Commits

Commit history

1. Commits on Oct 8, 2026

   • llama: support MoE cache over multiple GPUs (#30112)

     am17an

     authored
     c811cb8View commit details

     Copy full SHA for c811cb8

     View code at this pointBrowse repository at this point
2. Commits on Oct 7, 2026

   • llama : add a GPU cache for MoE experts kept in host memory (#29887)

     Show description for d6cf9ac
     am17an

     authored
     d6cf9acView commit details

     Copy full SHA for d6cf9ac

     View code at this pointBrowse repository at this point
3. Commits on Oct 6, 2026

   • vocab : implement PLaMo-3 tokenizer pre-segmentation (#30045)

     Show description for abeada3
     

     tokinasin

     and

     CISC

     authored
     abeada3View commit details

     Copy full SHA for abeada3

     View code at this pointBrowse repository at this point
4. Commits on Oct 5, 2026

   • kv-cache: fix restoring mismatched KV cache rotation by saving exact rotation metadata (#28498)

     Show description for 2107910
     

     eapache

     and

     ggerganov

     authored
     2107910View commit details

     Copy full SHA for 2107910

     View code at this pointBrowse repository at this point
5. Commits on Sep 29, 2026

   • llama : fix init in several tools/examples (#29632)

     ggerganov

     authored
     6d78fb0View commit details

     Copy full SHA for 6d78fb0

     View code at this pointBrowse repository at this point
6. Commits on Sep 28, 2026

   • tests : fix ggml init (#29554)

     Show description for 81ef10e
     ggerganov

     authored
     81ef10eView commit details

     Copy full SHA for 81ef10e

     View code at this pointBrowse repository at this point
7. Commits on Sep 27, 2026

   • server : allow RANK pooling batch splitting for causal LLM rerankers (ie. Qwen3 and Qwen3-VL) (#28876)

     Show description for 4da6337
     timothywang21

     authored
     4da6337View commit details

     Copy full SHA for 4da6337

     View code at this pointBrowse repository at this point
8. Commits on Sep 24, 2026

   • llama: add llama_batch_ext (#24669)

     Show description for fc343a8
     ngxson

     authored
     fc343a8View commit details

     Copy full SHA for fc343a8

     View code at this pointBrowse repository at this point
9. Commits on Sep 19, 2026

   • test-llama-archs : generate dummy test vocab (#29084)

     Show description for efa28e9
     ggerganov

     authored
     efa28e9View commit details

     Copy full SHA for efa28e9

     View code at this pointBrowse repository at this point
10. Commits on Sep 17, 2026

    • gguf : align the data section relative to the GGUF start, not the file (#28993)

      Show description for 81aeaeb
      

      ykhrustalev

      and

      JohannesGaessler

      authored
      81aeaebView commit details

      Copy full SHA for 81aeaeb

      View code at this pointBrowse repository at this point
11. Commits on Sep 9, 2026

    • llama : use int32_t for llama_sampler_chain_n return type (#28631)

      Show description for 4850c77
      

      linsen458-spec

      and

      linsen

      authored
      4850c77View commit details

      Copy full SHA for 4850c77

      View code at this pointBrowse repository at this point
12. Commits on Aug 28, 2026

    • bench: add --tensor-read-lazy (#27881)

      Show description for 50f068f
      ngxson

      authored
      50f068fView commit details

      Copy full SHA for 50f068f

      View code at this pointBrowse repository at this point
13. Commits on Aug 27, 2026

    • quantize: cap working memory size to avoid loading big tensors onto RAM (#27795)

      ngxson

      authored
      732707dView commit details

      Copy full SHA for 732707d

      View code at this pointBrowse repository at this point

    • llama: model_loader: add TENSOR_READ_LAZY (#27794)

      Show description for fac889f
      ngxson

      authored
      fac889fView commit details

      Copy full SHA for fac889f

      View code at this pointBrowse repository at this point
14. Commits on Aug 26, 2026

    • llama: add token ID tracking to KV cell (#27762)

      Show description for 925e117
      ngxson

      authored
      925e117View commit details

      Copy full SHA for 925e117

      View code at this pointBrowse repository at this point
15. Commits on Aug 23, 2026

    • DeepseekV4: fix rollback with multi-seq (#26756)

      Show description for b0539c4
      

      am17an

      and

      ggerganov

      authored
      b0539c4View commit details

      Copy full SHA for b0539c4

      View code at this pointBrowse repository at this point
16. Commits on Aug 12, 2026

    • cmake : introduce semantic versioning (#26839)

      Show description for 680a9ae
      

      danbev

      and

      ggerganov

      authored
      680a9aeView commit details

      Copy full SHA for 680a9ae

      View code at this pointBrowse repository at this point

    • server : support slot save/restore with media inputs (#26640)

      Show description for 5d9e5ac
      CHIPMUNK-T0T

      authored
      5d9e5acView commit details

      Copy full SHA for 5d9e5ac

      View code at this pointBrowse repository at this point
17. Commits on Aug 11, 2026

    • llama: add default load-mode auto, which avoids mmap on iGPUs (#26081)

      Show description for 153d324
      

      4 peopleauthored
      153d324View commit details

      Copy full SHA for 153d324

      View code at this pointBrowse repository at this point
18. Commits on Aug 10, 2026

    • Address review comment of PR 25532 (#26852)

      gaugarg-nv

      authored
      030ebb5View commit details

      Copy full SHA for 030ebb5

      View code at this pointBrowse repository at this point

    • llama : support multi-output backend sampling (#25532)

      Show description for dd1ea52
      

      gaugarg-nv

      and

      ggerganov

      authored
      dd1ea52View commit details

      Copy full SHA for dd1ea52

      View code at this pointBrowse repository at this point
19. Commits on Aug 4, 2026

    • sampler : remove "full-context windows" from history-based samplers (#26524)

      Show description for a6aa6f5
      ggerganov

      authored
      a6aa6f5View commit details

      Copy full SHA for a6aa6f5

      View code at this pointBrowse repository at this point

    • llama : move n_vocab from llama_sampler_data to penalty_sampler (#26520)

      Show description for 935cad6
      ggerganov

      authored
      935cad6View commit details

      Copy full SHA for 935cad6

      View code at this pointBrowse repository at this point
20. Commits on Aug 3, 2026

    • CUDA: Add backend sampler for penalties sampler (#25262)

      Show description for 96278e3
      kmorennv

      authored
      96278e3View commit details

      Copy full SHA for 96278e3

      View code at this pointBrowse repository at this point
21. Commits on Jul 31, 2026

    • llama : load MTP tensors only if they are really used (#26296)

      Show description for 82dbc4f
      

      fairydreaming

      and

      sszymczy

      authored
      82dbc4fView commit details

      Copy full SHA for 82dbc4f

      View code at this pointBrowse repository at this point
22. Commits on Jul 29, 2026

    • llama: move suppress_tokens handling to common/sampling (#26276)

      Show description for afeebe1
      ngxson

      authored
      afeebe1View commit details

      Copy full SHA for afeebe1

      View code at this pointBrowse repository at this point
23. Commits on Jul 27, 2026

    • args: add `-lm mlock` where it mlocks but doesnt mmap (#26135)

      Show description for ad256de
      taronaeo

      authored
      ad256deView commit details

      Copy full SHA for ad256de

      View code at this pointBrowse repository at this point
24. Commits on Jul 23, 2026

    • args: refactor mlock/mmap/directio into load-mode (#20834)

      Show description for e6dd0e2
      taronaeo

      authored
      e6dd0e2View commit details

      Copy full SHA for e6dd0e2

      View code at this pointBrowse repository at this point
25. Commits on Jul 7, 2026

    • Add Q2_0 quantization: type definition and CPU backend (#24448)

      khosravipasha

      authored
      bec4772View commit details

      Copy full SHA for bec4772

      View code at this pointBrowse repository at this point
26. Commits on Jul 2, 2026

    • llama : add llama_model_ftype_name() (#25134)

      Show description for fdb1db8
      angt

      authored
      fdb1db8View commit details

      Copy full SHA for fdb1db8

      View code at this pointBrowse repository at this point
27. Commits on Jun 21, 2026

    • spec : Support Step3.5/3.7 flash mtp3 (#24340)

      Show description for d789527
      

      forforever73

      and

      am17an

      authored
      d789527View commit details

      Copy full SHA for d789527

      View code at this pointBrowse repository at this point
28. Commits on Jun 7, 2026

    • llama : add Gemma4 MTP (#23398)

      am17an

      authored
      04eb4c4View commit details

      Copy full SHA for 04eb4c4

      View code at this pointBrowse repository at this point
29. Commits on Jun 2, 2026

    • llama : deprecate `llama_set_warmup` (#24009)

      Show description for 4f3a4be
      

      ggerganov

      and

      danbev

      authored
      4f3a4beView commit details

      Copy full SHA for 4f3a4be

      View code at this pointBrowse repository at this point
30. Commits on Jun 1, 2026

    • llama: limit max outputs of `llama_context` (#23861)

      Show description for de6f727
      am17an

      authored
      de6f727View commit details

      Copy full SHA for de6f727

      View code at this pointBrowse repository at this point
31. Commits on May 25, 2026

    • llama : document that only one on-device state can be saved per sequence (#23520)

      TimNN

      authored
      314e729View commit details

      Copy full SHA for 314e729

      View code at this pointBrowse repository at this point