* feat(gguf-py): Add Granite model and params to gguf-py
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(convert_hf_to_gguf): Add registration and param setup for Granite
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(llama.cpp): Add config parsing for Granite multiplier params
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(llama.cpp): First pass at full port of granite deviations from llama
Something is still not working right since the results are mostly terrible,
but on occasion it's producing relevant results at this point, so
_something_ is working.
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(llama.cpp): Determine granite language 3b instruct by vocab size
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(convert_hf_to_gguf): Use LlamaModel as base for GraniteModel
The defaults in LlamaModel are needed for Granite as well
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(llama.cpp): Switch Granite param names to use _scale for consistency
Other scalar multipliers are called *_scale, so this provides a more
consistent naming convention.
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(convert_hf_to_gguf/gguf-py): _multiplier -> _scale
The transformers names with _multiplier will now be converted to the _scale
equivalent during conversion.
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(llama.cpp): Use separate switch clause for granite in llm_load_hparams
Branch: GraniteLM
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* gguf-py, llama : add constants and methods related to Llama-3.1 <|eom_id|> token
* llama : find Llama-3.1 <|eom_id|> token id during vocab loading
* llama-vocab : add Llama-3.1 <|eom_id|> token to the set of tokens stopping the generation
---------
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
* gguf_writer.py: add_array() should not add to kv store if empty
* Apply suggestions from code review
I was wondering if there was a specific reason for `if val` but good to hear we can safely use `len(val == 0`
Co-authored-by: compilade <git@compilade.net>
---------
Co-authored-by: compilade <git@compilade.net>
Main thing is that the default output filename will take this form
{name}{parameters}{finetune}{version}{encoding}{kind}
In addition this add and remove some entries in the KV store and adds a metadata class with automatic heuristics capability to derive some values based on model card content
* No Change:
- Internal GGUF Spec
- `general.architecture`
- `general.quantization_version`
- `general.alignment`
- `general.file_type`
- General Model Details
- `general.name`
- `general.author`
- `general.version`
- `general.description`
- Licensing details
- `general.license`
- Typically represents the converted GGUF repo (Unless made from scratch)
- `general.url`
- Model Source during conversion
- `general.source.url`
* Removed:
- Model Source during conversion
- `general.source.huggingface.repository`
* Added:
- General Model Details
- `general.organization`
- `general.finetune`
- `general.basename`
- `general.quantized_by`
- `general.size_label`
- Licensing details
- `general.license.name`
- `general.license.link`
- Typically represents the converted GGUF repo (Unless made from scratch)
- `general.doi`
- `general.uuid`
- `general.repo_url`
- Model Source during conversion
- `general.source.doi`
- `general.source.uuid`
- `general.source.repo_url`
- Base Model Source
- `general.base_model.count`
- `general.base_model.{id}.name`
- `general.base_model.{id}.author`
- `general.base_model.{id}.version`
- `general.base_model.{id}.organization`
- `general.base_model.{id}.url` (Model Website/Paper)
- `general.base_model.{id}.doi`
- `general.base_model.{id}.uuid`
- `general.base_model.{id}.repo_url` (Model Source Repository (git/svn/etc...))
- Array based KV stores
- `general.tags`
- `general.languages`
- `general.datasets`
---------
Co-authored-by: compilade <git@compilade.net>
Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
* lora: load to devide buft
* add patch tensor function
* correct tensor patch
* llama_lora_adapter_apply
* correct ggml_backend_tensor_copy
* add llm_build_mm
* fix auto merge
* update based on review comments
* add convert script
* no more transpose A
* add f16 convert
* add metadata check
* add sanity check
* fix ftype
* add requirements
* fix requirements
* fix outfile
* conversion: only allow selected models
* fix types
* cuda : do not use dmmv if the tensor does not have enough cols
* llama : lora fixes
* do not disable mmap with lora
Co-authored-by: slaren <slarengh@gmail.com>
* llm_build_lora_mm_id
* convert_lora : MoE LoRA conversion support
* convert_lora : prefer safetensors, similarly to convert_hf
* convert_hf : simplify modify_tensors for InternLM2
* convert_lora : lazy conversion
* llama : load and use alpha from LoRA adapters
* llama : use llm_build_lora_mm in most model graphs
* auto scale
* Revert "auto scale"
This reverts commit 42415a4874.
* remove redundant params
* Apply suggestions from code review
Co-authored-by: slaren <slarengh@gmail.com>
* change kv metadata
* move add_type to __init__
* convert_hf : move add_type to main()
* convert_lora : use the GGUFWriter from Model instead of overwriting it
---------
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Francis Couture-Harpin <git@compilade.net>
* Initial OpenELM support (270M only so far)
* Fill out missing entries in llama_model_type_name
* fixup! Initial OpenELM support (270M only so far)
Fix formatting
* llama : support all OpenELM models
* llama : add variable GQA and variable FFN sizes
Some metadata keys can now also be arrays to support setting
their value per-layer for models like OpenELM.
* llama : minor spacing changes
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* llama : use std::array for per-layer hparams
* llama : fix save/load state
* llama : do not print hparams for vocab-only models
* llama : handle n_head == 0
* llama : use const ref for print_f and fix division by zero
* llama : fix t5 uses of n_head and n_ff
* llama : minor comment
---------
Co-authored-by: Francis Couture-Harpin <git@compilade.net>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add attention and final logit softcapping.
* fix
* Add custom add_ functions
* Disable flash attention for Gemma2
* Update src/llama.cpp
Co-authored-by: slaren <slarengh@gmail.com>
* Add default value for attention and final logit softcap value
* Add custom kq scaling from Gemma2Attention
* Remove custom pre attention scaling and use computed value instead.
---------
Co-authored-by: slaren <slarengh@gmail.com>
* support splits in convert.py
* Support split by size and dry run to write estimated shards/filesizes
* Move split functionality to new GGUFManager class
* fix improper function signature
* tentative push of convert-hf-to-gguf support
* resolve merge + SplitArguments for easier parsing
* Fix eager tensor memory leak and remove convert.py changes
Removed a memory leak caused by unexpected reference retention to eager tensors.
Also removed GGUFManager functionality in convert.py in favor of specializing for convert-hf-to-gguf.py.
* refactor SplitStrategy to be a deque
Instead of having SplitStrategy have a `data` field that is a deque, just have SplitStrategy be a subclass of deque itself.
* fix Q8 quantization
* remove unnecessary imports in gguf_manager
* fix final? merge issue
* fix gguf_writer placement and remove comments
* oops, actually fix gguf_writer placement
* reduce duplicated code from gguf_writer
* further simplify GGUFManager
* simplify even further and standardize with GGUFWriter
* reduce diffs with master
* form shards while adding tensors, SHA256 sums agree with master
* re-add type hint
Co-authored-by: compilade <git@compilade.net>
* GGUFWriter compatibility fix
Co-authored-by: compilade <git@compilade.net>
* Shard dataclass and un-negative dont_add_architecture
* type consistency in format_n_bytes_to_str
* move kv keys to constants.py
* make pathlib explicit
* base-1024 bytes to base-1000
* rename GGUFManager to GGUFWriterSplit
* Update gguf-py/gguf/constants.py
Co-authored-by: compilade <git@compilade.net>
* fix convert-hf-to-gguf.py permissions
* fix line endings
* Update gguf-py/gguf/gguf_writer_split.py
Co-authored-by: compilade <git@compilade.net>
* convert-hf : restore executable file permission
* examples/convert-legacy-llama.py: restore executable file permission
* reinstate original gguf package import and fix type annotation
* attempt to appease the linter
* attempt 2 to appease the linter
* attempt 3 to appease the linter
* comma consistency
* Update convert-hf-to-gguf.py
Co-authored-by: compilade <git@compilade.net>
* edit cmd line args
* use simplification from #7827
* kv/ti data are still wrong
* try to refactor kv data (still fails)
* fix ti data messiness
* tidy up
* fix linting
* actually make the linter happy
* cleanup round 1
* remove SplitStrategy, SplitArguments
* appease linter
* fix typing and clean up
* fix linting
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* progress bar, fix split logic
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* catch oversights
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* swap bar orders
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* compatibility fix
* Update gguf-py/gguf/gguf_writer.py
Co-authored-by: compilade <git@compilade.net>
* Update convert-hf-to-gguf.py
Co-authored-by: compilade <git@compilade.net>
---------
Co-authored-by: Brian <mofosyne@gmail.com>
Co-authored-by: compilade <git@compilade.net>
* gguf-py : add T5 model architecture
* gguf-py : add separate tensors for encoder and decoder
* gguf-py : add new model header parameters: decoder_start_token_id, attention.relative_buckets_count, tokenizer.ggml.remove_extra_whitespaces, tokenizer.ggml.precompiled_charsmap
* convert-hf : add model conversion support for T5ForConditionalGeneration and T5WithLMHeadModel
---------
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
* update: convert-hf-to-gguf.py to support Qwen2-57B-A14B
* fix: QWEN2MOE support for expert_feed_forward_length
previously, expert ff was taken from n_ff (intermediate size) but it is now properly taken from LLM_KV_EXPERT_FEED_FORWARD_LENGTH
n_ff_exp and n_ff_shared_exp are now properly calculated
* update: convert-hf-to-gguf.py cleanup for Qwen2MoeForCausalLM
* fix: QWEN2MOE support for expert_feed_forward_length
previously, expert ff was taken from n_ff (intermediate size) but it is now properly taken from LLM_KV_EXPERT_FEED_FORWARD_LENGTH
n_ff_exp and n_ff_shexp are now properly calculated
Main changes of this PR is to consolidate GGUFWriter.add_key and GGUFWriter.add_val into GGUFWriter.add_key_value.
In addition use_temp_file is now opt-in instead of opt-out defaulting to False.
Also GGUFWriter now does not require output file name until when actually writing to it.
And GGUFWriter doesn't really need to eagerly prepare the data layout of the metadata
* common : increase max number of experts to 160
* common : add tensors ATTN_Q_A, ATTN_Q_A_NORM, ATTN_Q_B, ATTN_KV_A_MQA, ATTN_KV_A_NORM, ATTN_KV_B needed by DeepSeek-V2 MLA (multi-head latent attention) architecture
* common : add model header parameters: leading_dense_block_count, expert_feed_forward_length, expert_shared_count, expert_weights_scale, attention.q_lora_rank, attention.kv_lora_rank, rope.scaling.yarn_log_multiplier
* convert-hf : add model conversion support for DeepseekV2ForCausalLM
* llama : add model types for DeepSeek-V2 and DeepSeek-V2-Lite models
* llama : add two new llm_build_moe_ffn() arguments: scale_w (whether to scale weights of selected MoE experts) and w_scale (numerical value of the scaling factor)
* llama : add inference support for LLM_ARCH_DEEPSEEK2
---------
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
* add phi3 128k support in convert-hf-to-gguf
* add phi3 128k support in cuda
* address build warnings on llama.cpp
* adjust index value in cuda long rope freq factors
* add long rope support in ggml cpu backend
* make freq factors only depend on ctx size
* remove unused rope scaling type 'su' frin gguf converter
* fix flint warnings on convert-hf-to-gguf.py
* set to the short freq factor when context size is small than trained context size
* add one line of comments
* metal : support rope freq_factors
* ggml : update ggml_rope_ext API to support freq. factors
* backends : add dev messages to support rope freq. factors
* minor : style
* tests : update to use new rope API
* backends : fix pragma semicolons
* minor : cleanup
* llama : move rope factors from KV header to tensors
* llama : remove tmp assert
* cuda : fix compile warning
* convert : read/write n_head_kv
* llama : fix uninitialized tensors
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* convert-hf : support q8_0 conversion
* convert-hf : add missing ftype
This was messing with the checksums otherwise.
* convert-hf : add missing ftype to Baichuan and Xverse
I didn't notice these on my first pass.
* convert-hf : support bfloat16 conversion
* gguf-py : flake8 fixes
* convert-hf : add missing space after comma
* convert-hf : get bit-exact same output as ./quantize
The quantization version was missing.
* convert-hf : don't round bf16 NANs
* convert-hf : save some memory with np.int16 intermediate bf16 weights
* convert-hf : more closely match llama.cpp with which weights to keep in f32
* convert-hf : add --outtype auto-f16
A reason for this to exist is for model quantizers who want an initial
GGUF with the most fidelity to the original model while still using
a 16-bit float type instead of 32-bit floats.
* convert-hf : remove a semicolon because flake8 doesn't like it
It's a reflex from when programming in C/C++, I guess.
* convert-hf : support outtype templating in outfile name
* convert-hf : rename --outtype auto-f16 to --outtype auto
* convert-hf : begin refactoring write_tensor
* convert : upgrade to sentencepiece v0.2.0
* convert-hf : remove unused n_dims in extra_*_tensors
* convert-hf : simplify MoE weights stacking
* convert-hf : flake8 linter doesn't like semicolons
* convert-hf : allow unusual model part names
For example, loading `model-00001-of-00001.safetensors` now works.
* convert-hf : fix stacking MoE expert tensors
`torch.stack` and `torch.cat` don't do the same thing.
* convert-hf : fix Mamba conversion
Tested to work even with a SentencePiece-based tokenizer.
* convert : use a string for the SentencePiece tokenizer path
* convert-hf : display tensor shape
* convert-hf : convert norms to f32 by default
* convert-hf : sort model part names
`os.listdir` is said to list files in arbitrary order.
Sorting the file names should let "model-00009-of-00042.safetensors"
be loaded before "model-00010-of-00042.safetensors".
* convert-hf : use an ABC for Model again
It seems Protocol can't be used as a statically type-checked ABC,
because its subclasses also can't be instantiated. (why did it seem to work?)
At least there's still a way to throw an error when forgetting to define
the `model_arch` property of any registered Model subclasses.
* convert-hf : use a plain class for Model, and forbid direct instantiation
There are no abstract methods used anyway,
so using ABC isn't really necessary.
* convert-hf : more consistent formatting of cmdline args
* convert-hf : align the message logged for converted tensors
* convert-hf : fix Refact conversion
* convert-hf : save memory with lazy evaluation
* convert-hf : flake8 doesn't like lowercase L as a variable name
* convert-hf : remove einops requirement for InternLM2
* convert-hf : faster model parts loading
Instead of pre-loading them all into a dict, iterate on the tensors
in the model parts progressively as needed in Model.write_tensors
Conversion for some architectures relies on checking for the presence
of specific tensor names, so for multi-part models, the weight map is read
from the relevant json file to quickly get these names up-front.
* convert-hf : minor changes for consistency
* gguf-py : add tqdm as a dependency
It's small, and used for a progress bar
in GGUFWriter.write_tensors_to_file
* convert.py: add python logging instead of print()
* convert.py: verbose flag takes priority over dump flag log suppression
* convert.py: named instance logging
* convert.py: use explicit logger id string
* convert.py: convert extra print() to named logger
* convert.py: sys.stderr.write --> logger.error
* *.py: Convert all python scripts to use logging module
* requirements.txt: remove extra line
* flake8: update flake8 ignore and exclude to match ci settings
* gh-actions: add flake8-no-print to flake8 lint step
* pre-commit: add flake8-no-print to flake8 and also update pre-commit version
* convert-hf-to-gguf.py: print() to logger conversion
* *.py: logging basiconfig refactor to use conditional expression
* *.py: removed commented out logging
* fixup! *.py: logging basiconfig refactor to use conditional expression
* constant.py: logger.error then exit should be a raise exception instead
* *.py: Convert logger error and sys.exit() into a raise exception (for atypical error)
* gguf-convert-endian.py: refactor convert_byteorder() to use tqdm progressbar
* verify-checksum-model.py: This is the result of the program, it should be printed to stdout.
* compare-llama-bench.py: add blank line for readability during missing repo response
* reader.py: read_gguf_file() use print() over logging
* convert.py: warning goes to stderr and won't hurt the dump output
* gguf-dump.py: dump_metadata() should print to stdout
* convert-hf-to-gguf.py: print --> logger.debug or ValueError()
* verify-checksum-models.py: use print() for printing table
* *.py: refactor logging.basicConfig()
* gguf-py/gguf/*.py: use __name__ as logger name
Since they will be imported and not run directly.
* python-lint.yml: use .flake8 file instead
* constants.py: logger no longer required
* convert-hf-to-gguf.py: add additional logging
* convert-hf-to-gguf.py: print() --> logger
* *.py: fix flake8 warnings
* revert changes to convert-hf-to-gguf.py for get_name()
* convert-hf-to-gguf-update.py: use triple quoted f-string instead
* *.py: accidentally corrected the wrong line
* *.py: add compilade warning suggestions and style fixes
* Support converting models with multiple chat templates
Adds the following metadata:
* tokenizer.chat_templates
* tokenizer.chat_template.<name1>
* tokenizer.chat_template.<name2>
* tokenizer.chat_template.<...>
Where `tokenizer.chat_templates` is an array of the template names (except `default`), `default` is added to the regular `tokenizer.chat_template`.
* replace filtered characters with underscore
* New script to add/modify/remove metadata
This scripts creates a copy of a GGUF file and allows you to add/modify/remove metadata in the process.
Most importantly this allows you to update chat templates, either as a string or directly from an updated tokenizer_config.json file.
* Add files via upload
add new script to project/readme
* flake--
This commit adds special token metadata for Fill-In-the-Middle
(FIM)/Infill to the GGUF model.
The motivation for this is that currently there is support for CodeLlama
but other models exist now like CodeGemma, but the different models use
different token ids for the special tokens and this commit allows for
supporting multiple models.
Signed-off-by: Daniel Bevenius <daniel.bevenius@gmail.com>
Information about the Command-R 35B model (128k context) can be found at:
https://huggingface.co/CohereForAI/c4ai-command-r-v01
Based on the llama2 model with a few changes:
1) New hyper parameter to scale output logits (logit_scale)
2) Uses LayerNorm instead of RMSNorm
3) Transfomer layers have a single shared LayerNorm that feeds into both the
self-attention and FFN layers in parallel. There is no post-attention LayerNorm.
4) No support for Rotary Position Embeddings (RoPE) scaling
5) No biases used
Find GGUF files here:
https://huggingface.co/andrewcanis/c4ai-command-r-v01-GGUF
To convert model to GGUF format yourself:
1) Download Command-R Hugging Face safetensors:
git lfs install
git clone https://huggingface.co/CohereForAI/c4ai-command-r-v01
2) Run:
python3 convert-hf-to-gguf.py --outtype f16 ./c4ai-command-r-v01
* gguf : add support for I64 and F64 arrays
GGML currently does not support I64 or F64 arrays and they are not often
used in machine learning, however if in the future the need arises, it
would be nice to add them now, so that the types are next to the other
types I8, I16, I32 in the enums, and it also reserves their type number.
Furthermore, with this addition the GGUF format becomes very usable for
most computational applications of NumPy (being compatible with the most
common NumPy dtypes: i8, i16, i32, i64, f32, f64), providing a faster,
and more versatile alternative to the `npz` format, and a simpler
alternative to the `hdf5` format.
The change in this PR seems small, not significantly increasing the
maintenance burden. I tested this from Python using GGUFWriter/Reader
and `gguf-dump`, as well as from C, everything seems to work.
* Fix compiler warnings
* additional methods to read model and ctx parameters
* vocab size as a part of a model metadata
* models without vocabulary, convert.py part
* models without vocabulary, llama.cpp part
* PR clean up
* converter scrypt fixes
* llama_vocab_type update (renamed the new key)
* pr review fixes
* revert function renaming
* one more NoVocab assert
* Refactor dtype handling to be extensible
This code is equivalent as before, but now it is prepared to easily add
more NumPy dtypes.
* Add support for I8, I16 and I32
These types are allowed in the GGUF specification.
* Add support for I8, I16 and I32 to gguf_writer
* Add support for I8, I16, I32 to gguf_reader
* mamba : begin working on support for Mamba SSM
* mamba : begin figuring out how to (ab)use the kv cache for Mamba
* mamba : recurrent inference almost works, but incoherent
* mamba : recurrent inference WORKS!!!
* convert : optionally use d_conv and d_state from config.json for Mamba
* mamba : refactor recurrent conv, resulting in 20% perf increase
It's still slower than I'd like, but I did not really optimize `ggml_exp` yet.
I also refactored `ggml_exp` to work with tensors with more than 2 dimensions.
* ggml : parallelize ggml_exp
This results in 8% faster token generation for Mamba-130M.
* mamba : simplify the conv step with a self-overlapping view
Turns out the conv_state can be made smaller by one column.
Note that this breaks existing GGUFs of Mamba,
because the key_value_length field is tied to the conv_state size.
Convolution with a self-overlapping view is cool!
And it's much simpler than what I initially thought would be necessary
to make the convolution step work with more than 1 token at a time.
Next step is to make the SSM step work on batches of tokens too,
and thus I need to figure out a way to make a parallel selective scan
which will keep the ssm_state small and won't make it bigger
by a factor of (n_layer * batch_size).
* llama : fix Mamba KV self size wrongly displaying as f16 instead of f32
Relatedly, I also tried to see if other types than f32 worked for the states,
but they don't, because of the operators used.
It's probably better anyway to keep lots of precision there,
since the states are small anyway.
* mamba : fix self-overlapping view depth stride
* mamba : handle batches of more than 1 token
This means running Mamba no longer crashes when using the default settings!
And probably also slightly faster prompt processing.
Both batched and non-batched processing yield the same output.
Previously, the state was not cleared when starting a sequence.
Next step is to make the KV cache API work as expected for Mamba models.
* ggml: add ggml_ssm_scan to help with parallel selective scan
If the selective scan was implemented without a custom operator,
there would be waaay too many nodes in the graph. For example,
for Mamba-130M, with a batch size of 512 (the default),
a naive selective scan could add at least 24*512=12288 nodes,
which is more than LLAMA_MAX_NODES (8192),
and that's only for the smallest Mamba model.
So it's much cleaner with a custom operator.
Not sure about the name, though.
* ggml : in ggml_ssm_scan, merge multiple rows in the same vec operation
This will help with performance on CPU if ggml_vec_mul_f32
and ggml_vec_add_f32 are ever optimized with SIMD.
* mamba : very basic quantization support
Mostly works, but there is currently no difference
between the variants of a k-quant (e.g. Q4_K_S and Q4_K_M are the same).
Most of the SSM-specific weights can be kept in f32 without affecting
the size that much, since they are relatively small.
(the linear projection weights are responsible for most of Mamba's size)
Too much quantization seems to make the state degrade quite fast, and
the model begins to output gibberish.
It seems to affect bigger models to a lesser extent than small models,
but I'm not sure by how much.
Experimentation will be needed to figure out which weights are more important
for the _M (and _L?) variants of k-quants for Mamba.
* convert : fix wrong name for layer norm weight of offical Mamba models
I was using Q-bert/Mamba-* models before, which have a slighlty different
naming scheme for the weights.
(they start with "model.layers" instead of "backbone.layers")
* mamba : fuse more steps of the SSM scan in the ggml_ssm_scan operator
This increases performance on CPU by around 30% for prompt processing,
and by around 20% for text generation.
However, it also makes the ggml_exp and ggml_soft_plus operators unused.
Whether or not they should be kept will be decided later.
* convert : for Mamba, also consider the "MambaLMHeadModel" arch name
It's the name of the class of the official implementation,
though they don't use it (yet) in the "architectures" field of config.json
* mamba : fix vocab size problems with official models
The perplexity was waaaay to high for models with a non-round vocab size.
Not sure why, but it needed to be fixed in the metadata.
Note that this breaks existing GGUF-converted Mamba models,
but **only if** the vocab size was not already rounded.
* ggml : remove ggml_exp and ggml_soft_plus
They did not exist anyway outside of this branch,
and since ggml_ssm_scan fused operations together, they are unused.
It's always possible to bring them back if needed.
* mamba : remove some useless comments
No code change.
* convert : fix flake8 linter errors
* mamba : apply suggestions from code review
* mamba : remove unecessary branch for row-wise ssm_state and C multiplication
It was previously done to avoid permuting when only one token is processed
at a time (like when generating text), but permuting is cheap,
and dynamically changing the compute graph is not future-proof.
* ggml : in ggml_ssm_scan, use more appropriate asserts
* ggml : rename the destination pointer in ggml_compute_forward_ssm_scan_f32
* mamba : multiple sequences, but one at a time
This is a step towards making this Mamba implementation usable
with the server example (the way the system prompt is kept when clearing
the client slots will need to be changed before this can work, though).
The KV cache size for this kind of model is tied to the maximum number
of sequences kept at any single time.
For now, this number is obtained from n_parallel (plus one,
to have an extra sequence to dedicate to the system prompt),
but there might be a better way to do this which won't also
make the main example use 2 cells even if only 1 is really used.
(for this specific case, --parallel 0 helps)
Simultaneous sequence processing will probably require changes to
ggml_ssm_scan, and possibly a new operator for the conv step.
* mamba : support llama_kv_cache_seq_cp
This (mis)uses the logic around K shifts, because tokens in a state
can't be shifted anyway, and because inp_K_shift has the right shape and type.
Using ggml_get_rows is a nice way to do copies, but copy chains can't work.
Fortunately, copy chains don't really seem to be used in the examples.
Each KV cell is dedicated to the sequence ID corresponding to its own index.
* mamba : use a state mask
It's cleaner than the previous heuristic of
checking for the pos of the first token in the batch.
inp_KQ_mask could not be re-used for this, because it has the wrong shape
and because it seems more suited to the next step of
simultaneous sequence processing (helping with the problem of
remembering which token belongs to which sequence(s)/state(s)).
* llama : replace the usage of n_ctx with kv_self.size in many places
* mamba : use n_tokens directly instead of n_tok
* mamba : in comments, properly refer to KV cells instead of slots
* mamba : reduce memory usage of ggml_ssm_scan
From 290.37 MiB to 140.68 MiB of CPU compute buffer size
with Mamba 3B with a batch size of 512.
The result tensor of ggml_ssm_scan was previously a big part
of the CPU compute buffer size. To make it smaller,
it does not contain the intermediate ssm states anymore.
Both y and the last ssm state are combined in the result tensor,
because it seems only a single tensor can be returned by an operator
with the way the graph is built.
* mamba : simultaneous sequence processing
A batch can now contain tokens from multiple sequences.
This is necessary for at least the parallel example, the server example,
and the HellaSwag test in the perplexity example.
However, for this to be useful, uses of llama_kv_cache_seq_rm/cp
will need to be changed to work on whole sequences.
* ggml : add ggml_ssm_conv as a new operator for the conv step of Mamba
This operator makes it possible to use and update the correct states
for each token of the batch in the same way as ggml_ssm_scan.
Other solutions which use existing operators would need loops which would
add too many nodes to the graph (at least the ones I thought of).
Using this operator further reduces the size of the CPU compute buffer
from 140.68 MiB to 103.20 MiB with Mamba 3B with a batch size of 512.
And (at least on CPU), it's a bit faster than before.
Note that "ggml_ssm_conv" is probably not the most appropriate name,
and it could be changed if a better one is found.
* llama : add inp_s_seq as a new input tensor
The most convenient implementation to select the correct state (for Mamba)
for each token is to directly get the correct index from a tensor.
This is why inp_s_seq is storing int32_t and not floats.
The other, less convenient way to select the correct state would be
to have inp_KQ_mask contain 1.0f for each state used by a token
and 0.0f otherwise. This complicates quickly fetching the first used
state of a token, and is also less efficient because a whole row
of the mask would always need to be read for each token.
Using indexes makes it easy to stop searching when there are
no more sequences for a token, and the first sequence assigned
is always very quickly available (it's the first element of each row).
* mamba : support llama_kv_cache_seq_cp copy chains
* mamba : support shifting and dividing the kv cache pos
* mamba : make the server and parallel examples work with whole sequences
A seq_id is dedicated to the system prompt in both cases.
* llama : make llama_kv_cache_seq_rm return whether it succeeded or not
* mamba : dedicate an input tensor for state copy indices
This is cleaner and makes it easier to adapt when/if token positions
(and by extension, inp_K_shift) are no longer integers.
* mamba : adapt perplexity, batched, and batched-bench examples
* perplexity : limit the max number of sequences
This adapts to what the loaded model can provide.
* llama : add llama_n_max_seq to get the upper limit for seq_ids
Used by the perplexity example.
* batched : pass n_parallel to the model's context params
This should have been there already, but it wasn't.
* batched-bench : reserve sequences to support Mamba
* batched-bench : fix tokens being put in wrong sequences
Generation quality isn't what's measured in there anyway,
but at least using the correct sequences avoids using non-consecutive
token positions.
* mamba : stop abusing attention metadata
This breaks existing converted-to-GGUF Mamba models,
but will allow supporting mixed architectures like MambaFormer
without needing to break Mamba models.
This will also allow changing the size of Mamba's states
without having to reconvert models in the future.
(e.g. using something else than d_conv - 1 columns for the conv_states
will not require breaking existing converted Mamba models again)
* gguf-py : add new KV metadata key-value pairs for Mamba
* llama : add new metadata key-value pairs for Mamba
* llama : guard against divisions by zero when n_head is 0
* mamba : rename "unlimited" KV cache property to "recurrent"
* mamba : more correctly update the "used" field of the KV cache
* ggml : in ggml_ssm_scan, use a threshold for soft_plus
This is how the official Mamba implementation does it,
and it's also what torch.nn.Softplus does.
* convert : for Mamba, fallback to internal NeoX tokenizer
The resulting models are exactly the same
as if the tokenizer.json and tokenizer_config.json of GPT-NeoX were there.
* mamba : support state saving and restoring
* ggml : implicitly pass src tensors through dst for Mamba-related ops
* mamba : clarify some comments
* server : fix cache_tokens not getting correctly resized
Otherwise, when the "we have to evaluate at least 1 token" special case
was triggered, an extra token was kept in cache_tokens even if it was
removed from the KV cache.
For Mamba, this caused useless prompt reprocessing when the previous
request triggered the above case.
* convert-hf : support new metadata keys for Mamba
For the models available at
https://huggingface.co/collections/state-spaces/transformers-compatible-mamba-65e7b40ab87e5297e45ae406
* mamba : rename metadata to be more similar to transformers library
This breaks existing converted-to-GGUF models,
but the metadata names are more "standard".
* mamba : support mamba-*-hf models
These models share their token_embd.weight with their output.weight
* mamba : add missing spaces
This is purely a formatting change.
* convert-hf : omit output.weight when identical with token_embd.weight
Only for Mamba for now, but it might be relevant for other models eventually.
Most Mamba models actually share these two tensors, albeit implicitly.
* readme : add Mamba to supported models, and add recent API changes
* mamba : move state_seq and state_mask views outside layer loop
A few tensors were also missing `struct` in front of `ggml_tensor`.
* fix(gguf-py): special tokens are no longer skipped when add_<token>_token is set to false
* fix(gguf-py): added missing cls and mask token ids to the gguf metadata
* batched embedding: pool outputs by sequence id. updated embedding example
* bring back non-causal attention
* embd : minor improvements
* llama : minor
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* BERT model graph construction (build_bert)
* WordPiece tokenizer (llm_tokenize_wpm)
* Add flag for non-causal attention models
* Allow for models that only output embeddings
* Support conversion of BERT models to GGUF
* Based on prior work by @xyzhang626 and @skeskinen
---------
Co-authored-by: Jared Van Bortel <jared@nomic.ai>
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add n_key_dim and n_value_dim
Some models use values that are not derived from `n_embd`.
Also remove `n_embd_head` and `n_embd_gqa` because it is not clear
which "head" is referred to (key or value).
Fix issue #4648.
* Fix `llm_build_kqv` to use `n_value_gqa`
* Rebase
* Rename variables
* Fix llm_build_kqv to be more generic wrt n_embd_head_k
* Update default values for n_embd_head_k and n_embd_head_v
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Fix llm_load_tensors: the asserts were not backcompat
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* convert : support Mixtral as LLAMA arch
* convert : fix n_ff typo
* llama : model loading
* ggml : sync latest ggml_mul_mat_id
* llama : update graph to support MoE
* llama : fix cur -> cur_expert
* llama : first working version
* llama : fix expert weighting in the FFN
* ggml : ggml_get_rows support 2D indexing [n_tokens, n_experts] (cpu only)
* ggml : add n_as argument to ggml_mul_mat_id
* ggml : fix ggml_get_rows to take into account ne02 / ne11
* metal : add more general support for ggml_get_rows + tests
* llama : add basic support for offloading moe with CUDA
* metal : add/mul/div use general kernel when src1 not cont
* metal : reduce the kernel launches for ggml_mul_mat_id
* ggml : get_rows : support non-contiguos tensors with gaps, generalize up to 3D
* ggml : update get_rows f16 and q
* cuda : support non-contiguous src1 in get_rows
* llama : offload missing ffn_moe_silu
* metal : fix ggml_get_rows to work with non-cont src1
* metal : add indirect mat-vec kernels for all quantization types
* llama : do not quantize expert gating tensors
* llama : add n_expert and n_expert_used to hparams + change quants
* test-backend-ops : add moe test
* cuda : fix get_rows when ncols is odd
* convert : determine n_ctx correctly
* metal : fix ggml_mul_mat_id for F32
* test-backend-ops : make experts more evenly probable (test_moe)
* test-backend-ops : cleanup, add moe test for batches
* test-backend-ops : add cpy from f32 -> all types test
* test-backend-ops : fix dequantize block offset
* llama : fix hard-coded number of experts
* test-backend-ops : simplify and disable slow tests to avoid CI timeout
* test-backend-ops : disable MOE test with thread sanitizer
* cuda : fix mul_mat_id with multi gpu
* convert : use 1e6 rope_freq_base for mixtral
* convert : fix style
* convert : support safetensors format
* gguf-py : bump version
* metal : add cpy f16 -> f32 kernel
* metal : fix binary ops for ne10 % 4 != 0
* test-backend-ops : add one more sum_rows test
* ggml : do not use BLAS with ggml_mul_mat_id
* convert-hf : support for mixtral-instruct (#4428)
* convert : typo fix, add additional hyperparameters, use LLaMA arch for Mixtral-instruct
* convert : use sentencepiece tokenizer for Mixtral-instruct
* convert : make flake8 happy
* metal : fix soft_max kernels
ref: 1914017863
* metal : limit kernels to not use more than the allowed threads
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Radek Pilar <github@mrkva.eu>
Disabled rules:
* E203 Whitespace before ':' - disabled because we often use 'C' Style where values are aligned
* E211 Whitespace before '(' (E211) - disabled because we often use 'C' Style where values are aligned
* E221 Multiple spaces before operator - disabled because we often use 'C' Style where values are aligned
* E225 Missing whitespace around operator - disabled because it's broken so often it seems like a standard
* E231 Missing whitespace after ',', ';', or ':' - disabled because we often use 'C' Style where values are aligned
* E241 Multiple spaces after ',' - disabled because we often use 'C' Style where values are aligned
* E251 Unexpected spaces around keyword / parameter equals - disabled because it's broken so often it seems like a standard
* E261 At least two spaces before inline comment - disabled because it's broken so often it seems like a standard
* E266 Too many leading '#' for block comment - sometimes used as "section" separator
* E501 Line too long - disabled because it's broken so often it seems like a standard
* E701 Multiple statements on one line (colon) - broken only in convert.py when defining abstract methods (we can use# noqa instead)
* E704 Multiple statements on one line - broken only in convert.py when defining abstract methods (we can use# noqa instead)
* gguf-py: Refactor and add file reading support
* Replay changes from #3871
Credit to @cebtenzzre for that pull
* Various type annotation fixes.
* sort imports with isort (again)
* Fix missing return statement in add_tensor
* style cleanup with flake8
* fix NamedTuple and Enum usage
* Fix an issue with state init in GGUFReader
Move examples to an examples/ directory
Clean up examples
Add an example of modifying keys in a GGUF file
Update documentation with info on examples
Try to support people importing gguf/gguf.py directly
* Damagage is not a word.
* Clean up gguf-py/examples/modify_gguf.py whitespace
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
* Update gguf-py/examples/modify_gguf.py formatting
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
* Update gguf-py/gguf/gguf_reader.py type hint
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
* Make examples executable, formatting changes
* Add more information to GGUFReader and examples comments
* Include a gguf Python package version bump
* Add convert-gguf-endian.py script
* cleanup
* gguf-py : bump minor version
* Reorganize scripts
* Make GGUFReader endian detection less arbitrary
* Add JSON dumping support to gguf-dump.py
Which I kind of regret now
* A few for gguf-dump.py cleanups
* Murder accidental tuple in gguf-py/scripts/gguf-dump.py
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
* cleanup
* constants : remove unneeded type annotations
* fix python 3.8 compat
* Set up gguf- scripts in pyproject.toml
* And include scripts/__init__.py, derp
* convert.py: We can't currently support Q8_0 on big endian.
* gguf-py: SpecialVocab: Always try available sources for special token ids
gguf-py: SpecialVocab: Try to load merges from merges.txt if not in tokenizer.json
gguf-py: SpecialVocab: Add 'add_bos_token' type bools to GGUF metadata
u
* cleanup
* Promote add_X_token to GGUF metadata for BOS and EOS
---------
Co-authored-by: Jared Van Bortel <jared@nomic.ai>
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
