-
We've been thinking about dynamic dataset replication it for a while. While this is definitely future work, @camhunt had the following question:
> Has there been any thought/discussion in leveragin…
-
# Problem
When I use mpirun to launch nccl-tests performance test, I got a poor performance compared with runing directly.
I am very confused with this, and I don't know what cause this. Can you off…
-
node version: 2.4.0
in config toml
```
# Output level for logging, including package level options
log_level = "consensus:info,main:info,state:info,*:error"
log_format = "json"
# Path to file f…
-
Globals are convenient, but lead to problems in maintainability.
- Their use can make two formerly unrelated modules dependent on each other
- They are difficult to document (little context)
- They…
-
Hello. I am running nodeos 5.0.0, compiled with GCC 11 and LLVM 11. Randomly throughout the day the node will stop responding to certain HTTP requests (but not others), and it also stops responding to…
-
Somehow transaction with more than 800 inputs are not correctly broadcasted.
This is my floonet node sending a 1670/2/1 tx.
```
20200714 17:37:00.212 DEBUG grin_servers::grin::seed - monitor_peer…
-
Distributed hash tables are powerful protocols that enable content discovering and routing in P2P networks. In "vanilla" DHT implementations (e.g. Kademlia), peer interactions leak a lot of informatio…
-
**Summary or problem description**
> If we are under attack, a large number of transactions are sent from a certain node. Parallel verification of transactions in Blockchain will result in fewer oppo…
-
**Describe the bug**
When I was testing the dryrun branch, 10 keys per node. I copied the bls.pass file to the .hmy/blskeys directory together with the corresponding .key file.
However, after res…
-
* paddle版本：1.5.1
* models版本：release/1.5
* 训练脚本：models/PaddleCV/image_classification/dist_train
使用nccl2在单机8卡V100训练resnet50，fp32在90epoch精度正常 top1@76.3，增加enable_dgc=True参数进行训练，实验多次在第38个epoch左右loss出现NA…