Yang Yu

icon indicating an email [email protected]

icon company Tencent icon location Beijing icon location I am the NLP infra leader for WeChat, was a core developer for Paddle. Accelerate and deploy NLP/ recommendation models.

Results 4 comments of Yang Yu

Parallel all reduce communication and backprop

@StellaAthena Also, the NIC is 100G RDMA/RoCE.

Fix bug `write_config` when self._config_path is None

@labteral Please take a review

normalizeNumbers panic

I actually face the same issue last year when using `jq` as a script language in an internal system. I just patch JQ, and manually copy `map` and `slice` instead....

deploy/runtime: use a background thread to run GC when interpreters aren't executing the forward pass

FYI, we actually call `gc.freeze()` after loading the inference model in our online system to reduce GC latency.