TODO: optimize this. - We can use async ops - We can use only one allgather Args: tensors: size: group:

Build the memory buffer given weight_buffer_meta Args: weight_buffer_meta: contains mapping from name to a dictionary containing shape an

for cuda memory alignment, make sure alignment by 128-bits

r"""Works like shutil.copy() for file, and shutil.copytree for dir, and supports hdfs. Copy data and mode bits ("cp src dst"). Return the file's

Filter responses based on accuracy and truncation criteria. Args: reward_tensor: Tensor containing accuracy scor

Gets the number of unfinished requests.

Returns an array of tuples, containing the string `input`, sympy `output` using the provided sympy `func`, and `symbolically` boolean for cal

Return world size for the tensor model parallel group.

Return world size for the tensor model parallel group.

Apply ceil() then return the ceil-ed expression. expr: Expr - sympy expression as an argument to ceil()

Apply floor() then return the floored expression. expr: Expr - sympy expression as an argument to floor()

Returns True if there are unfinished requests.

NOTE: This method is a hack from the open-sourced version without asertion of world_size = tp * pp Initialize model parallel groups.

Set input tensor to be used instead of forward()'s input. When doing pipeline parallelism the input from the previous stage comes fro

Update the policy with an iterator of DataProto Args: data (DataProto): an iterator over the DataProto that returns by

TODO(zhangchi.usc1992): We may change the implementation later

Writes an iterable of dictionaries to jsonl

Get Stats to be Logged to Prometheus.

Make minibatch iterator for updating the actor See PPO paper for details. https://arxiv.org/abs/1707.06347

Normalize answer expressions.

Normalize answer expressions.