boostr 0.1.0

ML framework built on numr - attention, quantization, model architectures
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45

//! Module names for CUDA kernel PTX files.

pub const DECODE_ATTENTION_MODULE: &str = "decode_attention";
pub const KV_INSERT_MODULE: &str = "kv_insert";
pub const PAGED_DECODE_ATTENTION_MODULE: &str = "paged_decode_attention";
pub const FLASH_V2_MODULE: &str = "flash_v2";
pub const FLASH_V2_BWD_MODULE: &str = "flash_v2_bwd";
pub const PAGED_ATTENTION_MODULE: &str = "paged_attention";
pub const PAGED_ATTENTION_BWD_MODULE: &str = "paged_attention_bwd";
pub const FLASH_V3_MODULE: &str = "flash_v3";
pub const FLASH_V3_BWD_MODULE: &str = "flash_v3_bwd";
pub const KV_CACHE_UPDATE_MODULE: &str = "kv_cache_update";
pub const VARLEN_ATTENTION_MODULE: &str = "varlen_attention";
pub const VARLEN_ATTENTION_BWD_MODULE: &str = "varlen_attention_bwd";
pub const MQA_GQA_MODULE: &str = "mqa_gqa";
pub const MQA_GQA_BWD_MODULE: &str = "mqa_gqa_bwd";
pub const ALIBI_MODULE: &str = "alibi";
pub const ALIBI_BWD_MODULE: &str = "alibi_bwd";
pub const KV_CACHE_INT4_MODULE: &str = "kv_cache_int4";
pub const KV_CACHE_FP8_MODULE: &str = "kv_cache_fp8";
pub const KV_CACHE_FP8_BWD_MODULE: &str = "kv_cache_fp8_bwd";
pub const KV_CACHE_QUANT_MODULE: &str = "kv_cache_quant";
pub const RESHAPE_AND_CACHE_MODULE: &str = "reshape_and_cache";
pub const FUSED_ADAMW_MODULE: &str = "fused_adamw";
pub const FUSED_SGD_MODULE: &str = "fused_sgd";
pub const FUSED_ADAGRAD_MODULE: &str = "fused_adagrad";
pub const FUSED_LAMB_MODULE: &str = "fused_lamb";
pub const FUSED_MULTI_TENSOR_MODULE: &str = "fused_multi_tensor";
pub const ROPE_MODULE: &str = "rope";
pub const ROPE_INTERLEAVED_MODULE: &str = "rope_interleaved";
pub const ROPE_YARN_MODULE: &str = "rope_yarn";
pub const SDPA_MODULE: &str = "sdpa";
pub const FUSED_QKV_MODULE: &str = "fused_qkv";
pub const MOE_ROUTING_MODULE: &str = "moe_routing";
pub const MOE_PERMUTE_MODULE: &str = "moe_permute";
pub const MOE_GROUPED_GEMM_MODULE: &str = "moe_grouped_gemm";
pub const SSD_STATE_PASSING_MODULE: &str = "ssd_state_passing";
pub const FUSED_GRAD_UNSCALE_CLIP_MODULE: &str = "fused_grad_unscale_clip";
pub const SPECULATIVE_VERIFY_MODULE: &str = "speculative_verify";
pub const SAMPLING_PENALTIES_MODULE: &str = "sampling_penalties";
pub const SAMPLING_MODULE: &str = "sampling";
pub const LOGITS_TO_TOKEN_MODULE: &str = "logits_to_token";
pub const CALIBRATION_MODULE: &str = "calibration";
pub const GRAMMAR_DFA_MODULE: &str = "grammar_dfa";
pub const PREFIX_CACHE_LOOKUP_MODULE: &str = "prefix_cache_lookup";