum.io/pkgs/free/, DNS: 0.0037 sec, LOAD: 0.0313 sec. Timing for Gluon Tutorial(en): http://gluon.mxnet.io, DNS: 0.1404 sec, LOAD: 0.0234 sec. Timing for Gluon Tutorial(cn): https://zh.gluon.ai, DNS: 0.2387 sec, LOAD: 0.4812 sec. ``` ## Error Message: ``` Traceback (most recent call last): File "test_gluon.py", line 45, in <module> new_net.load_parameters("net.params", ctx=ctx) File "/usr/local/lib/python2.7/dist-packages/mxnet-1.5.0-py2.7.egg/mxnet/gluon/block.py", line 402, in load_parameters params[name]._load_init(loaded[name], ctx) File "/usr/local/lib/python2.7/dist-packages/mxnet-1.5.0-py2.7.egg/mxnet/gluon/parameter.py", line 243, in _load_init self.name, str(self.dtype), str(data.dtype)) AssertionError: Failed loading Parameter 'multiinputmnist1_conv0_weight' from saved params: dtype incompatible expected <type 'numpy.float32'> vs saved <type 'numpy.float16'> ``` ## Minimum reproducible example ```python import numpy as np import mxnet as mx from mxnet import gluon mx.random.seed(1) from collections import namedtuple ##################################### #Create Model in Gluon ctx = mx.gpu() class MultiInputMnist(gluon.nn.HybridBlock): def __init__(self, **kwargs): gluon.nn.HybridBlock.__init__(self, **kwargs) # attention here to pass kwargs to initialization of hybridblock with self.name_scope(): self.conv = gluon.nn.Conv2D(channels=20, kernel_size=3, dilation=(1,1), use_bias= True, activation='relu') self.pool = gluon.nn.MaxPool2D(pool_size=2, strides=2) def hybrid_forward(self, F, input_1, input_2): # You don't really need *args, **kwards in this case add = F.broadcast_add(input_1, input_2) c = self.conv(add) p = self.pool(c) return p net = MultiInputMnist() net.collect_params().initialize(ctx=ctx) net.cast('float16') net.hybridize() np_data1 = np.random.randn(64,1,28,28) np_data2 = np.random.randn(64,1,28,28) inp1 = mx.nd.cast(mx.nd.array(np_data1, ctx=ctx), dtype='float16') #mx.nd.array(np_data1) inp2 = mx.nd.cast(mx.nd.array(np_data1, ctx=ctx), dtype='float16') #mx.nd.array(np_data2) #mx.nd.cast(mx.nd.array(np_data), dtype='float16') net.forward(inp1, inp2) #export for reloading in Gluon net.save_parameters("net.params") #export for reloading in Module API net.export("mnist16") ##################################### #load model in Gluon and infer new_net = MultiInputMnist() new_net.load_parameters("net.params", ctx=ctx) print(new_net.forward(inp1, inp2)) ``` ## Steps to reproduce 1. Run the above code ## What have you tried to solve it? 1. save the parameters as float32. Load the f32 params into the gluon block, and then cast it to f16.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
