biondizzle 658b12cb3d CRITICAL FIX: Remove double Q normalization and fix RoPE sin slice

2026-05-19 17:27:33 +00:00

..

fused_moe

Fix activation global scale: don't double-invert input_global_scale_inv

2026-05-19 06:03:08 +00:00

layers

CRITICAL FIX: Remove double Q normalization and fix RoPE sin slice

2026-05-19 17:27:33 +00:00

deepseek_v4_attention.py

FIX: Use vLLM's decode_swa_indices for correct paged KV cache access during decode

2026-05-19 16:55:44 +00:00

deepseek_v4.py

Fix hc_head mapping: checkpoint uses hc_head.hc_fn, model params are flat hc_head_fn

2026-05-19 03:58:25 +00:00

kernel.py

Add cutedsl to MoEBackend type in kernel config

2026-05-18 22:38:41 +00:00

patch_compressor_cache.py

Patch compressor cache for Blackwell (no FlashMLA alignment) - fixes 91 missing layers

2026-05-19 09:52:23 +00:00

patch_debug_layers.py

Add debug patch to print layer name mismatch

2026-05-19 09:45:10 +00:00

patch_indexer_cache.py

Patch SWA and Indexer cache specs for Blackwell (no FlashMLA alignment)

2026-05-19 09:29:57 +00:00

patch_kv_cache_utils.py

Fix KV cache page size patch: separate groups for large SWA pages

2026-05-19 09:05:14 +00:00

patch_swa_cache.py

Patch SWA and Indexer cache specs for Blackwell (no FlashMLA alignment)

2026-05-19 09:29:57 +00:00

register_cutedsl_kernel.py

Fix Blackwell: skip FlashMLA assertion + force CuTeDSL kernel

2026-05-19 08:19:23 +00:00