Flatten a multi-level nested collection of dictionaries and lists into a flat dictionary. The function traverses nested dictionaries and

Flatten a multi-level nested collection of dictionaries and lists into a flat dictionary. The function traverses nested dictionaries and

Flatten a multi-level nested collection of dictionaries and lists into a flat dictionary. The function traverses nested dictionaries and

Loads a data segment. This function checks if a cache directory is available and if the data is already cached. If the data

Divides an iterable into chunks of specified size or based on a given function. Useful for batching Parameters: - it

Converts `dtype` from `str` to torch.dtype when possible. Does not use an instantiated HF AutoConfig

Set weight_decay=0.0 for parameters in model.no_weight_decay, for parameters with attribute _no_weight_decay==True, for bias parameters if bias_we

Whether the task has a training set

Including files / directory 1-level below the path

Generate table of results.

Set the logits for none top-k values to -inf. Done in-place.

Set the logits for none top-k values to -inf. Done in-place.

Set the logits for none top-p values to -inf. Done in-place.

Take a single document and the LM results and evaluates, returning a dict where keys are the names of submetrics and values are the values of

Sample from top-k logits. Arguments: logits: Tensor of shape (batch_size, vocab_size)

Sample from top-k logits. Arguments: logits: Tensor of shape (batch_size, vocab_size)

Algorithm 1 from [1] [1] Fast Inference from Transformers via Speculative Decoding Yaniv Leviathan, Matan Kalman, Yossi Matias https://arx

Algorithm 1 from [1] [1] Fast Inference from Transformers via Speculative Decoding Yaniv Leviathan, Matan Kalman, Yossi Matias https://arx

if return_last_state is True, returns (out, last_state) last_state has shape (batch, dim, dstate). Note that the gradient of the last state is

if return_last_state is True, returns (out, last_state) last_state has shape (batch, dim, dstate). Note that the gradient of the last state is

Serializes a given function or string. If 'keep_callable' is True, the original callable is returned. Otherwise, attempts to return t

Set or update the configuration for a given key.

Query TextSynth API for completion. Retry with back-off until they respond.

:return: Iterable[obj] A iterable of any object, that doc_to_text can handle

Compute e = activation(x @ weight.T + bias). This wrapper kicks the `kernel_fwd` Triton kernel :param x: input tensor :param weight:

Update currently maintained parameters. Call this every time the parameters are updated, such as the result of the `optimizer

Splits a string into ngram words

:param engine: str OpenAI API engine (e.g. gpt-3.5-turbo-instruct) :param truncate: bool Truncate input if t

For prenorm=True, this Block has a slightly different structure compared to a regular prenorm Transformer block. The standard

Build an array with length = number-of-epochs * number-of-documents. Each index is mapped to a corresponding document.

Build doc-idx, sample-idx, and shuffle-idx. doc-idx: is an array (ordered) of documents to be used in training. sample-idx: is the start docum

Build the range [0, size) and shuffle.

Initializes an HF or HF-compatible PreTrainedModel from scratch inside HFLM, using the kwargs passed into self.__init__(). A

Helper method during initialization. Create a tokenizer object corresponding to the correct tokenizer for value of `pretrain