ggml-org/llama.cpp

Pull Requests Commits

Whitelist ggml_op_get_rows in metal instead of blacklisting

ORippler committed 6 hours ago

62a98751

Add dedicated encoder and decoder tests for FP8

ORippler committed 6 hours ago

599934cf

add explanatory comments and TODOs

ORippler committed 6 hours ago

42e74202

remove unneeded scratch buffer from FP8 cpu cpy/dup path

ORippler committed 6 hours ago

ce1dded8

Add supported for strided copies to/from FP8 in CPU

ORippler committed 6 hours ago

33a5ad6f

More FP8 conversion fixes

ORippler committed 6 hours ago

8f0d1eb8

Emit warnings when not fusing FP8 during conversion

ORippler committed 6 hours ago

a2f1f580

Avoid fusing FP8 for --fuse-gate-up-exps as we may have different ws

ORippler committed 6 hours ago

8d1e8ff0

Fix conversion of fused qkvz & FP8 for Qwen3-Next

ORippler committed 6 hours ago

6efa322a

Disallow QKV-fusion for FP8 temporarily

ORippler committed 6 hours ago

6efcdccc

Fix and test gguf-py behavior for FP8_E4M3

ORippler committed 6 hours ago

ac621c87

Allow tensor-strategy for FP8-only compressed-tensor ckpts

ORippler committed 6 hours ago

e3d8c171

Allow remaining FP8 tensors to be dequantized during conversion

ORippler committed 6 hours ago

17c77e1e

Fix conversion of https://huggingface.co/RedHatAI/Qwen3.8-27B-NVFP4

ORippler committed 6 hours ago

a068e50c

Materialize input_scales also for FP8 when converting

ORippler committed 6 hours ago

1234493c

Fix python type-checker

ORippler committed 6 hours ago

16e4b4f5

WIP OCP FP8 E4M3 support

ORippler committed 6 hours ago

334ba981

skill: Update `code-review` skill for UI development (#30509)

allozaur committed 7 hours ago

Verified 37d6d653

ggml: fix conv_1d result layout for batched input (#28738)

ServeurpersoCom committed 9 hours ago

Verified 6868a52c

common : add --spec-draft-n-ubatch (#30459)

ggerganov committed 9 hours ago

Verified 3fb47677

docs : clarify installation recommendation for Intel oneAPI Toolkit (#30407)

alosslessdev committed 10 hours ago

Verified 318a471a

ggml : add note about mul_mat_id + disable mul_mat_id_dup tests (#30501)

ggerganov committed 11 hours ago

Verified 8c2134a8

model: split layers evenly when no device reports free memory (#29677)

ServeurpersoCom committed 14 hours ago

Verified df18bee1

vendor : update cpp-httplib to 0.61.0 (#30494)

angt committed 14 hours ago

Verified 3eb10136

ggml : improve RPC tensor transfer logs (#30481)

ggerganov committed 14 hours ago

Verified af0579c7

cuda : use SIMT implicit GEMM conv2d on AMD without WMMA (#30325)

aisk committed 14 hours ago

Verified 62bacfa0

jinja : write floats in tojson as Python does (#30362)

ykhrustalev committed 14 hours ago

Verified 77390baa

opencl: use buf+offset instead of subbuffer for adreno moe expert sort (#30418)

lhez committed 14 hours ago

Verified c5a99762

opencl: fix q8_0 dp4a mul_mat_id and mark PREC_F32 mul_mat_id unsupported for adreno (#30365)

lhez committed 14 hours ago

Verified 21a19cd3

models: drop the t_embd expand of the embedding graphs (#30461)

ServeurpersoCom committed 14 hours ago

Verified 9cf967a0

Older