Auto-generated on 2026-07-23 10:53:30 UTC
Repo: https://github.com/ggml-org/llama.cpp
Common ancestor: 0badc06
Branches: 2
graph BT
COMMON["<a href='https://github.com/ggml-org/llama.cpp/commit/0badc06ab53a8eb96e01242b92ec1365c4465a2a'>0badc06</a><br/><span style='font-size:smaller;color:gray'>sync : ggml</span>"]
COMMON -->|master| C_a935fbf
C_a935fbf["<a href='https://github.com/ggml-org/llama.cpp/pull/25500'>#25500</a><br/><span style='font-size:smaller;color:gray'>server: remove loading.html (#25500)</span>"]
C_a935fbf --> C_9f623c6
C_9f623c6["<a href='https://github.com/ggml-org/llama.cpp/pull/24874'>#24874</a><br/><span style='font-size:smaller;color:gray'>ui: use server modalities in non-router mode (#24874)</span>"]
C_9f623c6 --> C_07d9378
C_07d9378["<a href='https://github.com/ggml-org/llama.cpp/pull/25492'>#25492</a><br/><span style='font-size:smaller;color:gray'>feat: pre-select models in the webui using alias (#25492)</span>"]
C_07d9378 --> C_3e706dd
C_3e706dd["<a href='https://github.com/ggml-org/llama.cpp/pull/24717'>#24717</a><br/><span style='font-size:smaller;color:gray'>mtmd: deepseek-ocr v1 multi-tile (#24717)</span>"]
C_3e706dd --> C_22b69b6
C_22b69b6["<a href='https://github.com/ggml-org/llama.cpp/pull/25527'>#25527</a><br/><span style='font-size:smaller;color:gray'>arg: prevent duplicate spec model downloads (#25527)</span>"]
C_22b69b6 --> C_67776ea
C_67776ea["<a href='https://github.com/ggml-org/llama.cpp/pull/25512'>#25512</a><br/><span style='font-size:smaller;color:gray'>hexagon: improve ARGSORT performance for small tensors (#25512)</span>"]
C_67776ea --> C_c749cb0
C_c749cb0["<a href='https://github.com/ggml-org/llama.cpp/pull/24710'>#24710</a><br/><span style='font-size:smaller;color:gray'>llama : make tensor-split regex patterns static (#24710)</span>"]
C_c749cb0 --> C_4f37f51
C_4f37f51["<a href='https://github.com/ggml-org/llama.cpp/pull/25538'>#25538</a><br/><span style='font-size:smaller;color:gray'>server: accept null sampling params (#25538)</span>"]
C_4f37f51 --> C_1d1d9a9
C_1d1d9a9["<a href='https://github.com/ggml-org/llama.cpp/pull/25537'>#25537</a><br/><span style='font-size:smaller;color:gray'>opencl: add int8 dp4 dense and MoE prefill optimization for Adreno GPUs (#25537)</span>"]
C_1d1d9a9 --> C_76f2798
C_76f2798["<a href='https://github.com/ggml-org/llama.cpp/pull/24877'>#24877</a><br/><span style='font-size:smaller;color:gray'>Vulkan: route large matmuls to medium tile on Adreno (#24877)</span>"]
C_76f2798 --> C_00f5442
C_00f5442["<a href='https://github.com/ggml-org/llama.cpp/pull/24231'>#24231</a><br/><span style='font-size:smaller;color:gray'>ggml : add GGML_OP_LIGHTNING_INDEXER that implements DeepSeek V3.2/V4 lightning indexer (#24231)</span>"]
C_00f5442 --> C_ea1f7bb
C_ea1f7bb["<a href='https://github.com/ggml-org/llama.cpp/pull/25541'>#25541</a><br/><span style='font-size:smaller;color:gray'>server: refactor server_stream (#25541)</span>"]
C_ea1f7bb --> C_c92e806
C_c92e806["<a href='https://github.com/ggml-org/llama.cpp/pull/25526'>#25526</a><br/><span style='font-size:smaller;color:gray'>server: allow stream for exec_shell_command (#25526)</span>"]
C_c92e806 --> C_13f2b28
C_13f2b28["<a href='https://github.com/ggml-org/llama.cpp/pull/25521'>#25521</a><br/><span style='font-size:smaller;color:gray'>DeepseekV4: clear cache only for seq rather than full (#25521)</span>"]
C_13f2b28 --> C_3cec3bc
C_3cec3bc["<a href='https://github.com/ggml-org/llama.cpp/pull/25157'>#25157</a><br/><span style='font-size:smaller;color:gray'>cuda: Don't crash when querying memory on device with no free memory. (#25157)</span>"]
C_3cec3bc --> C_d72bfa3
C_d72bfa3["<a href='https://github.com/ggml-org/llama.cpp/pull/24917'>#24917</a><br/><span style='font-size:smaller;color:gray'>gguf : reject empty metadata keys (#24917)</span>"]
C_d72bfa3 --> C_e3546c7
C_e3546c7["<a href='https://github.com/ggml-org/llama.cpp/pull/24926'>#24926</a><br/><span style='font-size:smaller;color:gray'>Fix conditional to display 'LLAMA_SPLIT_MODE_TENSOR not implemented for architecture' message (#24926)</span>"]
C_e3546c7 --> C_6b4dc21
C_6b4dc21["<a href='https://github.com/ggml-org/llama.cpp/pull/22536'>#22536</a><br/><span style='font-size:smaller;color:gray'>server : fix image blocks in tool_result being dropped during Anthropic OpenAI conversion (#22536)</span>"]
C_6b4dc21 --> C_0c4fa7a
C_0c4fa7a["<a href='https://github.com/ggml-org/llama.cpp/pull/25472'>#25472</a><br/><span style='font-size:smaller;color:gray'>server : evict checkpoints within min-step of each other (#25472)</span>"]
C_0c4fa7a --> C_4114ba1
C_4114ba1["<a href='https://github.com/ggml-org/llama.cpp/pull/25548'>#25548</a><br/><span style='font-size:smaller;color:gray'>mtmd: fix silent prompt truncation on embedded NUL (#25548)</span>"]
C_4114ba1 --> C_8014d2c
C_8014d2c["<a href='https://github.com/ggml-org/llama.cpp/pull/25590'>#25590</a><br/><span style='font-size:smaller;color:gray'>server: Don't consider models with --no-mmproj-auto as multimodal (#25590)</span>"]
C_8014d2c --> C_3455882
C_3455882["<a href='https://github.com/ggml-org/llama.cpp/pull/25576'>#25576</a><br/><span style='font-size:smaller;color:gray'>vendor : update cpp-httplib to 0.50.1 (#25576)</span>"]
C_3455882 --> C_99f3dc3
C_99f3dc3["<a href='https://github.com/ggml-org/llama.cpp/pull/23116'>#23116</a><br/><span style='font-size:smaller;color:gray'>server: honour per-request reasoning_budget_tokens in chat completions (#23116)</span>"]
C_99f3dc3 --> C_38fd5c9
C_38fd5c9["<a href='https://github.com/ggml-org/llama.cpp/pull/25535'>#25535</a><br/><span style='font-size:smaller;color:gray'>ui: Remove recommended MCP Servers + improve MCP Servers Settings UI/UX (#25535)</span>"]
C_38fd5c9 --> C_e474bba
C_e474bba["<a href='https://github.com/ggml-org/llama.cpp/pull/25064'>#25064</a><br/><span style='font-size:smaller;color:gray'>sycl: add Q2_K to DMMV reorder path (#25064)</span>"]
C_e474bba --> C_efb3036
C_efb3036["<a href='https://github.com/ggml-org/llama.cpp/pull/25217'>#25217</a><br/><span style='font-size:smaller;color:gray'>sycl: add fused top-k MoE (#25217)</span>"]
C_efb3036 --> C_91c631b
C_91c631b["<a href='https://github.com/ggml-org/llama.cpp/pull/24674'>#24674</a><br/><span style='font-size:smaller;color:gray'>chat : fix reasoning leak with force-opened bare <think> templates (#24674)</span>"]
C_91c631b --> C_ad8d821
C_ad8d821["<a href='https://github.com/ggml-org/llama.cpp/pull/24405'>#24405</a><br/><span style='font-size:smaller;color:gray'>gguf : add tensor shape accessor (#24405)</span>"]
C_ad8d821 --> C_f4253ef
C_f4253ef["<a href='https://github.com/ggml-org/llama.cpp/pull/25616'>#25616</a><br/><span style='font-size:smaller;color:gray'>tests: Harmonize header use (#25616)</span>"]
C_f4253ef --> C_4193ea6
C_4193ea6["<a href='https://github.com/ggml-org/llama.cpp/pull/25621'>#25621</a><br/><span style='font-size:smaller;color:gray'>readme : add link to maintainer PRs (#25621)</span>"]
C_4193ea6 --> C_259ae1d
C_259ae1d["<a href='https://github.com/ggml-org/llama.cpp/commit/259ae1d'>259ae1d</a><br/><span style='font-size:smaller;color:gray'>spec: add Minimax2 eagle3 support</span>"]
C_259ae1d --> C_e920c52
C_e920c52["<a href='https://github.com/ggml-org/llama.cpp/pull/25338'>#25338</a><br/><span style='font-size:smaller;color:gray'>vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)</span>"]
C_e920c52 --> C_6eddde0
C_6eddde0["<a href='https://github.com/ggml-org/llama.cpp/pull/24127'>#24127</a><br/><span style='font-size:smaller;color:gray'>CUDA: refactor MMQ kernel configuration (#24127)</span>"]
C_6eddde0 --> C_2969d6d
C_2969d6d["<a href='https://github.com/ggml-org/llama.cpp/pull/25395'>#25395</a><br/><span style='font-size:smaller;color:gray'>model: add Hy3 (hy_v3) support with MTP speculative decoding (#25395)</span>"]
C_2969d6d --> C_14d3ba4
C_14d3ba4["<a href='https://github.com/ggml-org/llama.cpp/pull/25419'>#25419</a><br/><span style='font-size:smaller;color:gray'>metal : add Q2_0 support (#25419)</span>"]
C_14d3ba4 --> C_c1063ac
C_c1063ac["<a href='https://github.com/ggml-org/llama.cpp/pull/25205'>#25205</a><br/><span style='font-size:smaller;color:gray'>sycl: set fattn_vec_nthreads to 256 for Battlemage (#25205)</span>"]
C_c1063ac --> C_ec0dbef
C_ec0dbef["<a href='https://github.com/ggml-org/llama.cpp/pull/25504'>#25504</a><br/><span style='font-size:smaller;color:gray'>arg: Flush log before exiting after usage() (#25504)</span>"]
C_ec0dbef --> C_cb489bc
C_cb489bc["<a href='https://github.com/ggml-org/llama.cpp/pull/25641'>#25641</a><br/><span style='font-size:smaller;color:gray'>convert_hf_to_gguf: support split MTP export for HY V3 (#25641)</span>"]
C_cb489bc --> C_c9330ed
C_c9330ed["<a href='https://github.com/ggml-org/llama.cpp/pull/25539'>#25539</a><br/><span style='font-size:smaller;color:gray'>ui: add reasoning effort control to mobile add sheet (#25539)</span>"]
C_c9330ed --> C_47c7869
C_47c7869["<a href='https://github.com/ggml-org/llama.cpp/pull/24414'>#24414</a><br/><span style='font-size:smaller;color:gray'>kleidiai : add SME2 f32 kernel (#24414)</span>"]
C_47c7869 --> C_47a3966
C_47a3966["<a href='https://github.com/ggml-org/llama.cpp/pull/23660'>#23660</a><br/><span style='font-size:smaller;color:gray'>ggml: uniformize im2col dst_type for all conv ops (#23660)</span>"]
C_47a3966 --> C_657e011
C_657e011["<a href='https://github.com/ggml-org/llama.cpp/pull/25619'>#25619</a><br/><span style='font-size:smaller;color:gray'>tests: export-graph-ops: exit gracefully when called w/o arguments (#25619)</span>"]
C_657e011 --> C_a7312ae
C_a7312ae["<a href='https://github.com/ggml-org/llama.cpp/pull/25650'>#25650</a><br/><span style='font-size:smaller;color:gray'>ggml : add a set of functions for checking contiguity of inner tensor dimensions (#25650)</span>"]
C_a7312ae --> C_8ff8c42
C_8ff8c42["<a href='https://github.com/ggml-org/llama.cpp/pull/25516'>#25516</a><br/><span style='font-size:smaller;color:gray'>tokenize : align usage by using common args (#25516)</span>"]
C_8ff8c42 --> C_7cbd610
C_7cbd610["<a href='https://github.com/ggml-org/llama.cpp/pull/25432'>#25432</a><br/><span style='font-size:smaller;color:gray'>vulkan/cpu: Support f16 as SET_ROWS src. (#25432)</span>"]
C_7cbd610 --> C_7f575c3
C_7f575c3["<a href='https://github.com/ggml-org/llama.cpp/pull/25588'>#25588</a><br/><span style='font-size:smaller;color:gray'>DeepseekV4: fix seq_rm (#25588)</span>"]
C_7f575c3 --> C_17a05e4
C_17a05e4["<a href='https://github.com/ggml-org/llama.cpp/pull/25631'>#25631</a><br/><span style='font-size:smaller;color:gray'>ui: fix mcp panel for toggle + timeout + proxy + ON/OFF state (#25631)</span>"]
C_17a05e4 --> C_00e79f6
C_00e79f6["<a href='https://github.com/ggml-org/llama.cpp/pull/25639'>#25639</a><br/><span style='font-size:smaller;color:gray'>opencl: fix a dp4a bug for devices where cl_khr_integer_dot_product is unavailable (#25639)</span>"]
C_00e79f6 --> C_dfba90d
C_dfba90d["<a href='https://github.com/ggml-org/llama.cpp/pull/25529'>#25529</a><br/><span style='font-size:smaller;color:gray'>webui: parse effective-parameter sizes (E2B, E4B) as params (#25529)</span>"]
C_dfba90d --> C_236ab57
C_236ab57["<a href='https://github.com/ggml-org/llama.cpp/pull/25634'>#25634</a><br/><span style='font-size:smaller;color:gray'>ui: Fix spacing in tool-call request (#25634)</span>"]
C_236ab57 --> C_6e52db5
C_6e52db5["<a href='https://github.com/ggml-org/llama.cpp/pull/25655'>#25655</a><br/><span style='font-size:smaller;color:gray'>server: add --cors-* options (#25655)</span>"]
C_6e52db5 --> C_bf2c86d
C_bf2c86d["<a href='https://github.com/ggml-org/llama.cpp/pull/25649'>#25649</a><br/><span style='font-size:smaller;color:gray'>server : refactor prompt cache state ownership (#25649)</span>"]
C_bf2c86d --> C_c718542
C_c718542["<a href='https://github.com/ggml-org/llama.cpp/pull/25677'>#25677</a><br/><span style='font-size:smaller;color:gray'>hexagon: fix hmx-queue signal enum-narrowing problem (#25677)</span>"]
C_c718542 --> C_a4ce259
C_a4ce259["<a href='https://github.com/ggml-org/llama.cpp/pull/25671'>#25671</a><br/><span style='font-size:smaller;color:gray'>opencl: avoid the vec path in GEMV for unaligned row stride (#25671)</span>"]
C_a4ce259 --> C_00fa7cb
C_00fa7cb["<a href='https://github.com/ggml-org/llama.cpp/pull/25640'>#25640</a><br/><span style='font-size:smaller;color:gray'>opencl: handle OOB write in noshuffle GEMV kernels (odd ne01) (#25640)</span>"]
C_00fa7cb --> C_12127de
C_12127de["<a href='https://github.com/ggml-org/llama.cpp/pull/25673'>#25673</a><br/><span style='font-size:smaller;color:gray'>opencl: do not use `clCreateBufferWithProperties` when targeting CL 2.x (#25673)</span>"]
C_12127de --> C_32b741c
C_32b741c["<a href='https://github.com/ggml-org/llama.cpp/pull/25222'>#25222</a><br/><span style='font-size:smaller;color:gray'>[SYCL] Flash Attention with XMX engine via oneDNN (#25222)</span>"]
C_32b741c --> C_0e148a5
C_0e148a5["<a href='https://github.com/ggml-org/llama.cpp/pull/25525'>#25525</a><br/><span style='font-size:smaller;color:gray'>sycl: Increase minimum buffer size for USM system allocations (#25525)</span>"]
C_0e148a5 --> C_22b208b
C_22b208b["<a href='https://github.com/ggml-org/llama.cpp/pull/25550'>#25550</a><br/><span style='font-size:smaller;color:gray'>sycl : implement xielu op (#25550)</span>"]
C_22b208b --> C_ae9291e
C_ae9291e["<a href='https://github.com/ggml-org/llama.cpp/pull/25653'>#25653</a><br/><span style='font-size:smaller;color:gray'>sycl : support kernel type fp16 for conv2d_dw (#25653)</span>"]
C_ae9291e --> C_d3fba0c
C_d3fba0c["<a href='https://github.com/ggml-org/llama.cpp/pull/25656'>#25656</a><br/><span style='font-size:smaller;color:gray'>sycl : fix get_rows Q2_K, Q4_K, Q5_K (#25656)</span>"]
C_d3fba0c --> C_33a75f4
C_33a75f4["<a href='https://github.com/ggml-org/llama.cpp/pull/25702'>#25702</a><br/><span style='font-size:smaller;color:gray'>DeepseekV4: reduce graph splits (#25702)</span>"]
C_33a75f4 --> C_f955e39
C_f955e39["<a href='https://github.com/ggml-org/llama.cpp/pull/23997'>#23997</a><br/><span style='font-size:smaller;color:gray'>ggml: add f16 out_prod support for CPU and out_prod op for Vulkan (#23997)</span>"]
C_f955e39 --> C_b3c9d1b
C_b3c9d1b["<a href='https://github.com/ggml-org/llama.cpp/pull/25459'>#25459</a><br/><span style='font-size:smaller;color:gray'>metal: fuse snake activation (mul, sin, sqr, mul, add) (#25459)</span>"]
C_b3c9d1b --> C_c810293
C_c810293["<a href='https://github.com/ggml-org/llama.cpp/pull/25706'>#25706</a><br/><span style='font-size:smaller;color:gray'>ci : add HF_TOKEN to self-hosted workflows (#25706)</span>"]
C_c810293 --> C_a3e5b96
C_a3e5b96["<a href='https://github.com/ggml-org/llama.cpp/pull/25678'>#25678</a><br/><span style='font-size:smaller;color:gray'>cuda : relax tensor contiguity requirements for quantized concat (#25678)</span>"]
C_a3e5b96 --> C_a05df0a
C_a05df0a["<a href='https://github.com/ggml-org/llama.cpp/pull/25637'>#25637</a><br/><span style='font-size:smaller;color:gray'>ui: fix thinking menu never appearing in single-model mode (#25637)</span>"]
C_a05df0a --> C_a582222
C_a582222["<a href='https://github.com/ggml-org/llama.cpp/pull/25705'>#25705</a><br/><span style='font-size:smaller;color:gray'>server: fix read_file append_loc space breaking edit_file match (#25705)</span>"]
C_a582222 --> C_956973c
C_956973c["<a href='https://github.com/ggml-org/llama.cpp/pull/25720'>#25720</a><br/><span style='font-size:smaller;color:gray'>Fix crash with draft-simple (#25720)</span>"]
C_956973c --> C_f6f12e4
C_f6f12e4["<a href='https://github.com/ggml-org/llama.cpp/pull/25613'>#25613</a><br/><span style='font-size:smaller;color:gray'>CUDA: tighter MMQ src1 buffer size for native fp4 (#25613)</span>"]
C_f6f12e4 --> C_c3d47e6
C_c3d47e6["<a href='https://github.com/ggml-org/llama.cpp/pull/25697'>#25697</a><br/><span style='font-size:smaller;color:gray'>opencl: fix two issues on flash attention for Adreno a7x (#25697)</span>"]
C_c3d47e6 --> C_aff6eb6
C_aff6eb6["<a href='https://github.com/ggml-org/llama.cpp/pull/25672'>#25672</a><br/><span style='font-size:smaller;color:gray'>tokenize : drop --stdin mutual-exclusion check (#25672)</span>"]
C_aff6eb6 --> C_3b53219
C_3b53219["<a href='https://github.com/ggml-org/llama.cpp/pull/25545'>#25545</a><br/><span style='font-size:smaller;color:gray'>cuda : CUDA GGML_OP_LIGHTNING_INDEXER implementation (generic vector kernel + wmma kernel) (#25545)</span>"]
C_3b53219 --> C_32beb24
C_32beb24["<a href='https://github.com/ggml-org/llama.cpp/pull/25450'>#25450</a><br/><span style='font-size:smaller;color:gray'>ui: Agentic Content UX improvements (#25450)</span>"]
C_32beb24 --> C_505b1ed
C_505b1ed["<a href='https://github.com/ggml-org/llama.cpp/pull/25698'>#25698</a><br/><span style='font-size:smaller;color:gray'>opencl: exclude some moe kernels on Adreno a7x (#25698)</span>"]
C_505b1ed --> C_602f828
C_602f828["<a href='https://github.com/ggml-org/llama.cpp/pull/25628'>#25628</a><br/><span style='font-size:smaller;color:gray'>cuda: extract Q1_0 elements via __byte_perm (#25628)</span>"]
C_602f828 --> C_3dafb58
C_3dafb58["<a href='https://github.com/ggml-org/llama.cpp/pull/25745'>#25745</a><br/><span style='font-size:smaller;color:gray'>opencl: disable FA and MoE weights repack to work around compiler issues for Adreno 850 GPU (#25745)</span>"]
C_3dafb58 --> C_56d6e9d
C_56d6e9d["<a href='https://github.com/ggml-org/llama.cpp/pull/25716'>#25716</a><br/><span style='font-size:smaller;color:gray'>quant : allow using manual tensor types with --pure (#25716)</span>"]
C_56d6e9d --> C_a320cbf
C_a320cbf["<a href='https://github.com/ggml-org/llama.cpp/pull/25728'>#25728</a><br/><span style='font-size:smaller;color:gray'>ci : add official website link to release notes (#25728)</span>"]
C_a320cbf --> C_5839ba3
C_5839ba3["<a href='https://github.com/ggml-org/llama.cpp/pull/25441'>#25441</a><br/><span style='font-size:smaller;color:gray'>CUDA: dedup MoE gate/up activation quantization (#25441)</span>"]
C_5839ba3 --> C_c7d8722
C_c7d8722["<a href='https://github.com/ggml-org/llama.cpp/pull/24233'>#24233</a><br/><span style='font-size:smaller;color:gray'>ggml-cuda : restore prop.integrated on HIP builds (#24233)</span>"]
C_c7d8722 --> C_8ee54c8
C_8ee54c8["<a href='https://github.com/ggml-org/llama.cpp/pull/25756'>#25756</a><br/><span style='font-size:smaller;color:gray'>server: Ignore empty / non-existing `Origin` headers (#25756)</span>"]
C_8ee54c8 --> C_3f08ef2
C_3f08ef2["<a href='https://github.com/ggml-org/llama.cpp/pull/25749'>#25749</a><br/><span style='font-size:smaller;color:gray'>Enable CUDA graphs on volta+turing (#25749)</span>"]
C_3f08ef2 --> C_79bba02
C_79bba02["<a href='https://github.com/ggml-org/llama.cpp/pull/25228'>#25228</a><br/><span style='font-size:smaller;color:gray'>CUDA: Support CUDA Virtual Devices (#25228)</span>"]
C_79bba02 --> C_a55a8c5
C_a55a8c5["<a href='https://github.com/ggml-org/llama.cpp/pull/25733'>#25733</a><br/><span style='font-size:smaller;color:gray'>convert : fix dflash target tokenizer mismatch during conversion (#25733)</span>"]
C_a55a8c5 --> C_a8dc0e3
C_a8dc0e3["<a href='https://github.com/ggml-org/llama.cpp/pull/25076'>#25076</a><br/><span style='font-size:smaller;color:gray'>server : allow text-only slot save/restore with mtmd (#25076)</span>"]
C_a8dc0e3 --> C_32e789f
C_32e789f["<a href='https://github.com/ggml-org/llama.cpp/pull/25758'>#25758</a><br/><span style='font-size:smaller;color:gray'>tests: actually exercise `test-recurrent-state-rollback` (#25758)</span>"]
C_32e789f --> C_86b719b
C_86b719b["<a href='https://github.com/ggml-org/llama.cpp/pull/25624'>#25624</a><br/><span style='font-size:smaller;color:gray'>vendor: update BoringSSL to 0.20260713.0 (#25624)</span>"]
C_86b719b --> C_2e1fd76
C_2e1fd76["<a href='https://github.com/ggml-org/llama.cpp/pull/25536'>#25536</a><br/><span style='font-size:smaller;color:gray'>TP: fix Phi3, Bert, Plamo2/3, ChatGLM (#25536)</span>"]
C_2e1fd76 --> C_3278e92
C_3278e92["<a href='https://github.com/ggml-org/llama.cpp/pull/25769'>#25769</a><br/><span style='font-size:smaller;color:gray'>conversion: accept BitNetForCausalLM architecture name (#25769)</span>"]
C_3278e92 --> C_b15ca93
C_b15ca93["<a href='https://github.com/ggml-org/llama.cpp/pull/25229'>#25229</a><br/><span style='font-size:smaller;color:gray'>vulkan: when using transfer queue for async copies, sync on event_wait to avoid race (#25229)</span>"]
C_b15ca93 --> C_f15bd60
C_f15bd60["<a href='https://github.com/ggml-org/llama.cpp/pull/25478'>#25478</a><br/><span style='font-size:smaller;color:gray'>kleidiai: Add SME vs SME2 distinction in kernel dispatch (#25478)</span>"]
C_f15bd60 --> C_b2dd28a
C_b2dd28a["<a href='https://github.com/ggml-org/llama.cpp/pull/25762'>#25762</a><br/><span style='font-size:smaller;color:gray'>hexagon: L2 cache handling rework (dirty bit tracking with lazy flushing) and more MUL_MAT updates (#25762)</span>"]
C_b2dd28a --> C_0dc74e3
C_0dc74e3["<a href='https://github.com/ggml-org/llama.cpp/pull/25585'>#25585</a><br/><span style='font-size:smaller;color:gray'>DeepseekV4: Add fused hyper-connection ops (#25585)</span>"]
C_0dc74e3 --> C_ac2557c
C_ac2557c["<a href='https://github.com/ggml-org/llama.cpp/pull/25786'>#25786</a><br/><span style='font-size:smaller;color:gray'>docs: added a note about using OpenCl with Adreno 810 (#25786)</span>"]
C_ac2557c --> C_e8f19cc
C_e8f19cc["<a href='https://github.com/ggml-org/llama.cpp/pull/25780'>#25780</a><br/><span style='font-size:smaller;color:gray'>opencl: loads quants as uint for q4_K and q5_K flat mv (optimization for Adreno A7x GPUs) (#25780)</span>"]
C_e8f19cc --> C_b85833e
C_b85833e["<a href='https://github.com/ggml-org/llama.cpp/pull/25115'>#25115</a><br/><span style='font-size:smaller;color:gray'>opencl: add ABS op (#25115)</span>"]
C_b85833e --> C_0bd0ec6
C_0bd0ec6["<a href='https://github.com/ggml-org/llama.cpp/pull/25690'>#25690</a><br/><span style='font-size:smaller;color:gray'>sycl: fix row calculation when K_QUANTS_PER_ITERATION is 1 (#25690)</span>"]
C_0bd0ec6 --> C_788e07d
C_788e07d["<a href='https://github.com/ggml-org/llama.cpp/pull/25430'>#25430</a><br/><span style='font-size:smaller;color:gray'>vulkan: Support Q2_0 (#25430)</span>"]
C_788e07d --> C_11fd0a6
C_11fd0a6["<a href='https://github.com/ggml-org/llama.cpp/pull/25710'>#25710</a><br/><span style='font-size:smaller;color:gray'>ggml-blas: default hadamard mul_mat to cpu routine (#25710)</span>"]
C_11fd0a6 --> C_635cdd5
C_635cdd5["<a href='https://github.com/ggml-org/llama.cpp/pull/25811'>#25811</a><br/><span style='font-size:smaller;color:gray'>common : auto-download dflash- and eagle3- HF sidecars (#25811)</span>"]
C_635cdd5 --> C_5d5306b
C_5d5306b["<a href='https://github.com/ggml-org/llama.cpp/pull/25822'>#25822</a><br/><span style='font-size:smaller;color:gray'>tests : initialize all tensors in test_dsv4_hc to avoid NaNs in sentinel tensors (#25822)</span>"]
C_5d5306b --> C_3727404
C_3727404["<a href='https://github.com/ggml-org/llama.cpp/commit/3727404'>3727404</a><br/><span style='font-size:smaller;color:gray'>ggml : bump version to 0.17.0 (ggml/1568)</span>"]
C_3727404 --> C_7d56da7
C_7d56da7["<a href='https://github.com/ggml-org/llama.cpp/commit/7d56da7'>7d56da7</a><br/><span style='font-size:smaller;color:gray'>sync : ggml</span>"]
C_7d56da7 --> C_86d86ed
C_86d86ed["<a href='https://github.com/ggml-org/llama.cpp/pull/25805'>#25805</a><br/><span style='font-size:smaller;color:gray'>opencl: transpose q4_K noshuffle scales for coalesced reads (#25805)</span>"]
C_86d86ed --> C_6bdd77f
C_6bdd77f["<a href='https://github.com/ggml-org/llama.cpp/pull/25810'>#25810</a><br/><span style='font-size:smaller;color:gray'>opencl: read/write MoE dp4a activation tiles to local memory as 128-bit (vectorized LD/ST perf opt) for Adreno GPUs (#25810)</span>"]
C_6bdd77f --> C_86a9c79
C_86a9c79["<a href='https://github.com/ggml-org/llama.cpp/pull/25797'>#25797</a><br/><span style='font-size:smaller;color:gray'>opencl: load and use `kernel_gemm_moe_q6_k_f32_ns` from bin kernel lib (#25797)</span>"]
C_86a9c79 --> C_4937ca8
C_4937ca8["<a href='https://github.com/ggml-org/llama.cpp/pull/25787'>#25787</a><br/><span style='font-size:smaller;color:gray'>llama-quant : exclude i32 ffn_gate_tid2eid routing table from quantization (#25787)</span>"]
C_4937ca8 --> C_571d0d5
C_571d0d5["<a href='https://github.com/ggml-org/llama.cpp/pull/25823'>#25823</a><br/><span style='font-size:smaller;color:gray'>model: rotate injected K/V cache for DFlash (#25823)</span>"]
C_571d0d5 --> C_178a6c4
C_178a6c4["<a href='https://github.com/ggml-org/llama.cpp/pull/25910'>#25910</a><br/><span style='font-size:smaller;color:gray'>opencl: Support broadcast for Adreno MUL_MAT and honor `view_offs` for Adreno Q8_0 MUL_MAT for llama-server multi-stream (#25910)</span>"]
C_178a6c4 --> C_1521a9a
C_1521a9a["<a href='https://github.com/ggml-org/llama.cpp/pull/25865'>#25865</a><br/><span style='font-size:smaller;color:gray'>ui: enable the agentic flow when only the JS sandbox is active (#25865)</span>"]
C_1521a9a --> C_43b5e63
C_43b5e63["<a href='https://github.com/ggml-org/llama.cpp/pull/25783'>#25783</a><br/><span style='font-size:smaller;color:gray'>UI: fix Settings/Display tool call content toggle (#25783)</span>"]
C_43b5e63 --> C_4ee6a9a
C_4ee6a9a["<a href='https://github.com/ggml-org/llama.cpp/pull/25869'>#25869</a><br/><span style='font-size:smaller;color:gray'>ui: fix collapsed user bubble with markdown rendering (#25869)</span>"]
C_4ee6a9a --> C_91d2fc3
C_91d2fc3["<a href='https://github.com/ggml-org/llama.cpp/pull/25325'>#25325</a><br/><span style='font-size:smaller;color:gray'>llama_dsv4: write only used rows in state (#25325)</span>"]
C_91d2fc3 --> C_2beefef
C_2beefef["<a href='https://github.com/ggml-org/llama.cpp/pull/25815'>#25815</a><br/><span style='font-size:smaller;color:gray'>ui: Sidebar Conversations Bulk Action + Improved Settings logic/UI (#25815)</span>"]
C_2beefef --> C_76f46ad
C_76f46ad["<a href='https://github.com/ggml-org/llama.cpp/pull/25934'>#25934</a><br/><span style='font-size:smaller;color:gray'>hexagon: add CLAMP op (#25934)</span>"]
C_76f46ad --> C_305ba51
C_305ba51["<a href='https://github.com/ggml-org/llama.cpp/pull/25929'>#25929</a><br/><span style='font-size:smaller;color:gray'>CUDA: vectorize same-type get_rows with int4 copy (#25929)</span>"]
C_305ba51 --> C_5735e10
C_5735e10["<a href='https://github.com/ggml-org/llama.cpp/pull/25795'>#25795</a><br/><span style='font-size:smaller;color:gray'>ggml-openvino: Add GGML_BACKEND_DL_IMPL invocation for OpenVINO backend (#25795)</span>"]
C_5735e10 --> C_f048010
C_f048010["<a href='https://github.com/ggml-org/llama.cpp/pull/23570'>#23570</a><br/><span style='font-size:smaller;color:gray'>vulkan: Refactor vk_queue to use per-instance mutexes and unique handles (#23570)</span>"]
C_f048010 --> C_40b740a
C_40b740a["<a href='https://github.com/ggml-org/llama.cpp/pull/25868'>#25868</a><br/><span style='font-size:smaller;color:gray'>server : properly handle null llama_context (#25868)</span>"]
C_40b740a --> C_fd41bf6
C_fd41bf6["<a href='https://github.com/ggml-org/llama.cpp/pull/25760'>#25760</a><br/><span style='font-size:smaller;color:gray'>server: return 400 instead of 500 on validation error with X-Conversation-Id (#25760)</span>"]
C_fd41bf6 --> C_60f6a17
C_60f6a17["<a href='https://github.com/ggml-org/llama.cpp/pull/25955'>#25955</a><br/><span style='font-size:smaller;color:gray'>common: resolve draft repo to its requested sidecar (#25955)</span>"]
C_60f6a17 --> C_fb0e6b6
C_fb0e6b6["<a href='https://github.com/ggml-org/llama.cpp/pull/25701'>#25701</a><br/><span style='font-size:smaller;color:gray'>kleidiai : warn once when a weight type has no KleidiAI kernel (#25701)</span>"]
C_fb0e6b6 --> C_846e991
C_846e991["<a href='https://github.com/ggml-org/llama.cpp/pull/25896'>#25896</a><br/><span style='font-size:smaller;color:gray'>cuda: add sqrt_softplus in topk-moe for dsv4 (#25896)</span>"]
C_846e991 --> C_71102a7
C_71102a7["<a href='https://github.com/ggml-org/llama.cpp/pull/25968'>#25968</a><br/><span style='font-size:smaller;color:gray'>hexagon: check tensor type when reusing descriptors (#25968)</span>"]
C_71102a7 --> C_b4aa7dd
C_b4aa7dd["<a href='https://github.com/ggml-org/llama.cpp/pull/25781'>#25781</a><br/><span style='font-size:smaller;color:gray'>mtmd : use align_corners for qwen3vl vision position embedding interpolation (#25781)</span>"]
C_b4aa7dd --> C_66e4bf7
C_66e4bf7["<a href='https://github.com/ggml-org/llama.cpp/pull/25514'>#25514</a><br/><span style='font-size:smaller;color:gray'>convert: fix handle HunyuanVL XD-RoPE config (#25514)</span>"]
C_66e4bf7 --> C_1f66c3c
C_1f66c3c["<a href='https://github.com/ggml-org/llama.cpp/pull/25165'>#25165</a><br/><span style='font-size:smaller;color:gray'>Add support for Laguna XS.2 & M.1 (#25165)</span>"]
C_1f66c3c --> C_67b9b0e
C_67b9b0e["<a href='https://github.com/ggml-org/llama.cpp/pull/25945'>#25945</a><br/><span style='font-size:smaller;color:gray'>llama-arch: fix DeepSeek4 APE tensor op (#25945)</span>"]
C_67b9b0e --> C_c5a4a0b
C_c5a4a0b["<a href='https://github.com/ggml-org/llama.cpp/pull/25962'>#25962</a><br/><span style='font-size:smaller;color:gray'>cuda: GET_ROWS quants (#25962)</span>"]
C_c5a4a0b --> C_7347430
C_7347430["<a href='https://github.com/ggml-org/llama.cpp/pull/25847'>#25847</a><br/><span style='font-size:smaller;color:gray'>webgpu : add CONV_2D_DW (depthwise conv2d) kernel (#25847)</span>"]
C_7347430 --> C_b4d6c7d
C_b4d6c7d["<a href='https://github.com/ggml-org/llama.cpp/pull/25987'>#25987</a><br/><span style='font-size:smaller;color:gray'>ci : fix SYCL package shared library lookup (#25987)</span>"]
C_b4d6c7d --> C_3ce7da2
C_3ce7da2["<a href='https://github.com/ggml-org/llama.cpp/pull/25983'>#25983</a><br/><span style='font-size:smaller;color:gray'>ggml: enable PowerPC backend variants on AIX (#25983)</span>"]
C_3ce7da2 --> C_f534da2
C_f534da2["<a href='https://github.com/ggml-org/llama.cpp/pull/25414'>#25414</a><br/><span style='font-size:smaller;color:gray'>Fix DeepSeek4 crafted template (#25414)</span>"]
C_f534da2 --> C_6d5a910
C_6d5a910["<a href='https://github.com/ggml-org/llama.cpp/pull/25989'>#25989</a><br/><span style='font-size:smaller;color:gray'>common: infer the speculative type from the draft repo sidecars (#25989)</span>"]
C_6d5a910 --> C_e8e6c7a
C_e8e6c7a["<a href='https://github.com/ggml-org/llama.cpp/pull/25999'>#25999</a><br/><span style='font-size:smaller;color:gray'>minor: fix reasoning preserve var for DS4 [no ci] (#25999)</span>"]
C_e8e6c7a --> C_61328e6
C_61328e6["<a href='https://github.com/ggml-org/llama.cpp/pull/25948'>#25948</a><br/><span style='font-size:smaller;color:gray'>feat(ui): add symbolic math support to JS sandbox via nerdamer (#25948)</span>"]
C_61328e6 --> C_e0833bf
C_e0833bf["<a href='https://github.com/ggml-org/llama.cpp/pull/25723'>#25723</a><br/><span style='font-size:smaller;color:gray'>mtmd: use RAII for setting and resetting non-causal attention (#25723)</span>"]
C_e0833bf --> C_0278d83
C_0278d83["<a href='https://github.com/ggml-org/llama.cpp/pull/25974'>#25974</a><br/><span style='font-size:smaller;color:gray'>hexagon: activation ops update (#25974)</span>"]
C_0278d83 --> C_1a064ab
C_1a064ab["<a href='https://github.com/ggml-org/llama.cpp/pull/25730'>#25730</a><br/><span style='font-size:smaller;color:gray'>CUDA: Improve NVFP4 W4A4 activation quantization (#25730)</span>"]
C_1a064ab --> C_cf51256
C_cf51256["<a href='https://github.com/ggml-org/llama.cpp/pull/25846'>#25846</a><br/><span style='font-size:smaller;color:gray'>ui: Add a 'Default' option for the reasoning selector (#25846)</span>"]
C_cf51256 --> C_4310aa4
C_4310aa4["<a href='https://github.com/ggml-org/llama.cpp/pull/26012'>#26012</a><br/><span style='font-size:smaller;color:gray'>contrib: allow all AI-generated code in general (#26012)</span>"]
C_4310aa4 --> C_d941f6e
C_d941f6e["<a href='https://github.com/ggml-org/llama.cpp/pull/25996'>#25996</a><br/><span style='font-size:smaller;color:gray'>conversion: fix non-MoE NomicBert GGUF conversion error (#25996)</span>"]
C_d941f6e --> C_c588c4f
C_c588c4f["<a href='https://github.com/ggml-org/llama.cpp/pull/25981'>#25981</a><br/><span style='font-size:smaller;color:gray'>metal : add f16 type support to leaky relu (#25981)</span>"]
C_c588c4f --> C_da296d6
C_da296d6["<a href='https://github.com/ggml-org/llama.cpp/pull/26030'>#26030</a><br/><span style='font-size:smaller;color:gray'>contrib: fix leftovers from the AI usage policy update (#26030)</span>"]
COMMON -->|dev-metal| C_3835214
C_3835214["<a href='https://github.com/ggml-org/llama.cpp/pull/24021'>#24021</a><br/><span style='font-size:smaller;color:gray'>metal : per-op source split + parallel compile (#24021)</span>"]
C_3835214 --> C_de75555
C_de75555["<a href='https://github.com/ggml-org/llama.cpp/pull/25176'>#25176</a><br/><span style='font-size:smaller;color:gray'>metal: add col2im_1d op (f32/f16/bf16) (#25176)</span>"]
C_de75555 --> C_e3ea5de
C_e3ea5de["<a href='https://github.com/ggml-org/llama.cpp/pull/25434'>#25434</a><br/><span style='font-size:smaller;color:gray'>metal : add set_rows with src0 f16 (#25434)</span>"]
C_e3ea5de --> C_65dd913
C_65dd913["<a href='https://github.com/ggml-org/llama.cpp/pull/21565'>#21565</a><br/><span style='font-size:smaller;color:gray'>metal : add CONV_2D_DW (depthwise convolution) support (#21565)</span>"]