Giancarlo DelfinandClaude Fable 5 f80b66f548 [Model Runner V2][Spec Decode] Add KV cache support for multi-layer MTP (#50062) ...

Signed-off-by: Giancarlo Delfin <gdelfin@inferact.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

2026-08-13 16:22:29 -07:00

..

prefix_cache

[Bugfix] Disable fine-grained prefix-cache hits for incompatible hybrid KV layouts (#51843)

2026-08-12 04:08:40 -07:00

__init__.py

Implement Async Scheduling (#19970)

2025-07-14 23:01:46 -07:00

test_async_scheduler.py

[Feature] Mask Replay (#49577)

2026-08-13 08:44:31 +00:00

test_contiguous_kv_packing.py

[Core][DSV4] Compact MXFP4 indexer KV cache and packed group overlays (#48993)

2026-07-22 16:58:44 -07:00

test_deferred_block_free.py

[2/N][Core] support partial prefix cache hit for hybrid model (#46384)

2026-07-12 05:37:51 +00:00

test_encoder_cache_manager.py

[Perf] Improve multimodal item handling from O(n) to O(log n) per step (#44212)

2026-06-03 11:00:26 +00:00

test_kv_cache_metrics.py

[Core][Observability] Add KV cache residency metrics (#27793)

2025-12-01 18:27:53 +00:00

test_kv_cache_utils.py

[BugFix][KV Cache] Fix hybrid prefix caching with hidden-state extraction (#51108)

2026-08-06 06:42:37 +00:00

test_kv_sharing.py

Decouple page_size_bytes calculation in AttentionSpec for TPU/RPA Compatibility. (#31635)

2026-01-08 09:00:07 +00:00

test_mamba_align_chunk_split.py

[Bugfix] Keep mamba align prefill chunks block-aligned past last_cache_position (#51113)

2026-08-06 10:21:00 -07:00

test_output.py

[BugFix] Fix ValueError in NewRequestData repr methods (#29392)

2025-11-28 13:42:30 +08:00

test_prefix_caching.py

[New model] Kimi K3 (#50000)

2026-07-30 03:49:58 -07:00

test_priority_preemption_bug.py

fix: resolve silent request skipping in PRIORITY scheduling (#49206)

2026-08-06 22:21:21 +00:00

test_priority_scheduler_random.py

[Bugfix][CI] fix typos (#34934)

2026-03-05 17:05:46 +00:00

test_repetition_detection.py

[Core] Add optional flags to check for repetitive token patterns in engine output (#35451)

2026-03-03 12:23:25 +08:00

test_reset_prefix_cache_e2e.py

[CI/Build][AMD] Use float16 in test_reset_prefix_cache_e2e to avoid accuracy issues (#29997)

2025-12-05 08:42:25 +00:00

test_scheduler_e2e.py

[CI] Replace large models with tiny alternatives in tests (#24057)

2025-10-16 15:51:27 +01:00

test_scheduler.py

[Model Runner V2][Spec Decode] Add KV cache support for multi-layer MTP (#50062)

2026-08-13 16:22:29 -07:00

test_single_type_kv_cache_manager.py

[Bugfix] Generalize KV block zeroing to AttentionSpec (#51749)

2026-08-11 12:13:09 -04:00

test_swa_inflight_window_free.py

[Bugfix][V1] Free out-of-window blocks on the processed-token basis under async scheduling (#47728)

2026-07-08 13:34:19 +01:00

test_worker_slot_overflow.py

[BugFix][MRV2] Ensure all req slots are accounted for when scheduling (#46974)

2026-07-02 10:19:24 -07:00

utils.py

[Perf] Adaptive budget for spec scheduled input tokens, ~60% better Kimi K3 DSpark TTFT (#51725)

2026-08-10 20:52:21 -07:00