This commit addresses an issue with the convert_hf_to_gguf script
which is currently failing with:
```console
AttributeError: module 'torch' has no attribute 'uint64'
```
This occurred because safetensors expects torch.uint64 to be available
in the public API, but PyTorch 2.2.x only provides limited support for
unsigned types beyond uint8 it seems. The torch.uint64 dtype exists but
is not exposed in the standard torch namespace
(see pytorch/pytorch#58734).
PyTorch 2.4.0 properly exposes torch.uint64 in the public API, resolving
the compatibility issue with safetensors. This also required torchvision
to updated to =0.19.0 for compatibility.
Chenguang Li [Wed, 6 Aug 2025 06:12:42 +0000 (14:12 +0800)]
CANN: add support for ACL Graph (#15065)
* feat(cann): add optional support for ACL Graph execution
This commit adds support for executing ggml computational graphs using
Huawei's ACL graph mode via the USE_CANN_GRAPH flag. The support can be
enabled at compile time using the CMake option:
-DUSE_CANN_GRAPH=ON
By default, ACL graph execution is **disabled**, and the fallback path
uses node-by-node execution.
Key additions:
- CMake option to toggle graph mode
- Graph capture and execution logic using
- Tensor property matching to determine whether graph update is required
- Safe fallback and logging if the environment variable LLAMA_SET_ROWS
is unset or invalid
This prepares the backend for performance improvements in repetitive graph
execution scenarios on Ascend devices.
Sam [Mon, 4 Aug 2025 18:29:25 +0000 (04:29 +1000)]
model: support GLM 4.5 family of models (#14939)
* model: Add GLM 4.5 (#14921)
Co-authored-by: Sigbjørn Skjæret <redacted>
* Merge in PR suggestions
Co-authored-by: Sigbjørn Skjæret <redacted>
* model: Add GLM 4.5 family of models (#14921)
1. Updated tensor_mapping.py with NextN tensor mappings
- Added proper tensor mappings for all NextN/MTP tensors in /Users/samm/git/llama.cpp/gguf-py/gguf/tensor_mapping.py
- Added mappings for: eh_proj, embed_tokens, enorm, hnorm, shared_head.head, shared_head.norm
2. Added num_nextn_predict_layers configuration
- Added LLM_KV_NUM_NEXTN_PREDICT_LAYERS constant to llama-arch.h and llama-arch.cpp
- Added num_nextn_predict_layers field to llama_hparams struct
- Updated GLM4_MOE parameter loading in llama-model.cpp to read this parameter
- Modified tensor loading logic to conditionally load NextN tensors based on num_nextn_predict_layers
- Added GGUF writer support in gguf_writer.py with add_num_nextn_predict_layers() method
- Updated conversion script to extract and write this parameter from HuggingFace config
3. Added FIM tokens for GLM4_MOE
- Added GLM-4.5's FIM tokens to llama-vocab.cpp:
- <|code_prefix|> for FIM_PRE
- <|code_suffix|> for FIM_SUF
- <|code_middle|> for FIM_MID
4. Removed manual NextN tensor handling
- Removed the special-case handling in convert_hf_to_gguf.py that manually mapped NextN tensors
- NextN tensors are now handled automatically through the proper tensor mapping system
* glm 4.5 update tensors names
* model: glm 4.5 apply suggestions from code review
Daniel Bevenius [Sat, 2 Aug 2025 14:14:57 +0000 (16:14 +0200)]
kv-cache : skip alignment of n_stream in kv-cache log msg [no ci] (#15040)
This commit removes the right alignment the `n_stream` value in the
log message in the `llama_kv_cache_unified` constructor.
The motivation for this change is to enhance the readability of log
message. Currently the output looks like this:
```console
llama_kv_cache_unified: size = 2048.00 MiB ( 4096 cells, 32 layers, 1/ 1 seqs), K (f16): 1024.00 MiB, V (f16): 1024.00 MiB
```
Notice that the `n_stream` value is right aligned, which makes it a
little harder to read.
With the change in this commit the output will look like
```console
llama_kv_cache_unified: size = 2048.00 MiB ( 4096 cells, 32 layers, 1/1 seqs), K (f16): 1024.00 MiB, V (f16): 1024.00 MiB
```
Jeff Bolz [Sat, 2 Aug 2025 09:21:37 +0000 (04:21 -0500)]
vulkan: coopmat2 mul_mat optimizations (#14934)
- Increase tile size for k-quants, to match non-k-quants
- Choose more carefully between large and medium tiles, considering how it
interacts with split_k
- Allow larger/non-power of two split_k, and make the splits a multiple of 256
- Use split_k==3 to when >1/2 and <=2/3 of the SMs would hae been used
Jeff Bolz [Sat, 2 Aug 2025 07:57:04 +0000 (02:57 -0500)]
vulkan: optimizations for direct convolution (#14933)
* vulkan: optimizations for direct convolution
- Empirically choose a better tile size. Reducing BS_K/BS_NPQ helps fill
the GPU. The new size should be amenable to using coopmat, too.
- Fix shmem bank conflicts. 16B padding should work with coopmat.
- Some explicit loop unrolling.
- Skip math/stores work for parts of the tile that are OOB.
- Apply fastdiv opt.
- Disable shuffles for NV.
* Three tiles sizes for CONV_2D, and a heuristic to choose
* reallow collectives for pre-Turing
* make SHMEM_PAD a spec constant
* fixes for intel perf - no shmem padding, placeholder shader core count
Daniel Bevenius [Wed, 30 Jul 2025 16:07:11 +0000 (18:07 +0200)]
server : add support for `embd_normalize` parameter (#14964)
This commit adds support for the `embd_normalize` parameter in the
server code.
The motivation for this is that currently if the server is started with
a pooling type that is not `none`, then Euclidean/L2 normalization will
be the normalization method used for embeddings. However, this is not
always the desired behavior, and users may want to use other
normalization (or none) and this commit allows that.
Example usage:
```console
curl --request POST \
--url http://localhost:8080/embedding \
--header "Content-Type: application/json" \
--data '{"input": "Hello world today", "embd_normalize": -1}
```
Kai Pastor [Wed, 30 Jul 2025 12:52:26 +0000 (14:52 +0200)]
vulkan : fix 32-bit builds (ggml/1313)
The pipeline member can be cast to VkPipeline.
This is a VkPipeline_T* on 64 bit but a uint64_t on 32 bit.
Cf. VK_DEFINE_NON_DISPATCHABLE_HANDLE documentation.
HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (#14930)
This is useful for testing for regressions on GCN with CDNA hardware.
With GGML_HIP_MMQ_MFMA=Off and GGML_CUDA_FORCE_MMQ=On we can conveniently test the GCN code path on CDNA. As CDNA is just GCN renamed with MFMA added and limited use ACC registers, this provides a good alternative for regression testing when GCN hardware is not available.
HIP: Ignore unsupported unroll transformation in fattn-vec (#14931)
llvm with the amdgcn target dose not support unrolling loops with conditional break statements, when those statements can not be resolved at compile time. Similar to other places in GGML lets simply ignore this warning.
Leonard Mosescu [Mon, 28 Jul 2025 16:04:27 +0000 (09:04 -0700)]
test-backend-ops : extend test case filtering (#14865)
* Extend test case filtering
1. Allow passing multiple (comma-separated?) ops to test-backend-ops. This can be convenient when working on a set of ops, when you'd want to test them together (but without having to run every single op). For example:
`test-backend-ops.exe test -o "ADD,RMS_NORM,ROPE,SILU,SOFT_MAX"`
2. Support full test-case variation string in addition to basic op names. This would make it easy to select a single variation, either for testing or for benchmarking. It can be particularly useful for profiling a particular variation (ex. a CUDA kernel), for example:
These two can be combined. As the current `-o`, this change doesn't try to detect/report an error if an filter doesn't name existing ops (ex. misspelled)
llama-bench : use local GPUs along with RPC servers (#14917)
Currently if RPC servers are specified with '--rpc' and there is a local
GPU available (e.g. CUDA), the benchmark will be performed only on the
RPC device(s) but the backend result column will say "CUDA,RPC" which is
incorrect. This patch is adding all local GPU devices and makes
llama-bench consistent with llama-cli.
SYCL: Add set_rows support for quantized types (#14883)
* SYCL: Add set_rows support for quantized types
This commit adds support for GGML_OP_SET_ROWS operation for various
quantized tensor types (Q8_0, Q5_1, Q5_0, Q4_1, Q4_0, IQ4_NL) and BF16
type in the SYCL backend.
The quantization/dequantization copy kernels were moved from cpy.cpp
to cpy.hpp to make them available for set_rows.cpp.
This addresses part of the TODOs mentioned in the code.
* Use get_global_linear_id() instead
ggml-ci
* Fix formatting
ggml-ci
* Use const for ne11 and size_t variables in set_rows_sycl_q
overview / pkg / ggml / sources / llama.cpp / log