nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
ik_llama_opt/src
History
Thireus ☠ 6d78a87c4c
perplexity: signal-driven hot-swap mode for persistent per-tensor PPL/KLD benchmarking (extends #1989) (#2131) 
* perplexity: add signal-driven hot-swap mode for persistent KLD/PPL benchmarking

llama-perplexity can now stay resident and be driven through control/status files (reload/compute/exit): it reloads only the tensors that changed on disk and recomputes PPL/KLD without ever reloading the full model. File-based signalling works on Windows, macOS and Linux. The reload returning-to-original path now refreshes tensor data from disk instead of
reattaching stale weights.

* Not Cygwin specific
 		2026-07-14 12:56:03 +03:00
..
graphs DSA: Better way to build the attention mask (#2119) 2026-07-13 13:28:05 +03:00
CMakeLists.txt model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-arch.cpp model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-arch.h model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-build-context.cpp model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-build-context.h model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-context.h DSA: Better way to build the attention mask (#2119) 2026-07-13 13:28:05 +03:00
llama-cparams.h Add --prefetch-experts to stream mmap'd MoE experts into page cache (#2101) 2026-07-11 10:43:42 +03:00
llama-delta-net.cpp MTP: faster recurrent state restore (#1791) 2026-05-13 11:00:24 +03:00
llama-delta-net.h MTP: faster recurrent state restore (#1791) 2026-05-13 11:00:24 +03:00
llama-dflash.cpp DFlash: support gpt-oss drafts with attention bias (#2061) 2026-06-30 09:02:18 +02:00
llama-dflash.h DFlash: use persistent FA-ready K/V cache (#1997) 2026-06-22 16:49:35 +02:00
llama-expert-io.h Add --defer-experts flag to defer expert mmap residency on Linux (#1634) 2026-04-16 08:54:44 +02:00
llama-grammar.cpp Bug fixes (#1940) 2026-06-10 07:45:49 +02:00
llama-grammar.h llama : add token matching support to llama-grammar (#1220) 2026-02-03 07:57:17 +02:00
llama-hparams.cpp model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-hparams.h model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-impl.h Split mode graph for dense Gemma4 assistant (#2022) 2026-06-24 18:29:32 +02:00
llama-load-tensors.cpp sm graph: reshape folded-head MLA tensors to 3D before the per-head split (#2130) 2026-07-14 12:17:24 +03:00
llama-mmap.cpp Add --prefetch-experts to stream mmap'd MoE experts into page cache (#2101) 2026-07-11 10:43:42 +03:00
llama-mmap.h Parallelize weight loading for weights targeted at anonymous host ram and also GPU (#2057) 2026-07-07 09:52:19 +02:00
llama-model-loader.cpp Fix: on Windows, fall back to read-buffer for sm graph loading (#2121) 2026-07-13 13:27:04 +03:00
llama-model-loader.h Split mode graph for dense Gemma4 assistant (#2022) 2026-06-24 18:29:32 +02:00
llama-model.cpp model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-model.h model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-quantize.cpp Fix per_layer_token_embedding (#2117) 2026-07-12 11:58:09 +03:00
llama-quantize.h Allow using -rtr and -muge together (#1444) 2026-03-16 18:26:26 +01:00
llama-reload-info.h on-demand tensor reload (#1989) 2026-06-22 16:36:34 +02:00
llama-reload.cpp perplexity: signal-driven hot-swap mode for persistent per-tensor PPL/KLD benchmarking (extends #1989) (#2131) 2026-07-14 12:56:03 +03:00
llama-sampling.cpp Bug fixes (#1940) 2026-06-10 07:45:49 +02:00
llama-sampling.h Add adaptive sampling clone and free functions to manage memory (#1851) 2026-05-21 08:11:17 +03:00
llama-spec-features-dflash.cpp DFlash: use persistent FA-ready K/V cache (#1997) 2026-06-22 16:49:35 +02:00
llama-spec-features-dflash.h clean redudance in dflash graph and small logics (#1994) 2026-06-19 09:04:54 +02:00
llama-spec-features.cpp remove duplicated code and unnecesary refactor 2026-06-14 16:02:02 -03:00
llama-spec-features.h move dflash for it own file 2026-06-02 10:22:13 -03:00
llama-vocab.cpp model: add openPangu-2.0-Flash (92B-A6B) with MLA-latent cache, DSA/SWA, mHC, and multi-head MTP (#2065) 2026-07-11 12:29:20 +03:00
llama-vocab.h Add Mellum2 architecture support (#1919) 2026-06-04 14:28:02 +02:00
llama.cpp Fix per_layer_token_embedding (#2117) 2026-07-12 11:58:09 +03:00
unicode-data.cpp Merge mainline llama.cpp (#3) 2024-07-27 07:55:01 +02:00
unicode-data.h Merge mainline llama.cpp (#3) 2024-07-27 07:55:01 +02:00
unicode-script-data.cpp Add Unicode allowlist (#1597) 2026-04-10 18:22:57 +02:00
unicode.cpp Gemma4 tokenizer fixes (#1603) 2026-04-09 15:33:28 +02:00
unicode.h Add Unicode allowlist (#1597) 2026-04-10 18:22:57 +02:00