realizar 0.8.5

Pure Rust ML inference engine built from scratch - model serving for GGUF and safetensors
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

//! Batched forward pass variants
//!
//! Contains forward_batch, forward_batch_gpu, forward_batch_with_cache,
//! and supporting batch matmul/attention helpers.

// Imports used by transitively included files (batched.rs, batch_size.rs, acceleration.rs, attention.rs)
#[allow(unused_imports)]
use crate::error::{RealizarError, Result};
#[allow(unused_imports)]
use crate::gguf::ops;
#[allow(unused_imports)]
use crate::gguf::{
    DispatchMetrics, OwnedQKVWeights, OwnedQuantizedKVCache, OwnedQuantizedModel,
    OwnedQuantizedTensor, QuantizedGenerateConfig, TokenBuffer, GGUF_TYPE_Q4_K, GGUF_TYPE_Q5_K,
    GGUF_TYPE_Q6_K,
};

include!("batch_tiled_causal_owned.rs");