This disables various destructive functions and prevents the generated code from interfering with the test (e.g. fork bomb, killing other pro

if test(generated_code) is not None it'll try to run the code. otherwise it'll just return an input and output pair.

This function sets the global logging format and level. It will be called when import verl

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

Set pad_token_id to eos_token_id if it is None. Args: tokenizer (transformers.PreTrainedTokenizer): The tokenizer to be set.

Sets up embedding layer in first stage and output layer in last stage. This function initalizes word embeddings in the final stage when we ar

Sort the placement groups by node ip, all bundles in a single placement group should be on the same node. FSDPCheckpointManager saves sharde

Test consistency between original forward pass and SP+rmpad forward passes. Args: trainer: The FSDPSFTTrainer instance to test

Unfold the first n dims as new batch dim

Union two dict. Will throw an error if there is an item not the same object with the same key. Args: dict1: dict2: Returns:

Compute Categorical entropy loss Args: logits: `(torch.Tensor)` shape: (bs, response_length, vocab_size) eos_mask: `(

Compute entropy when the logits are sharded in tp ranks Args: vocab_parallel_logits: (total_nnz, vocab_size // tp_size) Returns:

We will expand this function gradually based on the available datasets

We will expand this function gradually based on the available datasets

We will expand this function gradually based on the available datasets

We will expand this function gradually based on the available datasets

Note that we also return the raw_input_ids so that it can be combined with other chat template