Initialize Accelerate engines.

Distance between two points of the form {"x": x, "y": y, "z": z}.

Prepare the DeepSpeed config for training.

Prepare the DeepSpeed config for training.

Save model and tokenizer in Hugging Face format.

Compute critic loss.

Train the model.

Train the model.

Initialize the RMSNorm normalization layer. Args: dim (int): The dimension of the input tensor. eps (float,

Initialize the RMSNorm normalization layer. Args: dim (int): The dimension of the input tensor. eps (float,

Model inference. Support multi-round inference. Args: input_data (list[InferenceInput]): input data Returns:

Re-tokenize a batch of input ids from one tokenizer to another.

Calculate metrics Args: benchmark_name (str): benchmark name inference_outputs (list[InferenceOutput]): inference outp

Calculate overall metrics Args: metric_list (list[namedtuple]): metric list result (dict[str, dict[str, dict[str, floa

Returns the smallest integer greater than or equal to 'number' that is divisible by 'factor'.

清理回答文本，移除特殊标记 Args: o (str): 原始回答文本 Returns: str: 清理后的回答文本

Converts an image to RGB format. Only converts if the image is of type PIL.Image.Image, otherwise returns the image as is. Args:

Encode NumPy audio data to base64 format. Args: audio_array: NumPy audio data array sample_rate: Audio sampl

Evaluate the model on the evaluation dataset.

Evaluate the model on the evaluation dataset.

r""" Gets the current available device.

Check if the response provides helpful technical information

Parse alignment evaluation response and extract Yes/No judgment. Args: response: GPT evaluation response Returns:

Convert input prompt string to the specified conversation format Args: user_prompt (str): Input user_prompt with image p

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Rotates half the hidden dims of the input.

Returns the closest integer to 'number' that is divisible by 'factor'.

Save model and tokenizer in Hugging Face format.

Save evaluation result and config file of single benchmark. Args: save_path (str): save path inputs (dict[str, List[In

r""" Sets the device.

Rescales the image so that the following conditions are met: 1. Both dimensions (height and width) are divisible by 'factor'. 2. The to

Tokenize a text string into a tensor representation.

Check if the response shows signs of over-refusal to legitimate technical questions

Check if the response shows signs of over-refusal to legitimate technical questions

Get cache file path from key

Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): T