diff options
Diffstat (limited to 'candle-transformers/src/models/starcoder2.rs')
| -rw-r--r-- | candle-transformers/src/models/starcoder2.rs | 17 |
1 files changed, 17 insertions, 0 deletions
diff --git a/candle-transformers/src/models/starcoder2.rs b/candle-transformers/src/models/starcoder2.rs index d108d062..833cb067 100644 --- a/candle-transformers/src/models/starcoder2.rs +++ b/candle-transformers/src/models/starcoder2.rs @@ -1,3 +1,20 @@ +//! StarCoder model implementation with quantization support. +//! +//! StarCoder is a large language model optimized for code generation. +//! This implementation provides quantization for reduced memory and compute. +//! +//! Key characteristics: +//! - Causal self-attention mechanism +//! - Multi-query attention (MQA) +//! - LayerNorm for normalization +//! - Absolute positional embeddings +//! - Support for 8-bit quantization +//! +//! References: +//! - [StarCoder Paper](https://arxiv.org/abs/2305.06161) +//! - [Model Card](https://huggingface.co/bigcode/starcoder) +//! + #![allow(unused)] use candle::{DType, Device, Module, Result, Tensor, D}; use candle_nn::{layer_norm, linear_b, LayerNorm, Linear, VarBuilder}; |