CPU recorder that stores loss/gradient history on host. More...
#include <iteration_recorder.hpp>
Public Member Functions | |
| void | init (int capacity) |
Allocate buffers for up to capacity iterations. More... | |
| void | reset () |
| Reset recorded size without releasing memory. More... | |
| void | record (int idx, double loss, double grad_norm, double time_ms=0.0) |
| Record a loss/grad/time entry at iteration index. More... | |
| void | copy_to_host (std::vector< double > &loss_out, std::vector< double > &grad_norm_out) const |
| Copy recorded loss and gradient norm to output vectors. More... | |
| void | copy_to_host (std::vector< double > &loss_out, std::vector< double > &grad_norm_out, std::vector< double > &time_ms_out) const |
| Copy recorded loss, gradient norm, and time to output vectors. More... | |
| int | size () const |
| Current number of recorded entries. More... | |
Detailed Description
CPU recorder that stores loss/gradient history on host.
Member Function Documentation
◆ copy_to_host() [1/2]
|
inline |
Copy recorded loss and gradient norm to output vectors.
Here is the caller graph for this function:
◆ copy_to_host() [2/2]
|
inline |
Copy recorded loss, gradient norm, and time to output vectors.
◆ init()
|
inline |
Allocate buffers for up to capacity iterations.
Here is the caller graph for this function:
◆ record()
|
inline |
Record a loss/grad/time entry at iteration index.
- Parameters
-
idx Iteration index. loss Loss value. grad_norm Gradient norm value. time_ms Cumulative time in ms.
Here is the caller graph for this function:
◆ reset()
|
inline |
Reset recorded size without releasing memory.
Here is the caller graph for this function:
◆ size()
|
inline |
Current number of recorded entries.
The documentation for this class was generated from the following file:
