20241118 docs (#2629)

* module docs

* varbuilder gguf docs

* add a link to gguf files

* small additonal mod doc titles

* safetensor docs

* more core docs

* more module docs in canlde_core

* 2 more link fixes

candle-transformers/src/generation/mod.rs

@@ -1,3 +1,8 @@
+//! Logit Processing and Sampling
+//!
+//! Functionality for modeling sampling strategies and logits processing in text generation
+//! with support for temperature-based sampling, top-k filtering, nucleus sampling (top-p),
+//! and combinations thereof.
 use candle::{DType, Error, Result, Tensor};
 use rand::{distributions::Distribution, SeedableRng};
 

candle-transformers/src/object_detection.rs

@@ -1,3 +1,9 @@
+//! Bounding Boxes and Intersection
+//!
+//! This module provides functionality for handling bounding boxes and their manipulation,
+//! particularly in the context of object detection. It includes tools for calculating
+//! intersection over union (IoU) and non-maximum suppression (NMS).
+
 /// A bounding box around an object.
 #[derive(Debug, Clone)]
 pub struct Bbox<D> {

candle-transformers/src/quantized_nn.rs

@@ -1,3 +1,9 @@
+//! Utilities for quanitized network layers
+//!
+//! This module contains various implementations of standard neural network layers, modules and
+//! utilities including embedding, linear layers, and various normalization techniques.
+//! Most implementations provide quantized weights support.
+
 use crate::models::with_tracing::QMatMul;
 use crate::quantized_var_builder::VarBuilder;
 use candle::quantized::QTensor;

candle-transformers/src/quantized_var_builder.rs

@@ -1,3 +1,9 @@
+//! Varbuilder for Loading gguf files
+//!
+//! VarBuilder is a utility to store quantized tensors from a [GGUF model file](https://huggingface.co/docs/hub/gguf).
+//! These tensors can be loaded from disk using `from_gguf` or from an in-memory
+//! buffer using `from_gguf_buffer`.
+
 use candle::quantized::QTensor;
 use candle::{Device, Result, Shape};
 use std::sync::Arc;

candle-transformers/src/utils.rs

@@ -1,3 +1,5 @@
+//! Apply penalty and repeat_kv
+
 use candle::{Result, Tensor};
 
 pub fn apply_repeat_penalty(logits: &Tensor, penalty: f32, context: &[u32]) -> Result<Tensor> {