Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
22 commits
Select commit Hold shift + click to select a range
bf97a98
feat(kernels): custom-mask prefill/paged attention for EAGLE-3 drafter
Jul 10, 2026
57d6a93
fix(kernels): guard EAGLE-3 FlashInfer dispatches + validate head div…
Jul 10, 2026
6f4e725
fix(kernels): overflow-check offsets + validate backing capacity in E…
Jul 11, 2026
c7baf67
feat(eagle3): drafter config + safetensors loading
Jul 10, 2026
cb3d9e8
Merge branch 'openinfer-project:main' into feat/eagle3-drafter-load
scatyf3 Jul 12, 2026
ab5ed17
fix(eagle3): clear clippy -D warnings on the drafter loader
Jul 13, 2026
b79822b
fix(eagle3): correct aux capture layers to post-layer [1,17,32]; drop…
Jul 13, 2026
2c2071c
fix(eagle3): clear clippy on the drafter test build
Jul 13, 2026
2a36a56
fix(core): validate dtype and rank in typed host tensor loaders
Jul 17, 2026
1359bf5
fix(eagle3): validate drafter tensor shapes and budget the RoPE cache…
Jul 17, 2026
4abbf86
feat(eagle3): drafter forward pass (single-step + chain rollout)
Jul 10, 2026
56f7f6b
fix(eagle3): clear clippy on the drafter forward
Jul 17, 2026
fec62b3
refactor(eagle3): rename seed_feature -> aux_hidden_states
Jul 17, 2026
2f09be2
fix(eagle3): assert drafter state/scratch geometry in the forward path
Jul 17, 2026
90358e3
docs(eagle3): number the prefill_batched steps to match draft_step
Jul 17, 2026
317b62b
docs(eagle3): tidy the prefill_prompt feature-shift explanation
Jul 17, 2026
e97297c
refactor(eagle3): rename aux_hidden_states -> last_aux_hidden_states
Jul 18, 2026
3989cfd
refactor(eagle3): rename draft_chain seed_hidden -> fused_target_hidden
Jul 18, 2026
b564787
refactor(eagle3): drop "seed" vocab for field-standard feature/input-…
Jul 18, 2026
58af3fe
fix(eagle3): drop double-space doc linebreak to clear clippy -D warnings
Jul 18, 2026
0809c86
Merge branch 'main' into feat/eagle3-drafter-forward
Jul 18, 2026
19a9a2b
fix(eagle3): reserve the commit slot in draft chain capacity
scatyf3 Jul 26, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion openinfer-qwen3/src/eagle3.rs
Original file line number Diff line number Diff line change
Expand Up @@ -6,10 +6,15 @@ use anyhow::Result;
use crate::config::Eagle3Config;
use openinfer_core::tensor::{DeviceMatrix, DeviceVec};

mod forward;
mod loading;
mod reservation;

// Wired into the KV budget by the forward-pass PR; kept here as the skeleton lands.
// Consumed by the scheduler PR that drives the drafter; exported as the forward
// path lands.
#[allow(unused_imports)]
pub(crate) use forward::{Eagle3RequestState, Eagle3Scratch};
// Wired into the KV budget by the scheduler PR; kept here as the skeleton lands.
#[allow(unused_imports)]
pub(crate) use reservation::Eagle3MemoryReservation;

Expand Down
Loading
Loading