Summary: 61 instances, 49 unique

Text	Count
"""TODO: Load checkpoint model."""	1
# TODO unrolling by omg, eps, nu may improve performance but	1
# TODO: support defined intermediate buffer later	1
# check_correctness=True, # TODO: re-enable when check_correctness works again.	5
// TODO we would prefer to blanket impl From/TryFrom ArgValue/RetValue, but we	1
# check_correctness=True, # TODO: re-enable when check_correctness works again.	2
# TODO: unify prelude and main IRModules	1
# TODO: Currently VM compiler is likely to stack overflow for large models.	1
self.diag_ctx.render()  # Raise exception on the first error we hit. TODO remove	1
// TODO: Support multiple outputs.	1
# TODO: Add BSR transpose support	1
/* TODO: Thread extent unitest not generated.*/	1
# TODO expand to other supported devices, as listed in tvm.rpc.client (@leandron)	1
// FIXME (anijain2305) - The lowering can be further optimized. Instead of inserting requantize in	2
# TODO:	1
# TODO add more schedule opts (similar to the NHWC template)	1
# TODO: (codeislife99) : Fix different dynamic batch size inputs	1
# TODO (t-vi): should this also be part of the type promotion?	1
# TODO (yongwww): add new op to avoid the squeeze overhead	1
# TODO: add checking here, datatype casting may cause precision loss	1
# TODO (t-vi): handle non-int constants? (like numpy.intXX)	1
/* TODO: the warp size of OpenCL device is not always 1	1
* TODO: shutdown hook won't work on Windows	1
// TODO: consider `new_with_params` to avoid ever allocating	1
# TODO: we need to support more colorspace, such as rgb.	1
# TODO: ideally we could vectorize this	1
# FIXME - https://github.com/apache/tvm/issues/4122	1
# TODO (@leandron) This is a path to a physical file, but	2
// FIXME: Right now memory_type is a 2-bit field which means that	1
# TODO (t-vi): why is the cast to int needed? similarly elsewhere	1
# TODO use march for arm(https://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html)?	1
# FIXME - https://github.com/apache/tvm/issues/3598	1
# TODO should we add a max_factor attr to these splits?	1
// TODO: type check	3
# TODO lib.get_source call have inconsistent behavior for unsupported	1
# TODO: add checking here, datatype casting may cause precision loss	3
// TODO: figure out how to type the last argument runtime::TypedPackedFunc<String(ObjectRef)> annotate)	1
# TODO: Change the output shape calculation based on keep_dim option	1
# TODO: scalar inputs	1
// FIXME: for no-copy, need setup_op_execs to not capture tensor ptr	1
# TODO: add support for a prod_scan	1
# TODO --device needs to be extended and tested to support other targets,	1
// TODO (mehrdadh): Update this when zephyr version has updated to 2.6.	1
# TODO (jwfromm) Test with strided slice once type unifier for this case is fixed.	1
"""Bitserial dense implementation. TODO: Why are these separate	1
# TODO - Naive softmax int8 implementation leads to bad accuracy. Currently, we can	1
# TODO:	1
# FIXME: pad_value is ignored...	1
# TODO enforce 4-way padding in topi/nn/conv2d after #4644 merged	2