Commit Graph

77 Commits

Author SHA1 Message Date
Ren Xuancheng
229ffff872
llama : add BPE pre-tokenization for Qwen2 (#7114)
* Add BPE pre-tokenization for Qwen2.

* minor : fixes

---------

Co-authored-by: Ren Xuancheng <17811943+jklj077@users.noreply.github.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-05-08 15:06:43 +03:00
DAN™
4cd621c26d
convert : add BPE pre-tokenization for DBRX (#7132)
* Add BPE pre-tokenization for DBRX.

* Add vocab GGUFs.

* Remove test.

* Remove GGUFs.
2024-05-08 13:43:23 +03:00
nopperl
b6aa670203
Fix OLMo HF to GGUF conversion (#6910) 2024-05-07 21:39:43 +02:00
DAN™
889bdd7686
command-r : add BPE pre-tokenization (#7063)
* Add BPE pre-tokenization for Command-R/R+.

* Bump transformers convert requirement.

* command-r : add individual digits regex

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-05-05 08:19:30 +03:00
Georgi Gerganov
92139b90af
tests : add test-tokenizer-0.sh + fix some tokenizers (#7036)
* tests : add test-tokenizer-0.sh

* unicode : add all unicode number ranges

* starcoder : fix pre-tokenizer

* tests : add test that fails with DeepSeek tokenizers

* falcon : fix regex

* unicode : regenerate unicode tables

* refact : add tokenizer model

* lint : fix

* tests : disable failing tests

ggml-ci

* refact : add tests files

ggml-ci

* convert : print -> logging

ggml-ci

* lint : fix

* unicode : digit -> number

* phi-3 : update
2024-05-04 08:32:32 +03:00
Brian
a2ac89d6ef
convert.py : add python logging instead of print() (#6511)
* 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
2024-05-03 22:36:41 +03:00
Bartowski
60325fa56f
Remove .attention from skipped tensors to match more accurately (#7051) 2024-05-03 01:49:09 +02:00
Georgi Gerganov
952d03dbea
convert : use utf8 encoding (#7000)
* convert : use utf8 encoding

* convert : update instructions and warning message
2024-04-30 11:05:25 +03:00
Georgi Gerganov
f4ab2a4147
llama : fix BPE pre-tokenization (#6920)
* merged the changes from deepseeker models to main branch

* Moved regex patterns to unicode.cpp and updated unicode.h

* Moved header files

* Resolved issues

* added and refactored unicode_regex_split and related functions

* Updated/merged the deepseek coder pr

* Refactored code

* Adding unicode regex mappings

* Adding unicode regex function

* Added needed functionality, testing remains

* Fixed issues

* Fixed issue with gpt2 regex custom preprocessor

* unicode : fix? unicode_wstring_to_utf8

* lint : fix whitespaces

* tests : add tokenizer tests for numbers

* unicode : remove redundant headers

* tests : remove and rename tokenizer test scripts

* tests : add sample usage

* gguf-py : reader prints warnings on duplicate keys

* llama : towards llama3 tokenization support (wip)

* unicode : shot in the dark to fix tests on Windows

* unicode : first try custom implementations

* convert : add "tokenizer.ggml.pre" GGUF KV (wip)

* llama : use new pre-tokenizer type

* convert : fix pre-tokenizer type writing

* lint : fix

* make : add test-tokenizer-0-llama-v3

* wip

* models : add llama v3 vocab file

* llama : adapt punctuation regex + add llama 3 regex

* minor

* unicode : set bomb

* unicode : set bomb

* unicode : always use std::wregex

* unicode : support \p{N}, \p{L} and \p{P} natively

* unicode : try fix windows

* unicode : category support via std::regex

* unicode : clean-up

* unicode : simplify

* convert : add convert-hf-to-gguf-update.py

ggml-ci

* lint : update

* convert : add falcon

ggml-ci

* unicode : normalize signatures

* lint : fix

* lint : fix

* convert : remove unused functions

* convert : add comments

* convert : exercise contractions

ggml-ci

* lint : fix

* cmake : refactor test targets

* tests : refactor vocab tests

ggml-ci

* tests : add more vocabs and tests

ggml-ci

* unicode : cleanup

* scripts : ignore new update script in check-requirements.sh

* models : add phi-3, mpt, gpt-2, starcoder

* tests : disable obsolete

ggml-ci

* tests : use faster bpe test

ggml-ci

* llama : more prominent warning for old BPE models

* tests : disable test-tokenizer-1-bpe due to slowness

ggml-ci

---------

Co-authored-by: Jaggzh <jaggz.h@gmail.com>
Co-authored-by: Kazim Abrar Mahi <kazimabrarmahi135@gmail.com>
2024-04-29 16:58:41 +03:00
Christian Zhou-Zheng
3055a41805
convert : fix conversion of some BERT embedding models (#6937) 2024-04-29 16:34:41 +03:00
Junyang Lin
3fec68be4e
convert : add support of codeqwen due to tokenizer (#6707)
* add support of codeqwen due to tokenizer

* override load_hparams

* fix typo

* fix load_params

* convert : fix whitespace

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-24 10:16:21 +03:00
liuwei-git
c8297c6af5
llama : add phi3 support (#6852)
* add explicit phi3 support

* add explicit phi3 support

* remove unused code

* convert : add BOS token

* llama : match EOT token <|end|>

* llama : minor / style

* llama : tabs -> spaces

* convert : fix lint checks

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-24 10:00:37 +03:00
Pedro Cuenca
b97bc3966e
llama : support Llama 3 HF conversion (#6745)
* Support Llama 3 conversion

The tokenizer is BPE.

* style

* Accept suggestion

Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>

* llama : add llama_token_is_eog()

ggml-ci

* llama : auto-detect more EOT tokens when missing in KV data

* convert : replacing EOS token is a hack

* llama : fix codegemma EOT token + add TODOs

* llama : fix model type string for 8B model

---------

Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-21 14:50:41 +03:00
nopperl
9958c81b79
Implement the OLMo architecture (#6741)
* implement olmo architecture

* remove unused variable

* remove unused moe branch

* remove check for weight

* remove superfluous moe, bias and rope tensors

* clarified comment

* fix clamp_kqv setting

* remove obsolete parameter name filter
2024-04-19 11:35:54 +02:00
Zheng.Deng
facb8b56f8
convert : fix autoawq gemma (#6704)
* fix autoawq quantized gemma model convert error

using autoawq to quantize gemma model will include a lm_head.weight tensor in model-00001-of-00002.safetensors. it result in this situation that convert-hf-to-gguf.py can't map lm_head.weight. skip loading this tensor could prevent this error.

* change code to full string match and print necessary message

change code to full string match and print a short message to inform users that lm_head.weight has been skipped.

---------

Co-authored-by: Zheng.Deng <32841220+CUGfred@users.noreply.github.com>
2024-04-16 23:51:07 +03:00
Ashish
dbceec87c0
llama : add StableLM2 12B (#6635)
* StableLM2 12B support for huggingface -> GGUF

* StableLM12 tensormapping and constants

* StableLM-2-12b model support

* fix

* Added 12B support

* Removed autoformatting; resolved bug where model_arch was not selecting StableLM2

* Formatting

* Do QK norm stacking in model conversion step

* Converge StableLM and StableLM2 code to simplify graph construction

* Fix accidental removal

* Removed warnings

* Revert formatter

* Move QK norm stack to private function so it's easier to read

* refactor stablelm graph builder to support 1.6, 3b and 12b more efficiently

* Proper check for None type for new_name to avoid crash; formatting; revert change to base class `write_tensors()`

* Format

* Formatting

* format

Co-authored-by: compilade <git@compilade.net>

* Fix incorrect check for K norm

* space after commas; Keep indentation multiple of 4 spaces

* Flake8 format

* Removed unnecessary conditional branches

* Removed unused comment

* Fixed incorrect tensor passing

* Format

---------

Co-authored-by: compilade <git@compilade.net>
2024-04-16 18:48:35 +03:00
Shijie
f4dea7da18
llama : add qwen2moe (#6074)
* support qwen2moe

* fix-review

* metal : support unary ops for nelements % 4 != 0

* metal : require contiguousness for float4 unary kernels

* metal : require contiguousness for float4 unary kernels (cont)

* fix-review

* names : for brevity "SHARED_EXP" -> "SHEXP"

* llama : reuse build_moe_ffn()

* llama : add model type name

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-16 18:40:48 +03:00
Daniel Bevenius
4fbd8098e6
gguf : add special tokens metadata for FIM/Infill (#6689)
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>
2024-04-16 09:13:13 +03:00
James A Capozzoli
a4ec34e1cd
convert : enable the --use-temp-file cli flag (#6645) 2024-04-14 11:40:18 +03:00
Pierrick Hymbert
4bd0f93e4a
model: support arch DbrxForCausalLM (#6515)
* model: dbrx convert to gguf
#6344

* llama: support dbrx
#6344

* doc: dbrx: add the model as supported

* scripts: get-wikitext-2 add unzip

* llama: increase maximum experts allowed

* llama: factorize moe graph implementation between grok, mixtral and dbrx


---------

Co-authored-by: Megha Agarwal <16129366+megha95@users.noreply.github.com>
2024-04-13 11:33:52 +02:00
Jared Van Bortel
1b67731e18
BERT tokenizer fixes (#6498)
Key changes:
* BERT conversion: fix abuse of LlamaHfVocab, do not set BOS or EOS
* Nomic Embed conversion: pad vocab instead of slicing embedding tensor
* llama_tokenize: handle added special tokens like HF does
2024-04-09 13:44:08 -04:00
Carolinabanana
5dc9dd7152
llama : add Command R Plus support (#6491)
* Add Command R Plus GGUF

* Add Command R Plus GGUF

* Loading works up to LayerNorm2D

* Export new tensors in 1D so they are not quantized.

* Fix embedding layer based on Noeda's example

* Whitespace

* Add line

* Fix unexpected tokens on MPS. Re-add F16 fix. ((Noeda)

* dranger003: Fix block index overflow in CUDA dequantizing.

* Reverted blocked multiplication code as it still has issues and could affect other Llama arches

* export norms as f32

* fix overflow issues during quant and other cleanup

* Type convention

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* dranger003: Fix more int overflow during quant.

---------

Co-authored-by: S <seast@Ss-Mac-Studio.local>
Co-authored-by: S <s@example.com>
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-09 11:16:13 +03:00
Clint Herron
72d73af651
convert : fix for lint error complaining of bare except (#6470) 2024-04-04 09:32:53 +03:00
bryanSwk
bb43cf7e9d
llama : add SEA-LION support (#6448)
* initial commit for sealion support

* add sealion support

* minor fix

* q/k ln and pos_embd only if required

* Apply suggestions from code review

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* minor : clear whitespaces

---------

Co-authored-by: bryan <bryansiow@aisingapore.org>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-03 21:05:10 +03:00
Abhishek Gopinath K
db214fa578
Missing tokenizer.model error during gguf conversion (#6443)
Co-authored-by: Jared Van Bortel <jared@nomic.ai>
2024-04-03 11:42:52 -04:00
slaren
08a0c02060
ggml : mul_mat_id use the same tensor for all the experts (#6387)
* ggml : update mul_mat_id to use the same tensor for all the experts

* update cuda

* minor

* update metal

* update test-backend-ops

* fix cuda

* Update ggml-metal.m

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* update convert.py

* update convert-hf-to-gguf.py

* update convert.py for mixtral hf models

* Update convert-hf-to-gguf.py

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* cuda : support non-pow-2 number of experts

* allow quantize to work for split and merged experts models in the same way

* cleanup + disable mmap automatically with split tensors models

* update imatrix

* test-backend-ops : test qwen argsort

* update grok model loading

* llama : add merged experts tensors to the grok tensor map

* minor

* gguf : bump version

* fix quantizing of merged experts

* convert-hf-to-gguf.py : update grok (untested)

* make linter happy

* cuda/argsort : use shared memory instead of pool memory

* convert : fix grok tensor names

* metal : add support for non-pow-2 argsort

* llama : more loader cleanup, better error checking

* cuda : fix warning

* llama : still use mmap for loading old models, but copy the data to a host buffer

* add review note

* llama : remove ffn tensor counting + add sanity check

ggml-ci

* convert : fix handling of n_experts == None

ggml-ci

* imatrix : fix ncall counters

* llama : produce error if imatrix size does not match

* quantize : terminate on errors + trace logs

ggml-ci

* metal : pad shared memory to 16 bytes

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-03 16:07:05 +03:00
hxer7963
069574775c
[Model] Add support for xverse (#6301)
* Support xverse model convert to gguf format.

* 1. Convert xverse models to gguf;
2. Add LLM_ARCH_XVERSE inference in llama.cpp;
3. Add xverse item in Supported models in README.md;

* * gguf-py: remove redundant logs
* llama: remove the init_mapping_prefetch custom parameter

* llama.cpp: Include the changes from #6122 to exclude the unused outputs of the last layers.

* - Fix format issues
- Remove duplicate set kqv_out to llm_build_kv

* Update llama.cpp

---------

Co-authored-by: willhe <willhe@xverse.cn>
Co-authored-by: willhe <hexin@xverse.cn>
2024-03-29 14:37:03 +01:00
Pedro Cuenca
b75c38166c
convert : allow conversion of Mistral HF models (#6144)
* Allow conversion of Mistral HF models

* Homogenize Llama, Mistral, Mixtral under the same entry.

* Fix tokenizer, permute tensors

* Use sentencepiece tokenizer, or fall back to hfft.

* convert-hf : small fix for mypy

* convert-hf : fix duplicated block_count

* convert-hf : add vocab size to metadata

---------

Co-authored-by: Jared Van Bortel <jared@nomic.ai>
2024-03-29 09:15:00 +02:00
Jared Van Bortel
be55134a53
convert : refactor vocab selection logic (#6355) 2024-03-28 11:44:36 -04:00
Pedro Cuenca
e097633f63
convert-hf : fix exception in sentencepiece with added tokens (#6320) 2024-03-26 14:32:19 +02:00
Julius Arkenberg
476b0251b2
llama : add grok-1 support (#6204)
* Add support for Grok model architecture

* Revert convert-hf-to-gguf to default options

* Fixed f_norm_rms_eps bug

* Fix whitespaces

* llama : fix grok rope type

* llama : minor

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-23 18:41:53 +02:00
Thérence
9b03719ad7
convert : add support for CamembertModel architecture (#6119)
Adding support for CamembertModel architecture used by :
https://huggingface.co/dangvantuan/sentence-camembert-large
2024-03-18 10:17:00 +02:00
Andrew Canis
12247f4c69
llama : add Command-R support (#6033)
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
2024-03-15 22:41:22 +02:00
compilade
c2101a2e90
llama : support Mamba Selective State Space Models (#5328)
* 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`.
2024-03-08 17:31:00 -05:00
Georgi Gerganov
e0843afe1b
flake : fix 2024-03-04 21:50:50 +02:00
Douglas Hanley
475df1d6cf
llama : allow for user specified embedding pooling type (#5849)
* allow for user specified pooling type

* llama : use enum types over int

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-03 12:40:27 +02:00
Jared Van Bortel
c7a0ad8ec9
convert-hf : make model class definitions self-contained (#5825) 2024-03-02 12:21:47 -05:00
Sourab Mangrulkar
c29af7e225
llama : add StarCoder2 support (#5795)
* Add support for starcoder2

* handle rope type

* skip rope freq and rotary embeddings from being serialized

* resolve comments

* Update llama.cpp

* remove redundant changes

* handle `rope-theta`

* llama : change starcoder2 rope type

* address comment

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-01 21:30:46 +02:00
kunal-vaishnavi
e743386728
gemma : fix bfloat16 -> float16 conversion issue (#5810) 2024-03-01 16:08:08 +02:00
Anas Ahouzi
69917dfa55
py : fix StableLM conversion after config.json changes (#5703)
* Fix issues during StableLM models conversion

* Fix hard coded layer_norm_eps

* Support layer_norm_eps for LlavaStableLM

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>

* Add missing parenthesis

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>

* Support rotary_factor for LlavaStableLM

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>

* fix typo

* Add StableLMEpochForCausalLM for safety

Co-authored-by: compilade <113953597+compilade@users.noreply.github.com>

* Add StableLMEpochForCausalLM for safety 2

Co-authored-by: compilade <113953597+compilade@users.noreply.github.com>

---------

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
Co-authored-by: Jared Van Bortel <jared@nomic.ai>
Co-authored-by: compilade <113953597+compilade@users.noreply.github.com>
2024-02-25 11:54:04 +02:00
Jared Van Bortel
54fbcd2ce6
convert : fix missing ftype for gemma (#5690) 2024-02-23 20:39:14 +02:00
Jared Van Bortel
15499eb942
mpt : do not duplicate token_embd.weight on disk (#5670) 2024-02-22 17:05:23 -05:00
Georgi Gerganov
847eedbdb2
py : add Gemma conversion from HF models (#5647)
* py : add gemma conversion from HF models

* Update convert-hf-to-gguf.py

Co-authored-by: Aarni Koskela <akx@iki.fi>

* Update convert-hf-to-gguf.py

Co-authored-by: Aarni Koskela <akx@iki.fi>

* Update convert-hf-to-gguf.py

Co-authored-by: Jared Van Bortel <jared@nomic.ai>

---------

Co-authored-by: Aarni Koskela <akx@iki.fi>
Co-authored-by: Jared Van Bortel <jared@nomic.ai>
2024-02-22 23:22:48 +02:00
Georgi Gerganov
5a9e2f60ba
py : minor fixes (#5668) 2024-02-22 20:13:25 +02:00
Douglas Hanley
4524290e87
Use correct type of pooling for embedding models (#5500)
Use correct type of pooling for embedding models
2024-02-15 12:21:49 -05:00
Jared Van Bortel
ea9c8e1143
llama : add support for Nomic Embed (#5468) 2024-02-13 12:03:53 -05:00
Douglas Hanley
03bf161eb6
llama : support batched embeddings (#5466)
* 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>
2024-02-13 14:06:58 +02:00
Douglas Hanley
2891c8aa9a
Add support for BERT embedding models (#5423)
* 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>
2024-02-11 11:21:38 -05:00
runfuture
4aa43fab56
llama : fix MiniCPM (#5392)
* fix bug for norm_rms_eps missing

* to align with the same order as convert.py for model write

* fix: undo HF models permute tensor

* update for flake8 lint
2024-02-08 12:36:19 +02:00
runfuture
316c7faf77
llama : add MiniCPM support (#5346)
* support minicpm arch.

* fix tab/space typo.

* convert minicpm model via convert-hf-gguf.py

* try to make tokenizer work

* fix bug for quantize minicpm

* fix for flake8 lint

* remove convert-minicpm.py

* fix for editorconfig

* correct minicpm model type (size)

* constants expanded for minicpm

* Minor change of the constant names for minicpm
2024-02-07 08:15:56 +02:00