Commit Graph

268 Commits

Author SHA1 Message Date
Pierrick Hymbert
24ee66ed0d
server : coherent log output for KV cache full (#6637) 2024-04-12 14:49:21 +03:00
Ralph Soika
b3a96f27f0
minor layout improvements (#6572)
* minor layout improvements

* added missing file, run deps.sh locally
2024-04-10 19:18:25 +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
Ed Lee
400d5d722d
server : detect search query to start webchat (#6554) 2024-04-09 10:31:47 +02:00
Jan Boon
beea6e1b16
llama : save and restore kv cache for single seq id (#6341)
* llama : save and restore kv cache for single seq id

* remove trailing whitespace

* respond error in case there's no space in the kv cache

* add kv seq save restore to test case

* add --slot-save-path arg to enable save restore and restrict save location

* Returning 0 for some cases, instead of asserting.

* cleanup error cases

* rename sequence state functions

* rename state get set functions

* add previous function names back in with DEPRECATED notice

* update doc

* adjust endpoints to preferred style

* fix restoring zero cell count

* handle seq rm return value

* unused param

* keep in the size check

* fix return types

* add server test case for slot save restore

* cleanup

* add cake

* cleanup style

* add special

* removing a whole sequence never fails

* move sequence state file functionality from server to llama to match session api and add version tags

* catch exceptions on save as well

* error log messages

* check types for stricter restore

* update server doc

* readme : update API changes date

* strict filename validation

* move include, reject bom as well

* also reject empty filename

* reject whitespace and trailing dot

---------

Co-authored-by: Martin Evans <martindevans@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-08 15:43:30 +03:00
Pierrick Hymbert
75cd4c7729
ci: bench: support sse and fix prompt processing time / server: add tokens usage in stream OAI response (#6495)
* ci: bench: support sse and fix prompt processing time
server: add tokens usage in stream mode

* ci: bench: README.md EOL

* ci: bench: remove total pp and tg as it is not accurate

* ci: bench: fix case when there is no token generated

* ci: bench: change to the 95 percentile for pp and tg as it is closer to what the server exports in metrics

* ci: bench: fix finish reason rate
2024-04-06 05:40:47 +02:00
Shakhar Dasgupta
2e66913e5f
server: allow penalizing repetition of newlines on server webpage (#6431) 2024-04-04 17:03:00 +02:00
Pierrick Hymbert
7a2c92637a
ci: bench: add more ftype, fix triggers and bot comment (#6466)
* ci: bench: change trigger path to not spawn on each PR

* ci: bench: add more file type for phi-2: q8_0 and f16.
- do not show the comment by default

* ci: bench: add seed parameter in k6 script

* ci: bench: artefact name perf job

* Add iteration in the commit status, reduce again the autocomment

* ci: bench: add per slot metric in the commit status

* Fix trailing spaces
2024-04-04 12:57:58 +03:00
Georgi Gerganov
4399f13fb9
server : remove obsolete --memory-f32 option 2024-04-04 09:34:58 +03:00
Xiao-Yong Jin
1a43c7254e
server : add option to disable KV offload (#6468) 2024-04-04 09:33:48 +03:00
Fattire
5fb1574c81
A few small fixes to server's README docs (#6428)
* Typo fix to server's README.md

Fix minor typo ("tonen") in server README.

* server readme grammar/style fixes.

Quickly went through this file to look for inconsistencies in
presentation of defaults, flag options, and looked for typos
and grammar issues.

Not perfect, but hopefully improved.

* Update README.md

Remove an extra space before newline.
2024-04-03 22:22:57 +02:00
JH23X
60cdf40cc3
server : handle exception on wrong type in request (#6452)
Co-authored-by: Jonas Holzner <jonas.holzner.external@hensoldt.net>
2024-04-03 21:09:52 +03:00
Eric Zhang
6902cb7f2e
server : stop gracefully on SIGTERM (#6348) 2024-03-28 09:50:48 +01:00
Pierrick Hymbert
a016026a3a
server: continuous performance monitoring and PR comment (#6283)
* server: bench: init

* server: bench: reduce list of GPU nodes

* server: bench: fix graph, fix output artifact

* ci: bench: add mermaid in case of image cannot be uploaded

* ci: bench: more resilient, more metrics

* ci: bench: trigger build

* ci: bench: fix duration

* ci: bench: fix typo

* ci: bench: fix mermaid values, markdown generated

* typo on the step name

Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>

* ci: bench: trailing spaces

* ci: bench: move images in a details section

* ci: bench: reduce bullet point size

---------

Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
2024-03-27 20:26:49 +01:00
Eric Zhang
0642b22cd1
server: public: use relative routes for static files (#6325)
server: public: support custom `api_url`, default to relative base path
2024-03-27 06:55:29 +01:00
compilade
557410b8f0
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage

* llama : more compact state saving and reloading

* llama : fix lctx.n_outputs not being set before building graph

* perplexity : adapt to the logits API changes

* perplexity : fix Winogrande, use correct logits for second choice start

The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.

The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.

This is simpler now, and the outlier scores aren't there anymore.

* perplexity : normalize spaces and punctuation in Winogrande sentences

* llama : fix embedding conditions

* llama : fix llama_get_embeddings_ith when the resulting id is 0

* llama : fix wrong n_outputs in llama_set_inputs

A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.

* llama : when saving the state, recalculate n_outputs

This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.

* llama : fix not-skipping outputs of non-causal models

* llama : fix running a batch with n_outputs == 0

It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.

* llama : keep same graph topology even when n_outputs == 0

* ggml : saner ggml_can_repeat with empty tensors

*  ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1

* ggml : do not multi-thread ops returning empty tensors

* ggml : make ggml_is_empty public and work with views

* llama : use a vector for ctx->output_ids

* llama : rework reallocation logic for llama_output_reserve

Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.

* ggml : skip empty tensors in all backends

* llama : fix llama_output_reserve nullptr deref when new_size is 0

* perplexity : make Winogrande work as it does on master

The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.

* llama : clearer error messages for invalid logits or embeddings ids

* llama : assert all models that can have inp_out_ids

Since the graph topology is now constant, this presence check
can be done even when there are no outputs.

* llama : assert logits and embd buffers exist before writing to them

* llama : handle errors from llama_output_reserve at call sites

* perplexity : make hellaswag and multiple-choice outputs identical to master

Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.

This will probably be changed back in the future to make these benchmarks
a tiny bit faster.

* perplexity : fix division by zero when using less than 100 multiple-choice tasks

* llama : allow loading state saved with a different ctx size

When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.

Doing this enables the use-case of extending or shrinking the context size
of a saved session.

This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.

* llama : minor

ggml-ci

* readme : update recent API changes, and warn about Vulkan

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 16:46:41 +02:00
Jan Boon
3d032ece8e
server : add n_discard parameter (#6300) 2024-03-26 10:47:43 +02:00
slaren
280345968d
cuda : rename build flag to LLAMA_CUDA (#6299) 2024-03-26 01:16:01 +01:00
Xuan Son Nguyen
ad3a0505e3
Server: clean up OAI params parsing function (#6284)
* server: clean up oai parsing function

* fix response_format

* fix empty response_format

* minor fixes

* add TODO for logprobs

* update docs
2024-03-25 09:42:17 +01:00
Pierrick Hymbert
f482bb2e49
common: llama_load_model_from_url split support (#6192)
* llama: llama_split_prefix fix strncpy does not include string termination
common: llama_load_model_from_url:
 - fix header name case sensitive
 - support downloading additional split in parallel
 - hide password in url

* common: EOL EOF

* common: remove redundant LLAMA_CURL_MAX_PATH_LENGTH definition

* common: change max url max length

* common: minor comment

* server: support HF URL options

* llama: llama_model_loader fix log

* common: use a constant for max url length

* common: clean up curl if file cannot be loaded in gguf

* server: tests: add split tests, and HF options params

* common: move llama_download_hide_password_in_url inside llama_download_file as a lambda

* server: tests: enable back Release test on PR

* spacing

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

* spacing

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

* spacing

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

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-23 18:07:00 +01:00
Pierrick Hymbert
1997577d5e
server: docs: --threads and --threads, --ubatch-size, --log-disable (#6254) 2024-03-23 18:00:38 +01:00
Pierrick Hymbert
1b26aebe4d
server: flush stdout after logging in both text and json layout (#6253) 2024-03-23 13:18:45 +01:00
Olivier Chafik
72114edf06
json-schema-to-grammar : fix order of props + non-str const/enum (#6232)
* json: ordered json in server/schema converter to respect orig order

* json: ws nits

* json: support non-string const / enums
2024-03-22 15:07:44 +02:00
Jan Boon
6b8bb3a31d
server : fix n_keep always showing as 0 in response (#6211) 2024-03-22 13:12:05 +02:00
Georgi Gerganov
68e210b354
server : enable continuous batching by default (#6231) 2024-03-22 13:08:28 +02:00
Jan Boon
be07a03217
server : update readme doc from slot_id to id_slot (#6213) 2024-03-21 23:41:24 +01:00
Olivier Chafik
5b7b0ac8df
json-schema-to-grammar improvements (+ added to server) (#5978)
* json: fix arrays (disallow `[,1]`)

* json: support tuple types (`[number, string]`)

* json: support additionalProperties (`{[k: string]: [string,number][]}`)

* json: support required / optional properties

* json: add support for pattern

* json: resolve $ref (and support https schema urls)

* json: fix $ref resolution

* join: support union types (mostly for nullable types I think)

* json: support allOf + nested anyOf

* json: support any (`{}` or `{type: object}`)

* json: fix merge

* json: temp fix for escapes

* json: spaces in output and unrestricted output spaces

* json: add typings

* json:fix typo

* Create ts-type-to-grammar.sh

* json: fix _format_literal (json.dumps already escapes quotes)

* json: merge lit sequences and handle negatives

{"type": "string", "pattern": "^({\"question\": \"[^\"]+\", \"response\": \"[^\"]+\"}\\n)+$"}

* json: handle pattern repetitions

* Update json-schema-to-grammar.mjs

* Create regex-to-grammar.py

* json: extract repeated regexp patterns to subrule

* Update json-schema-to-grammar.py

* Update json-schema-to-grammar.py

* Update json-schema-to-grammar.py

* json: handle schema from pydantic Optional fields

* Update json-schema-to-grammar.py

* Update json-schema-to-grammar.py

* Update ts-type-to-grammar.sh

* Update ts-type-to-grammar.sh

* json: simplify nullable fields handling

* json: accept duplicate identical rules

* json: revert space to 1 at most

* json: reuse regexp pattern subrules

* json: handle uuid string format

* json: fix literal escapes

* json: add --allow-fetch

* json: simplify range escapes

* json: support negative ranges in patterns

* Delete commit.txt

* json: custom regex parser, adds dot support & JS-portable

* json: rm trailing spaces

* Update json-schema-to-grammar.mjs

* json: updated server & chat `( cd examples/server && ./deps.sh )`

* json: port fixes from mjs to python

* Update ts-type-to-grammar.sh

* json: support prefixItems alongside array items

* json: add date format + fix uuid

* json: add date, time, date-time formats

* json: preserve order of props from TS defs

* json: port schema converter to C++, wire in ./server

* json: nits

* Update json-schema-to-grammar.cpp

* Update json-schema-to-grammar.cpp

* Update json-schema-to-grammar.cpp

* json: fix mjs implementation + align outputs

* Update json-schema-to-grammar.mjs.hpp

* json: test C++, JS & Python versions

* json: nits + regen deps

* json: cleanup test

* json: revert from c++17 to 11

* json: nit fixes

* json: dirty include for test

* json: fix zig build

* json: pass static command to std::system in tests (fixed temp files)

* json: fix top-level $refs

* json: don't use c++20 designated initializers

* nit

* json: basic support for reserved names `{number:{number:{root:number}}}`

* Revamp test cmake to allow args (WORKING_DIRECTORY needed for JSON test)

* json: re-ran server deps.sh

* json: simplify test

* json: support mix of additional props & required/optional

* json: add tests for some expected failures

* json: fix type=const in c++, add failure expectations for non-str const&enum

* json: test (& simplify output of) empty schema

* json: check parsing in test + fix value & string refs

* json: add server tests for OAI JSON response_format

* json: test/fix top-level anyOf

* json: improve grammar parsing failures

* json: test/fix additional props corner cases

* json: fix string patterns (was missing quotes)

* json: ws nit

* json: fix json handling in server when there's no response_format

* json: catch schema conversion errors in server

* json: don't complain about unknown format type in server if unset

* json: cleaner build of test

* json: create examples/json-schema-pydantic-example.py

* json: fix date pattern

* json: move json.hpp & json-schema-to-grammar.{cpp,h} to common

* json: indent 4 spaces

* json: fix naming of top-level c++ function (+ drop unused one)

* json: avoid using namespace std

* json: fix zig build

* Update server.feature

* json: iostream -> fprintf

* json: space before & refs for consistency

* json: nits
2024-03-21 11:50:43 +00:00
Xuan Son Nguyen
91f8ad167d
Server: version bump for httplib and json (#6169)
* server: version bump for httplib and json

* fix build

* bring back content_length
2024-03-20 13:30:36 +01:00
Georgi Gerganov
bc0baab2ea
server : allow to override -ngl in tests (#6170) 2024-03-20 14:14:32 +02:00
Karthick
47cc7a7bf9
Server: Handle n_keep parameter in the request (#6174) 2024-03-20 12:02:34 +01:00
Jared Van Bortel
bd60d82d0c
server tests : more pythonic process management; fix bare except: (#6146)
* server tests : remove seemingly redundant newlines in print()

* server tests : use built-in subprocess features, not os.kill and psutil

* server tests : do not catch e.g. SystemExit; use print_exc

* server tests: handle TimeoutExpired exception

* server tests: fix connect on dual-stack systems

* server: tests: add new tokens regex on windows generated following new repeat penalties default changed in (#6127)

* server: tests: remove the hack on windows since now we get the good socket family

* server: tests: add new tokens regex following new repeat penalties default changed in (#6127)

* server: tests: add new tokens regex following new repeat penalties default changed in (#6127)

---------

Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-03-20 06:33:49 +01:00
Pierrick Hymbert
d01b3c4c32
common: llama_load_model_from_url using --model-url (#6098)
* common: llama_load_model_from_url with libcurl dependency

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-17 19:12:37 +01:00
Pierrick Hymbert
43241adf22
server: disable debug release type sanitizer, simplify trigger (#6047)
- increase time out for server
 - do not fail fast
2024-03-14 13:15:39 +02:00
slaren
f30ea47a87
llama : add pipeline parallelism support (#6017)
* llama : add pipeline parallelism support for batch processing with multiple CUDA GPUs

ggml-ci

* server : add -ub, --ubatch-size parameter

* fix server embedding test

* llama : fix Mamba inference for pipeline parallelism

Tested to work correctly with both `main` and `parallel` examples.

* llama : limit max batch size to n_batch

* add LLAMA_SCHED_MAX_COPIES to configure the number of input copies for pipeline parallelism
default increase to 4 (from 2)

changing this value may improve performance for some systems, but increases memory usage

* fix hip build

* fix sycl build (disable cpy_tensor_async)

* fix hip build

* llama : limit n_batch and n_ubatch to n_ctx during context creation

* llama : fix norm backend

* batched-bench : sync after decode

* swiftui : sync after decode

* ggml : allow ggml_get_rows to use multiple threads if they are available

* check n_ubatch >= n_tokens with non-casual attention

* llama : do not limit n_batch to n_ctx with non-casual attn

* server : construct batch with size of llama_n_batch

* ggml_backend_cpu_graph_compute : fix return value when alloc fails

* llama : better n_batch and n_ubatch comment

* fix merge

* small fix

* reduce default n_batch to 2048

---------

Co-authored-by: Francis Couture-Harpin <git@compilade.net>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-13 18:54:21 +01:00
Xuan Son Nguyen
99b71c068f
Server: Use multi-task for embeddings endpoint (#6001)
* use multitask for embd endpoint

* specify types

* remove redundant {"n_predict", 0}
2024-03-13 11:39:11 +01:00
Jakub N
828defefb6
Update server docker image URLs (#5997) 2024-03-11 14:40:42 +01:00
Xuan Son Nguyen
caa106d4e0
Server: format error to json (#5961)
* server: format error to json

* server: do not crash on grammar error

* fix api key test case

* revert limit max n_predict

* small fix

* correct coding style

* update completion.js

* launch_slot_with_task

* update docs

* update_slots

* update webui

* update readme
2024-03-11 10:56:41 +01:00
Minsoo Cheong
332bdfd798
server : maintain chat completion id for streaming responses (#5988)
* server: maintain chat completion id for streaming responses

* Update examples/server/utils.hpp

* Update examples/server/utils.hpp

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-11 10:09:32 +02:00
Pierrick Hymbert
fa8a809a91
server: ci: windows build and tests (#5968)
* server: ci: windows build and tests

* server: ci: remove tmp push branch

* server: ci: EOF EOL

* Use builti

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

* server: tests: server graceful shutdown, then kill, then hard kill

* server: tests: remove python2 unicode string

* server: tests: remove wrong comment on server starting,  close_fds is always true

* server: tests: server kill, if pid exists

* server: tests: remove dependency to killall

* server: tests: ci windows: pid exists better handling

---------

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
2024-03-10 18:17:47 +01:00
Pierrick Hymbert
621e86b331
server: benchmark: chat/completions scenario and other llm servers comparison (#5941)
* server: bench: Init a bench scenario with K6
See #5827

* server: bench: EOL EOF

* server: bench: PR feedback and improved k6 script configuration

* server: bench: remove llamacpp_completions_tokens_seconds as it include prompt processing time and it's misleading

server: bench: add max_tokens from SERVER_BENCH_MAX_TOKENS

server: bench: increase truncated rate to 80% before failing

* server: bench: fix doc

* server: bench: change gauge custom metrics to trend

* server: bench: change gauge custom metrics to trend
server: bench: add trend custom metrics for total tokens per second average

* server: bench: doc add an option to debug http request

* server: bench: filter dataset too short and too long sequences

* server: bench: allow to filter out conversation in the dataset based on env variable

* server: bench: fix assistant message sent instead of user message

* server: bench: fix assistant message sent instead of user message

* server : add defrag thold parameter

* server: bench: select prompts based on the current iteration id not randomly to make the bench more reproducible

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-09 23:41:49 +01:00
Georgi Gerganov
77d1ac7e00
server : print chat template info 2024-03-09 22:04:00 +02:00
Georgi Gerganov
58308a0ecc
server : fix metrics init (#5964) 2024-03-09 17:34:15 +02:00
Georgi Gerganov
97c09585d6
server : clarify some items in the readme (#5957)
* server : clarify some items in the readme

* server : fix typo
2024-03-09 15:47:47 +02:00
SeungWon Jeong
fb215c3832
server : normalize embeddings (#5956)
* output normalize embedding in '/v1/embeddings'

* common : reuse llama_embd_normalize

* common : better normalize impl

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-09 14:27:58 +02:00
Alexey Parfenov
0db32beaf0
server : fix passing prompt as tokens (#5955)
* server: fix passing prompt as tokens

* Update examples/server/server.cpp

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-09 13:16:53 +02:00
Georgi Gerganov
9674aaf35c
server : simplify logic for empty prompts (#5953) 2024-03-09 12:34:18 +02:00
Xuan Son Nguyen
950ba1ab84
Server: reorganize some http logic (#5939)
* refactor static file handler

* use set_pre_routing_handler for validate_api_key

* merge embedding handlers

* correct http verb for endpoints

* fix embedding response

* fix test case CORS Options

* fix code style
2024-03-09 11:27:53 +01:00
Gabe Goodhart
e1fa9569ba
server : add SSL support (#5926)
* add cmake build toggle to enable ssl support in server

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* add flags for ssl key/cert files and use SSLServer if set

All SSL setup is hidden behind CPPHTTPLIB_OPENSSL_SUPPORT in the same
way that the base httlib hides the SSL support

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* Update readme for SSL support in server

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* Add LLAMA_SERVER_SSL variable setup to top-level Makefile

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

---------

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
2024-03-09 11:57:09 +02:00
Pierrick Hymbert
fd72d2d2a5
server: tests: add truncated prompt tests, better kv cache size (#5933)
* server: tests: add truncated prompt tests, better size

* server, tests : update regex

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-09 11:30:04 +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