* Make a few GLM tensors not required
layer.nextn.shared_head_head and layer.nextn.embed_tokens are both excluded from GLM 4.6 resulting in the model not loading after conversion/quantization, this marks those tensors as not required which makes it work
* Update llama-model.cpp
layer.nextn.shared_head_norm also not required in case of future models
* minicpm: make GGUF scaling keys optional with legacy defaults
Older MiniCPM GGUFs do not include the scaling metadata keys (minicpm.embedding_scale, minicpm.residual_scale, minicpm.logit_scale). The loader currently treats these as required, so quantization fails with:
key not found in model: minicpm.embedding_scale
This change restores backward compatibility by treating these keys as optional in the loader and using the older MiniCPM scaling values:
embedding_scale = 12.0f
residual_scale = 1.4f / sqrt(n_layer)
logit_scale = 256.0f / n_embd
When the GGUF provides the keys, their values override the defaults; otherwise the legacy defaults are used. Newer GGUFs that already include these keys are unaffected.
Fixes: #16192
Signed-off-by: Vinkal Chudgar <vinkal.chudgar@gmail.com>
* Update src/llama-model.cpp
Committed as suggested. Thanks!
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Signed-off-by: Vinkal Chudgar <vinkal.chudgar@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* devops: move s390x and ppc64le ci build
we have access to ubuntu-24.04-s390x and ppc64le images now
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: disable ppc64le for now since they have compiler errors
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: stop warnings as errors
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: switch to non-macro flag
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: going the llama macro route
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add big-endian gguf test models
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: disable ppc64le to test s390x, check test build
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: dup .gguf.inp files for big-endian tests
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: dup .gguf.out files for big-endian too
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add python setup and endian byteswap
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: pooring thing does not have s390x python3
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add missing rust compiler for s390x
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: try rust actions runner
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Revert "devops: try rust actions runner"
This reverts commit 3f8db04356033d6c1d7eccc75ca396bc5298250c.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: try a different path for rust
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: dump home directory and user info
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: install gguf-py only
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: missed relative path
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: remove big-endian files since local swapping is working
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: revert test-tokenizer-0 cmakelists
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix unicode flags conversion from and to uint16_t
Bitfields are allocated in different order on s390x
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Simplify byteswap command
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Add byteswapping and git-lfs for test-tokenizers-ggml-vocabs
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix endianness detection in vocab loader
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Disable test-thread-safety on s390x
In this test a model is downloaded,
then immediately loaded to check if more downloads are needed,
and then used for test.
There is no clean way to separate all those steps
to add byteswapping between them, so just skip this test.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix q8_0 test in test-quantize-fns
vec_signed uses unexpected rounding mode.
Explicitly use different rounding function.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add big-endian stories260K
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add s390x test-eval-callback
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: fix test does not exist
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: fix model not found llama-eval-callback
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix q3_K dot product error in test-quantize-fns on s390x
Array q8bytes had only 4 elements allocated, but 8 elements accessed.
This lead to write out of bounds and later read of overwritten values out of bounds
and incorrect result.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: re-enable ppc64le for testing
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: activate test-thread-safety for s390x
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: disable ppc64le tests
for some reason it keeps failing test-thread-safety tests and I do not
have a machine that is able to replicate the tests.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: LLAMA_FATAL_WARNINGS=ON
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Correct repository URL for s390x for test-thread-safety model
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix fs_get_cache_directory
Ensure it works even if both XDG_CACHE_HOME and HOME are unset.
This might happen in containers.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Re-enable CI for ppc64le
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fortify ggml_rope_impl
Only memcpy data from sections argument if it's non-NULL.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Add TODO in struct unicode_cpt_flags to reimplement it in endian-independent way
* Update URL for big-endian model
* Update .github/workflows/build.yml
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update remaining mentions of BE models to ggml-org/models repo
---------
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
Co-authored-by: Aleksei Nikiforov <aleksei.nikiforov@linux.ibm.com>
Co-authored-by: Aleksei Nikiforov <103434461+AlekseiNikiforovIBM@users.noreply.github.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* CUDA: add a fused top-K MoE kernel
This kernel does the following:
1. softmax over the logits per token [n_experts, n_tokens]
2. argmax reduce over the top-k (n_experts_used) logits
3. write weights + ids to global memory
It is intended as fusion of softmax->top-k->get_rows pipeline for MoE models
* Refactor into ggml_cuda_should_use_topk_moe
* Review: Use better coalescing pattern, use WARP_SIZE, store logits into registers before
* Review: format + micro-optimizations
* Fix bug: fix tie breakers
* Add optional norm + clean-up code
* Use smem for final write
* Add bounds check
* Use better memory pattern for writeback
* add grok-2 support
* type fix
* type fix
* type fix
* "fix" vocab for invalid sequences
* fix expert tensor mapping and spaces in vocab
* add chat template
* fix norm tensor mapping
* rename layer_out_norm to ffn_post_norm
* ensure ffn_post_norm is mapped
* fix experts merging
* remove erroneous FFN_GATE entry
* concatenate split tensors and add more metadata
* process all expert layers and try cat instead of hstack
* add support for community BPE vocab
* fix expert feed forward length and ffn_down concat
* commit this too
* add ffn_up/gate/down, unsure if sequence is right
* add ffn_gate/down/up to tensor names
* correct residual moe (still not working)
* mess--
* fix embedding scale being applied twice
* add built in chat template
* change beta fast for grok if default value
* remove spm vocab in favor of community bpe vocab
* change attention temp length metadata type to integer
* update attention temp length metadata
* remove comment
* replace M_SQRT2 with std::sqrt(2)
* add yarn metadata, move defaults to hparams
* ggml-backend : add GGML_BACKEND_DEVICE_TYPE_IGPU device type
ggml-backend : add device id to device props
llama : only use iGPU devices if there are no GPU devices
llama : do not use multiple devices from different backends with the same device id
* support non-contiguous Q in build_attn_mha
* Update src/llama-graph.cpp
ggml-ci
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit adds check for two function pointers returned from
ggml_backend_reg_get_proc_address.
The motivation for this is that the function pointer could be nullptr if
the get proc address function changes in the future. This is also
consistent with all the other calls to ggml_backend_reg_get_proc_address
in the code base.
* feat: Add python-side constants and conversion for adapter.lora.invocation_string
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add c++ side constants for adapter.lora.invocation_string
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Parse invocation string for adapters from GGUF
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(python): Update conversion to alora_invocation_tokens
This is the preferred method in PEFT which is the source of ground truth
https://github.com/huggingface/peft/pull/2609/files#diff-13380145401d203d5935c5189dd09879f990b81aa63e8e3aaff8ce9110333f0e
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(cpp): Update to alora_invocation_tokens on c++ side
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add C APIs to get alora invocation token array from lora
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Initial implementation of alora cache logic in server
This does not yet do the part to identify the invocation tokens and only
apply the lora adapter afterwards, but it does seem to produce correct
results if the invocation tokens are the beginning of the uncached input.
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Identify alora invocation sequences
This currently limits to a single enabled alora per slot. Multiple aloras
with different invocation sequences would be possible, but it would require
a more complex integration of the adapter toggling and is not really a well
studied case for alora since it's unclear if one alora can reuse cache from
previous prefill computed with a different alora.
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Only reuse cache for tokens before the alora invocation start
This is a bit of an edge case, but theoretically a user could try the same
query with the alora disabled (just using the base model), then retry with
the alora. The cached tokens from the first pass should be invalid.
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Handle un-cached tokens that come before the alora activation
The solution is to only fill up to the token before the invocation start in
the batch if there are any tokens to be prefilled between those pulled from
cache and the invocation start. When this is detected, the alora is
temporarily disabled with a scale of 0.0, then immediately re-enabled after
it has been initialized for the internal graph. Since the batch does not
complete the prompt tokens, the remaining prompt tokens are handled in the
next task, pulling all of the non-alora tokens from cache and proceeding
with prefill for the alora tokens.
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use || instead of 'or'
Too much python 🤦
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fix off-by-one for limiting cached tokens to before alora start
This was the cause of the inconsistent results from the dummy test script
with and without the turn that runs the prompt without the adapter before
running it with the adapter.
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Support backwards-compatibility for "invocation_string" in adapter_config.json
While this has been replaced in the PEFT PR in favor of
alora_invocation_tokens, the existing adapters in the ibm-granite org on HF
use "invocation_string," so this will enable backwards compatibility and
enable testing now (before PEFT PR changes have percolated everywhere).
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Remove duplicate logging
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* feat: Report alora_invocation_string and alora_invocation_tokens from /lora-adapters
Branch: gabe-l-hart/alora-support
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit add support for the EmbeddingGemma 300m. This model supports
sliding window attention (SWA) and a new swq_type is introduced to
support symmetric SWA masking.
This commit also extracts the code from the function
llama_is_masked_swa in llama-impl.h, so that the logic can be shared
by both llm_graph_input_attn_no_cache::set_input and
llama_kv_cache::set_input_kq_mask.
With this commit the EmbeddingGemma 300m model can be converted to
to GGUF and used with llama.cpp.
Once the model has been uploaded to HuggingFace it can be used like
this:
```console
./build/bin/llama-cli -hf ggml-org/embeddinggemma-300m-GGUF:Q8_0
```
* llama : set n_outputs to 1 to avoid 0 outputs mean-pooling
This commit modifies the llama_context constructor to set n_outputs to
1.
The motivation for this is that when using pooling, and specifically
mean pooling, for embeddings having n_outputs set to 0 can lead to the
following error:
```console
$ build/bin/llama-embedding -m models/nomic-embed-text-1.5-Q4_K_M.gguf \
--pooling mean -p "Hello, how are you?"
...
llama_context: CPU output buffer size = 0.12 MiB
/home/danbev/work/ai/llama.cpp/ggml/src/ggml.c:3023: GGML_ASSERT(ggml_can_mul_mat(a, b)) failed
0x0000743c96d107e3 in __GI___wait4 (pid=292978, stat_loc=0x0, options=0, usage=0x0) at ../sysdeps/unix/sysv/linux/wait4.c:30
warning: 30 ../sysdeps/unix/sysv/linux/wait4.c: No such file or directory
30 in ../sysdeps/unix/sysv/linux/wait4.c
196 waitpid(child_pid, NULL, 0);
230 ggml_print_backtrace();
3023 GGML_ASSERT(ggml_can_mul_mat(a, b));
1823 cur = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, inp)), inp_mean);
18983 llm->build_pooling(cls, cls_b, cls_out, cls_out_b);
1399 auto * gf = model.build_graph(gparams);
292 auto * gf = graph_reserve(1, n_seqs, n_outputs, mctx.get(), true);
2329 auto * ctx = new llama_context(*model, params);
913 llama_context * lctx = llama_init_from_model(model, cparams);
105 common_init_result llama_init = common_init_from_params(params);
[Inferior 1 (process 292976) detached]
Aborted (core dumped)
```
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add comment about not reserving graphs with zero outputs
* add assert in graph_reserve to ensure n_outputs >= 1
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit fixes the model type for the Gemma 270M model in
llama_model.cpp which should be LLM_TYPE_270M. I incorrectly added this
previously as LLM_TYPE_537M which was wrong.
The motivation for this is that it causes the model to not be identified
properly when using tools like llama-bench. For example:
```console
$ ./build/bin/llama-bench -m models/gemma-3-270m-Q8_0.gguf
| model | size | ...
| ------------------------------ | ---------: | ...
| gemma3 ?B Q8_0 | 271.81 MiB | ...
| gemma3 ?B Q8_0 | 271.81 MiB | ...
```
With the changes in this commit the output will be:
```console
$ ./build/bin/llama-bench -m models/gemma-3-270m-Q8_0.gguf
| model | size | ...
| ------------------------------ | ---------: | ...
| gemma3 270M Q8_0 | 271.81 MiB | ...
| gemma3 270M Q8_0 | 271.81 MiB | ...
```
* sampling : optimize sorting using bucket sort in more places
ggml-ci
* sampling : do not sort in dist sampler
ggml-ci
* sampling : avoid heap allocations for sort buffers
ggml-ci
* common : add option to sort sampling candidates by probability
ggml-ci
* sampling : revert the change for preserving sort buffers
* sampling : use std::copy instead of memcpy
* sampling : clarify purpose of partial sort helpers
ggml-ci
* cont : remove wrong comment [no ci]
* common : update comment
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Exposes ggml_backend_sched_split_graph() to allow splitting the graph without allocating compute buffers and uses it to split the graph for the automatic Flash Attention check.
* feat: Add NEMOTRONH to python arch enum
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add NEMOTRONH to c++ arch enum
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add NEMOTRONH to llama-arch layer map
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: First pass at conversion for nemotronh
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add a verbose log for each tensor loaded
This is really helpful for diagnosing mismatches between the expected and
received tensors
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: First (broken) pass at nemotronh model architecture
It generates tokens, just not valid ones!
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Explicitly enable add_bos_token during conversion
The `tokenizer.json`/`tokenizer_config.json` in the model are a bit
contradictory. In the config, add_bos_token is set to False, but the
tokenizer model itself has a post_processor that adds the BOS token via
type: TemplateProcessing
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use relu2 (LLM_FFN_RELU_SQR) for activation in FFN layers
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Only allocate attention cache for attention layers (not non-recurrent)
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Move residual add to after every block
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use the correct norm tensor for the MLP blocks
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* Nemotron-H: MLP gate cleanup (pass NULL for unused gate)
This model does not use a gate in MLP blocks; pass NULLs for gate tensors to make intent clear and avoid unused-pointer noise.
* SSM: respect ssm_dt_rank for dt_dim when provided
Use GGUF-provided time_step_rank (ssm_dt_rank) to set dt_dim when > 0; fallback to max(64, n_embd/16).
* fix: plamo2 - revert dt_dim to default (remove ssm_dt_rank usage)
* Rename nemotronh to nemotron_h for consistency
- Update architecture name from NEMOTRONH to NEMOTRON_H in constants.py
- Change architecture string from 'nemotronh' to 'nemotron_h' in all files
- Update enum LLM_ARCH_NEMOTRONH to LLM_ARCH_NEMOTRON_H
- Update class name llm_build_nemotronh to llm_build_nemotron_h
- Consistent naming with underscore convention (nemotron_h vs nemotronh)
* feat: Support conversion for older NemotronH models
https://github.com/ggml-org/llama.cpp/issues/nemotron-nano-15409
Branch: gabe-l-hart/nvidia-nemotron-nano-15409
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Maicon Domingues <dominguesm@outlook.com>
Co-authored-by: weatherman <fxdstudios@gmail.com>
This commit adds support for the 18-layer model type in the Gemma3
series, which is the size of the Gemma3-270m model.
The motivation for this commit is was the only change required for
Gemma3-270m to be converted to GGUF format and used with llama.cpp.
Once the model has been converted and uploaded to Huggingface it can be
used like this:
```console
$ ./build/bin/llama-cli -hf ggml-org/gemma-3-270m-GGUF:Q8_0
```
* model : add harmony parser for gpt-oss
* gpt-oss : fix grammar trigger from causing empty stack
* gpt-oss: tweak the grammar trigger again
* gpt-oss : add support for recipient in role header
* gpt-oss : fix ungrouped tool calls in grammar
* gpt-oss : loosen function name matching during parse
* gpt-oss : clean up workarounds
* gpt-oss : add template tests
* gpt-oss : simulate thinking and tool call tags
* gpt-oss : undo think tags when reasoning_format is none
* gpt-oss : set special tokens back to user defined
* gpt-oss : update openai-gpt-oss template
* server : filter out harmony thought messages
* gpt-oss : simplify parsing
* server : add SWA checkpoints
ggml-ci
* cont : server clean-up
* server : handle state restore fails
* llama : add extended llama_state_seq_ API
* server : do not make checkpoints if --swa-full
ggml-ci
* llama : remove flags value for NONE
* server : configure number of SWA checkpoints with CLI arg
ggml-ci
* args : fix scope of new argument
When attempting to do llama-perplexity on certain tasks which have coupled sequences there is a cryptic error that does not tell you what to do, which is to set the -kvu flag. This adds a hint about that fact.
* examples/finetune -opt SGD (stochastic gradient descent) memory opt
add unit tested GGML_OPT_OPTIMIZER_SGD to ggml - avoids allocating
m, v tensors.
support finetune.cpp arg -opt SGD (or sgd). (default adamw as before)
llama 3.2-1b-F32 result: observed 11gb gpu ram (41 sec/epoch)
when using SGD instead of 19gb (55 sec/epoch) using adamw.
(wikipedia 100 lines finetune)
(
using the same GPU memory, adamw can only do before OOM 512
batch/context, reaching:
train: [███████▉] data=0000140/0000140 loss=0.02575±0.00099 acc=99.52±0.03% t=00:00:47 ETA=00:00:00
val: [███████▉] data=0000008/0000008 loss=4.76565±0.28810 acc=41.46±0.77% t=00:00:00 ETA=00:00:00
SGD is superior, though it converges slower, with max before OOM 1728
batch/context (esp see the better validation perf):
train: [███████▉] data=0000039/0000039 loss=0.00371±0.00010 acc=99.96±0.01% t=00:00:41 ETA=00:00:00
val: [███████▉] data=0000003/0000003 loss=5.11406±0.76034 acc=48.01±0.69% t=00:00:01 ETA=00:00:00
)
note: when finetuning long enough (or w/ enough -lr),
validation accuracy *eventually* drops ('catastrophic forgetting')
-lr-half (halflife) option useful for SGD to avoid oscillation or
super slow underdamped learning (makes setting -lr more forgiving).
terminal -lr for now is set by lr-halvings i.e. if you want at most
1/8 the inital -lr you set -lr-halvings 3.
note: objective loss not directly comparable between adamw, sgd? -
check perplexity or accuracy or consider relative improvements
for convergence
new finetune args -wd 1e-9 to enable weight decay in sgd or adamw,
and max -epochs N (default 2 as before)
cache (1 - wd*alpha) in 'adamw' opt struct -
no noticeable perf benefit, disabled (still done
for new SGD though)
since opt. memory is pre-allocated, the ggml_opt_get_optimizer_params
would probably be able to change between SGD and AdamW with each epoch
but would need to use adamw for the first (unconfirmed - no cmdline arg
to set such a policy yet)
test-opt checks adamw as before and now sgd (except for a few disabled
tests for sgd only; probably just needs logging values and adding
alternate reference values); tolerance on the 'regression'
test is broader for sgd (so we don't need many more epochs)
* Vulkan: Implement GGML_OP_OPT_STEP_SGD
* tests: Fix OPT_STEP_SGD test-backend-ops
* SGD op param store weight-decay and not 1-alpha*wd
* minor + cosmetic changes
* fix vulkan sgd
* try CI fix
---------
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
This commit updates `llama_kv_cache_unified::find_slot` to log
information for all streams when debug is enabled.
The motivation for this change is that currently if a non-unified
kv-cache is used, then only one stream will be logged because the
code was currently uses `seq_to_stream[1]`.
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
```
* Add support for Llada-8b: diffusion model
* Add README
* Fix README and convert_hf_to_gguf
* convert_hf_to_gguf.py: address review comments
* Make everything in a single example
* Remove model-specific sampling
* Remove unused argmax
* Remove braced initializers, improve README.md a bit
* Add diffusion specific gguf params in set_vocab, remove setting rope_theta and rms_norm_eps
* Remove adding the mask token
* Move add_add_bos_token to set_vocab
* use add_bool in gguf_writer.py
* support smallthinker
* support 20b softmax, 4b no sliding window
* new build_moe_ffn_from_probs, and can run 4b
* fix 4b rope bug
* fix python type check
* remove is_moe judge
* remove set_dense_start_swa_pattern function and modify set_swa_pattern function
* trim trailing whitespace
* remove get_vocab_base of SmallThinkerModel in convert_hf_to_gguf.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* better whitespace
Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* use GGML_ASSERT for expert count validation
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Improve null pointer check for probs
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* use template parameter for SWA attention logic
* better whitespace
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* move the creation of inp_out_ids before the layer loop
* remove redundant judge for probs
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* llama : clarify comment about pp and tg graphs [no ci]
This commit clarifies the comment in `llama-context.cpp` regarding the
prefill prompt (pp), and token generation (tg) graphs.
The motivation for this is that I've struggled to remember these and had
to look them up more than once, so I thought it would be helpful to add
a comment that makes it clear what these stand for.
* squash! llama : clarify comment about pp and tg graphs [no ci]
Change "pp" to "prompt processing".
MiniCPM models use the llm_build_granite constructor which was changed
in the Granite Four PR to use hparams.rope_finetuned instead of a
use_rope parameter. MiniCPM models need rope enabled by default.
Fixes inference from gibberish to correct responses.
* Support diffusion models: Add Dream 7B
* Move diffusion to examples
* Move stuff to examples. Add patch to not use kv-cache
* Address review comments
* Make sampling fast
* llama: remove diffusion functions
* Add basic timings + cleanup
* More cleanup
* Review comments: better formating, use LOG instead std::cerr, re-use batch, use ubatch instead of max_length
* fixup!
* Review: move everything to diffusion-cli for now
* Add PLaMo-2 model using hybrid memory module
* Fix z shape
* Add cmath to include from llama-vocab.h
* Explicitly dequantize normalization weights before RoPE apply
* Revert unnecessary cast because the problem can be solved by excluding attn_k, attn_q when quantizing
* Use ATTN_K/Q_NORM for k,q weights to prevent quantization
* Remove SSM_BCDT that is not used from anywhere
* Do not duplicate embedding weights for output.weight
* Fix tokenizer encoding problem for multibyte strings
* Apply suggestion from @CISC
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Use LLM_FFN_SWIGLU instead of splitting ffn_gate and ffn_up
* Remove unnecessary part for Grouped Query Attention
* Fix how to load special token id to gguf
* Remove unused tensor mapping
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Remove llama_vocab_plamo2 class and replace it with llm_tokenizer_plamo2_session to follow the other tokenizer implementations
* Update src/llama-vocab.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update convert_hf_to_gguf.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update convert_hf_to_gguf.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update convert_hf_to_gguf.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Fix plamo2 tokenizer session to prevent multiple calls of build()
---------
Co-authored-by: Francis Couture-Harpin <git@compilade.net>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* wip: llama : separate recurrent states from the KV cache
This will be necessary to support Jamba
(and other recurrent models mixed with Attention).
Doesn't compile yet, and finding a slot isn't yet done correctly for recurrent states.
* llama : use std::find for seq_nodes in llama_rs_cache
* llama : state checkpoints for recurrent models
* llama : correctly handle more edge cases for the rs cache
* llama : rename many llama_kv_cache_* functions
* llama : remove useless return value for some llama_cache_* functions
* llama : rethink recurrent state cell counts
* llama : begin work on support for variable GQA
This will also be useful for Jamba if we consider the Mamba layers
to have 0 KV heads.
* llama : gracefully fail when not finding hybrid slot
* llama : support Jamba
* llama : fix BERT inference without KV cache
* convert-hf : check for unprocessed Jamba experts
* convert-hf : support Mini-Jamba conversion
* llama : fix Jamba quantization sanity checks
* llama : sequence-length-aware batch splitting
* llama : use equal-sequence-length sub-batches for recurrent models
* ggml : simplify SSM-related operators
* llama : make recurrent state slot allocation contiguous
* llama : adapt internal uses of batches to llama_ubatch
* llama : fix batch split output count for embeddings
* llama : minimize swaps when reordering logits
This reduces overhead when running hellaswag
on thousands of sequences with very small 100k params Mamba models.
* llama : fix edge case finding batch seq_id of split recurrent cell
This otherwise was a problem when running the HellaSwag benchmark
with small batch sizes, making it crash.
* llama : avoid copies for simple batch splits
* llama : use im2col and mul_mat to perform convolution for Mamba
This removes the need for ggml_ssm_conv!!!
But performance seems slighly worse on my system,
especially for prompt processing.
Maybe ggml_mul_mat isn't optimized for small row sizes?
More performance testing is necessary until GGML_OP_SSM_CONV is removed.
* ggml : make ggml_ssm_scan not modify its source tensors
* llama : fix shared recurrent tail cell count for small ubatch sizes
Otherwise it was impossible to run the 'parallel' example with '-ub 1'
with a Mamba or Jamba model.
* llama : fix .base() compilation error on Windows
* llama : allow doing the equivalent of SSM_CONV with SUM_ROWS and MUL
* ggml : allow GGML_OP_CONCAT to work on non-contiguous tensors
The implementation already supported it,
and this makes Mamba's conv step slightly faster.
* llama : rename llama_cache to llama_past
This can be changed back later if the name change is wrong.
I was renaming the functions anyway to generalize kv-cache-related
functions to hybrid and recurrent model architectures.
I think llama_past is a better name than llama_cache for a combined
kv cache and recurrent state cache, because the states it contains
pretty much always come before the newly-added ones for any particular
sequence. Also 'llama_past_clear' sounds more obvious in what it does
than 'llama_kv_cache_clear'. The future is what the models generate.
(For embeddings, the kv cache isn't really used anyway)
Still, I'm open to better suggestions.
* examples : replace llama_kv_cache_seq_* with llama_past_seq_*
* mamba : fix non-contiguous usage of ggml_silu
* llama : initial Mamba-2 support
* ggml : SIMD ggml_ssm_scan for Mamba-2
* ggml : improve ggml_mul speed when masking recurrent states
* llama : support running Mamba-Codestral-7B-v0.1
* llama : fix Mamba-2 conv state saving
* ggml : make the ggml_mul fast broadcast path more consistently formatted
* llama : remove unused variable
* llama : add missing break
* convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present
The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires
workarounds to work correctly.
* llama : session saving and reloading for hybrid models
* convert_hf : fix Jamba conversion
* llama : fix mixed signedness comparison
* llama : use unused n_embd_k_gqa in k_shift
This also slightly reduces the diff from the master branch
* llama : begin renaming llama_past back to llama_kv_cache
* llama : avoid redundant state copy for Mamba 1 and 2
* metal : attempt to adapt SSM_SCAN for Mamba-2
* metal : fix SSM_SCAN pipeline scope
* metal : use log and exp instead of log1pf and expf in SSM_SCAN
* metal : remove unused arguments for SSM_SCAN
The max index is 31, so trimming the arguments is necessary.
* metal : add back n_seqs to SSM_SCAN args
Whoops, this is needed for the offset in the concatenated output.
* metal : fix SSM_SCAN state head offset
* metal : fix wrong number of tokens per sequence in SSM_SCAN
* ggml : remove unused fast broadcast path in GGML_MUL
This was initially added because states were masked with ggml_mul,
but this is no longer done and so this "optimisation" is no longer
necessary, or at least not worth the additional code complexity.
* ggml : avoid multiply by D in GGML_OP_SSM_SCAN
This makes the weight buft detection in src/llama.cpp simpler.
* convert : transpose Mamba-2 A, D and reshape SSM_NORM
This breaks existing conversions of Mamba-2 models
to avoid some reshapes.
Not sure if it's a good idea,
but it makes the graph slightly cleaner.
* llama : more appropriate SSM_SCAN and SSM_CONV buft support checks
* convert : fix flake8 lint
* llama : remove implicit recurrent state rollbacks
* llama : partially apply clang-format style
* metal : fix confusion between ; and ,
* metal : add missing args for nb references in ssm_scan_f32_group
* metal : single-user mamba2 inference works
* kv-cache : remove const_cast when setting inputs for s_copy
And also fix multi-user inference for recurrent models
by using cell_id instead of i as the kv cell index
when populating s_copy.
* convert : avoid AutoConfig for Mamba and Mamba2 hparams
* kv-cache : allow context shift for recurrent models
* graph : fix recurrent state copies when avoiding copies
Works, but using lambda functions might not be that clean.
* ggml : fix mamba2 ssm scan when compiled with SVE
* ggml-cpu : reorder SVE FMA for consistency with other SIMD arches
* cuda : implement ssm scan for Mamba2
There is still room for improvement, but it works!
* cuda : adapt Mamba1 ssm scan to shape changes from Mamba2
* feat: Add conversion for Bamba models
This is borrowed and adapted from the original implementation
https://github.com/ggml-org/llama.cpp/pull/10810
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add Granite 4 conversion
This is a manual copy from my draft branch
https://github.com/gabe-l-hart/llama.cpp/blob/GraniteFourDraft/convert_hf_to_gguf.py#L5076
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Plumb bamba through llama-arch
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add bamba to llama_arch_is_hybrid_recurrent
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add optional mamba ssm_in bias tensor
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add template specialization for get_arr to load a vector<uint32_t> for layer index arr in hparams
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Use an explicit bool to determine mamaba vs mamba2
This allows other architectures like bamba and granitemoehybrid to use
mamab2 without a growing architecture `if` statement inside the mamba
implementation.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Isolate mamba(2) and granite attention layer building in static methods
This will allow these layer-builder methods to be used from other build
structs without complex inheritance.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use per-layer sizes in granite build_attention_layer
Also no need to pass in kv cache since it's already in the inp_attn
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: First (broken) pass at end-to-end Bamba implementation
It generates (garbage) tokens! Still lots of debugging to do.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Only do Granite multipliers if set
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Pull granite ffn portion into a static function and reuse in hybrid
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(py): Allow gguf duplicate keys if they match by value and type
This is helpful for hybrid models that want to do gguf param setting by
calling multiple parent classes without needing to make those parent
classes try/except on every attempt to set a gguf value.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor(py): Simplify granitemoehybrid conversion to use parents better
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add GRANITE_MOE_HYBRID through llama-arch
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Support GRANITE_MOE_HYBRID in llama-model
This re-uses the Bamba code paths heavily and simply adds the missing parts
for loading MoE and the shared expert.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* style: Fix flake8 errors
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fix recurrent cache get after rebase
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fix hybrid granite implementation for signature changes in build_mamba*_layer
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Refactor relationship between non-hybrid classes and hybrid impl to use mixins
The challenge here is to give both the non-hybrid classes (llm_build_mamba
and llm_build_granite) AND the hybrid class (llm_build_hybrid_mamba) access
to the same intermediate "base class" functionality (build_mamba*_layer,
build_granite_attention_layer) without running into trouble with diamond
inheritance of llm_graph_context. Due to the non-trivial initialization
that happens in llm_graph_context, diamond inheritance results in multiple
initializations of the common base which cause problems around the unique
ptrs. I wanted to get away from `self->` everywhere, but this is still a
bit cleaner than making those methods static I think.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Implement the full copy-paste version to duplicate the layer builders
This follows the pattern where the type of input is pinned to the type of
memory and that is used to dispatch to the correct version of `build_rs` /
`build_attn`. There's a lot of code duplication that can hopefully be
pulled into common functions in the graph later.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Rename llm_build_hybrid_mamba -> llm_build_granite_hybrid
I've got back-and-forth a lot about how/if to try to implement reuse of the
"child model" layer types for hybrid models. At the end of the day, I think
hybrid models are their own beast and even if their layers are inspired by
other models, they should maintain control of their own layer building (in
other words, the copy-paste method). Given that, the name should reflect
that this is not a generic hybrid model builder, but rather a granite-
specific hybrid model builder that can do MoE (granite 4) or dense (bamba).
As part if this, I also cleaned up dangling comments from previous attempts
at using static methods for reusability.
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* mamba : fix mismatched new and delete size for llm_build_mamba
Subclasses of llm_graph_context cannot have extra fields,
because the called destructor is not the one from the subclass.
This otherwise would cause problems when runnning Mamba-(1|2) inference
when compiled -DGGML_SANITIZE_ADDRESS=ON
* memory : correctly handle failure in apply()
ggml-ci
* style: Remove TODO for adding first hybrid models to the switch
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fix bad merge in tensor_mapping.py w/ SSM_NORM
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fix bad merge resolution with variable renames/moves in llm_build_mamba
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* docs: Fix comment about duplicate key check
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Conform to standard way of initializing inp_out_ids
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* convert : fix jamba conv1d shape squeezing
* fix: Fix input initialization in granite_hybrid after removal of hybrid inputs
Branch: GraniteFourWithJamba
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use llm_graph_context_mamba in llm_build_granite_hybrid
Branch: GraniteFourWithJamba
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Refactor mamba2/granite/jamba/granite_hybrid relationships as mixins
The key is for the mixin classes (llm_graph_context_mamba,
llm_graph_context_granite) to use virtual inheritance from
llm_graph_context. This allows the common members to exist only once in the
class hierarchy. The downside is that llm_graph_context will be
re-initialized once for each parent (ie 2x for single mixin, 3x for two
mixins, etc...).
Branch: GraniteFourWithJamba
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* graph : add back hybrid memory graph input
But this time it contains the sub-cache graph inputs.
This *should* make it easier to handle updating the inputs
when caching the graph (eventually).
* model : add Jamba to Mamba-specific hparams printing
* fix: Fix input setup after upstream merge
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* jamba : remove redundant nullptr initializations
* model : remove unnecessary prefix for tensor loading constants
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* model : use ggml_swiglu_split for Mamba
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* feat: Add support for dense FFN in GraniteMoeHybrid
This was already partially supported via reusing the granite ffn builder,
and there may be models that leverage this architecture going forward. The
naming is a bit odd, but in the transformers version, it reuses the same
model class and simply has zero regular experts and a single shared expert
(which is the same as a single dense FFN).
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add support for dense FFN tensor names on c++ side
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use child inputs for Falcon H1 after merge resolution
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Remove unnecessary prefix on tensor constants
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* model : make falcon-h1 use shared mamba2 layer builder
* memory : avoid referring to KV in recurrent cache logs
* fix: Revert order changes for Falcon H1 to stay consistent with upstream
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* gguf-py : avoid adding duplicate tensor mappings for Jamba
Some of the tensor names are common with Llama4
* refactor: Collapse Bamba and GraniteMoeHybrid into GraniteHybrid
The only key difference is the use of rope which is now set via
rope_finetuned in the hparams
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Remove use of diamond inheritance
Per PR discussion, it's simpler to keep this with basic inheritance and not
introduce the complexity of virtual inheritance and multiple inheritance
https://github.com/ggml-org/llama.cpp/pull/13550#issuecomment-3053787556
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Log mamba params for Granite Hybrid
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Remove unused ssm_in_b
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Remove ATTENTION_LAYER_INDICES hparam in favor of n_head_kv
This matches how recurrent vs attention heads are identified for Jamba
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Remove unused template expansion for get_arr
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Review cleanup in convert_hf_to_gguf
The gist is to be explicit about which base class is being used with the
multiple inheritance setup
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Undo hidden warnings about duplicate identical keys in add_key_value
After further discussion, this encourages sloppy overwriting in the model
converters
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: If not using ROPE, context is "infinite"
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* doc: Add a comment outlining expected duplicate key warnings
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Remove unnecessary duplicate keys in converter
Co-authored-by: Francis Couture-Harpin <git@compilade.net>
(thanks for the sharp eyes and patience!)
Branch: GraniteFour
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Francis Couture-Harpin <git@compilade.net>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* wip: llama : separate recurrent states from the KV cache
This will be necessary to support Jamba
(and other recurrent models mixed with Attention).
Doesn't compile yet, and finding a slot isn't yet done correctly for recurrent states.
* llama : use std::find for seq_nodes in llama_rs_cache
* llama : state checkpoints for recurrent models
* llama : correctly handle more edge cases for the rs cache
* llama : rename many llama_kv_cache_* functions
* llama : remove useless return value for some llama_cache_* functions
* llama : rethink recurrent state cell counts
* llama : begin work on support for variable GQA
This will also be useful for Jamba if we consider the Mamba layers
to have 0 KV heads.
* llama : gracefully fail when not finding hybrid slot
* llama : support Jamba
* llama : fix BERT inference without KV cache
* convert-hf : check for unprocessed Jamba experts
* convert-hf : support Mini-Jamba conversion
* llama : fix Jamba quantization sanity checks
* llama : sequence-length-aware batch splitting
* llama : use equal-sequence-length sub-batches for recurrent models
* ggml : simplify SSM-related operators
* llama : make recurrent state slot allocation contiguous
* llama : adapt internal uses of batches to llama_ubatch
* llama : fix batch split output count for embeddings
* llama : minimize swaps when reordering logits
This reduces overhead when running hellaswag
on thousands of sequences with very small 100k params Mamba models.
* llama : fix edge case finding batch seq_id of split recurrent cell
This otherwise was a problem when running the HellaSwag benchmark
with small batch sizes, making it crash.
* llama : avoid copies for simple batch splits
* ggml : make ggml_ssm_scan not modify its source tensors
* llama : fix shared recurrent tail cell count for small ubatch sizes
Otherwise it was impossible to run the 'parallel' example with '-ub 1'
with a Mamba or Jamba model.
* llama : fix .base() compilation error on Windows
* llama : allow doing the equivalent of SSM_CONV with SUM_ROWS and MUL
* ggml : allow GGML_OP_CONCAT to work on non-contiguous tensors
The implementation already supported it,
and this makes Mamba's conv step slightly faster.
* mamba : fix non-contiguous usage of ggml_silu
* llama : session saving and reloading for hybrid models
* convert_hf : fix Jamba conversion
* llama : fix mixed signedness comparison
* llama : use unused n_embd_k_gqa in k_shift
This also slightly reduces the diff from the master branch
* llama : begin renaming llama_past back to llama_kv_cache
* llama : remove implicit recurrent state rollbacks
* llama : partially apply clang-format style
* convert : fix jamba conv1d shape squeezing
* graph : add back hybrid memory graph input
But this time it contains the sub-cache graph inputs.
This *should* make it easier to handle updating the inputs
when caching the graph (eventually).
* model : add Jamba to Mamba-specific hparams printing
* jamba : remove redundant nullptr initializations
* model : remove unnecessary prefix for tensor loading constants
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* model : use ggml_swiglu_split for Mamba
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* model : make falcon-h1 use shared mamba2 layer builder
* memory : avoid referring to KV in recurrent cache logs
* gguf-py : avoid adding duplicate tensor mappings for Jamba
Some of the tensor names are common with Llama4
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Splits producing more than one ubatch per batch for recurrent models
were broken with #14512.
This fixes it by moving the completeness check after the ubatch split loop.
* kv-cache : use ggml_set_rows
ggml-ci
* graph : separate k and v indices
ggml-ci
* cont : remove redundant ifs
ggml-ci
* kv-cache : improve find_slot impl
* kv-cache : bounds-check when accessing slot_info indices
* kv-cache : add comments
ggml-ci
* ggml : add TODOs for adding GGML_OP_SET_ROWS support in the backends
ggml-ci
* llama : initial Mamba-2 support
* ggml : SIMD ggml_ssm_scan for Mamba-2
* ggml : improve ggml_mul speed when masking recurrent states
* llama : support running Mamba-Codestral-7B-v0.1
* llama : fix Mamba-2 conv state saving
* ggml : make the ggml_mul fast broadcast path more consistently formatted
* llama : remove unused variable
* llama : add missing break
* convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present
The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires
workarounds to work correctly.
* llama : avoid redundant state copy for Mamba 1 and 2
* metal : attempt to adapt SSM_SCAN for Mamba-2
* metal : fix SSM_SCAN pipeline scope
* metal : use log and exp instead of log1pf and expf in SSM_SCAN
* metal : remove unused arguments for SSM_SCAN
The max index is 31, so trimming the arguments is necessary.
* metal : add back n_seqs to SSM_SCAN args
Whoops, this is needed for the offset in the concatenated output.
* metal : fix SSM_SCAN state head offset
* metal : fix wrong number of tokens per sequence in SSM_SCAN
* ggml : remove unused fast broadcast path in GGML_MUL
This was initially added because states were masked with ggml_mul,
but this is no longer done and so this "optimisation" is no longer
necessary, or at least not worth the additional code complexity.
* ggml : avoid multiply by D in GGML_OP_SSM_SCAN
This makes the weight buft detection in src/llama.cpp simpler.
* convert : transpose Mamba-2 A, D and reshape SSM_NORM
This breaks existing conversions of Mamba-2 models
to avoid some reshapes.
Not sure if it's a good idea,
but it makes the graph slightly cleaner.
* llama : more appropriate SSM_SCAN and SSM_CONV buft support checks
* convert : fix flake8 lint
* metal : fix confusion between ; and ,
* metal : add missing args for nb references in ssm_scan_f32_group
* metal : single-user mamba2 inference works
* kv-cache : remove const_cast when setting inputs for s_copy
And also fix multi-user inference for recurrent models
by using cell_id instead of i as the kv cell index
when populating s_copy.
* convert : avoid AutoConfig for Mamba and Mamba2 hparams
* kv-cache : allow context shift for recurrent models
* graph : fix recurrent state copies when avoiding copies
Works, but using lambda functions might not be that clean.
* ggml : fix mamba2 ssm scan when compiled with SVE
* ggml-cpu : reorder SVE FMA for consistency with other SIMD arches
* cuda : implement ssm scan for Mamba2
There is still room for improvement, but it works!
* cuda : adapt Mamba1 ssm scan to shape changes from Mamba2
* mamba : fix mismatched new and delete size for llm_build_mamba
Subclasses of llm_graph_context cannot have extra fields,
because the called destructor is not the one from the subclass.
This otherwise would cause problems when runnning Mamba-(1|2) inference
when compiled -DGGML_SANITIZE_ADDRESS=ON
* cuda : graceful fallback for Mamba-1 models with weird embd size
* implement unary REGLU/GEGLU/SWIGLU cpu ops
* relax constraints
* duplicate shape of source
* fix ggml_vec_geglu_f16
* special case gated ops
* implement unary REGLU/GEGLU/SWIGLU cuda ops
* tighten constraints again
* refactor into GGML_GLU_OP
* metal : add glu kernels
ggml-ci
* add CUDA_GLU_BLOCK_SIZE [no ci]
* more constraints and use 64bit ints
ggml-ci
* 64bit multiplication [no ci]
* implement swapped variants (cpu/cuda)
* update comment [no ci]
ggml-ci
* Vulkan: Add GLU ops and shaders
* SYCL: Implement fused kernel GEGLU, SWIGLU and REGLU for single up+gate
* ggml : implement GLU for split up/gate (#14181)
* implement GLU for split up/gate
* add tests for ggml_glu_split
* Vulkan: Implement glu_split logic and shader support
* add split to logging [no ci]
* SYCL: refactor element_size ops and add split up and gate support to gated kernels
* SYCL: switch GEGLU to use tanh approximation
---------
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Akarshan <akarshan@menlo.ai>
* GGML: increase OP count in assertion
* Refactor: Optimize SYCL element-wise operations with unary function inlining
This commit refactors the SYCL element-wise operations to improve performance by:
- Inlining unary operations (sgn, abs, elu, gelu, silu, etc.) to reduce kernel launch overhead.
- Introducing helper functions `op_xxx` for each unary operation to encapsulate the logic.
- Replacing direct kernel calls with calls to these inlined functions.
- Using `__dpct_inline__` to encourage compiler inlining.
- Minor code cleanup and consistency improvements.
The changes aim to reduce kernel launch overhead and improve the overall efficiency of element-wise operations on SYCL devices.
* vulkan: Increase workgroup size for GLU, for performance (#14345)
* vulkan: Increase workgroup size for GLU, for performance
* vulkan: change GLU shaders to do one element per invocation rather than one row per workgroup
* merge fix
* metal : add support for split and swap
ggml-ci
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Akarshan <akarshan@menlo.ai>
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
Bartowski
anavp-nvidia
Vinkal
Aaron Teo
Sigbjørn Skjæret
Aman Gupta
Douglas Hanley
Johannes Gäßler
Tarek Dakhran
Georgi Gerganov
Xuan-Son Nguyen
Jie Fu (傅杰)
Shane A
Haiyue Wang
Diego Devesa
ddh0
Daniel Bevenius
Gabe Goodhart
Piotr Wilkin (ilintar)
LaffeyNyaa
Tarek Dakhran
Aldehir Rojas
kallewoof
Jonathan Graehl
stevenkuang
Juk Armstrong
compilade
Sam
Csaba Kecskemeti
Ed Addario
Dongliang Wei
Gabriel Larson
Shunta Saito
yummy
l3utterfly
Molly Sophia
lgai-exaone
Nexes the Elder
Diner Burger
tempstudio
Dowon
Ryan Mangeno
ibrahim khadraoui
Weizhao Ouyang
Ruikai Peng