baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
3724 commits 1 branch 0 tags 14 MiB
Commit graph

751 commits

Author SHA1 Message Date
Daniel Hiltgen
3518aaef33
Merge pull request #4218 from dhiltgen/auto_parallel 
Enable concurrency by default
 2024-07-01 08:32:29 -07:00
Michael Yang
123a722a6f
zip: prevent extracting files into parent dirs (#5314) 2024-06-26 21:38:21 -07:00
Blake Mizerany
cb42e607c5
llm: speed up gguf decoding by a lot (#5246) 
Previously, some costly things were causing the loading of GGUF files
and their metadata and tensor information to be VERY slow:

  * Too many allocations when decoding strings
  * Hitting disk for each read of each key and value, resulting in a
    not-okay amount of syscalls/disk I/O.

The show API is now down to 33ms from 800ms+ for llama3 on a macbook pro
m3.

This commit also prevents collecting large arrays of values when
decoding GGUFs (if desired). When such keys are encountered, their
values are null, and are encoded as such in JSON.

Also, this fixes a broken test that was not encoding valid GGUF.
 2024-06-24 21:47:52 -07:00
Daniel Hiltgen
642cee1342 Sort the ps output 
Provide consistent ordering for the ps command - longest duration listed first
 2024-06-21 15:59:41 -07:00
Daniel Hiltgen
9929751cc8 Disable concurrency for AMD + Windows 
Until ROCm v6.2 ships, we wont be able to get accurate free memory
reporting on windows, which makes automatic concurrency too risky.
Users can still opt-in but will need to pay attention to model sizes otherwise they may thrash/page VRAM or cause OOM crashes.
All other platforms and GPUs have accurate VRAM reporting wired
up now, so we can turn on concurrency by default.
 2024-06-21 15:45:05 -07:00
Daniel Hiltgen
17b7186cd7 Enable concurrency by default 
This adjusts our default settings to enable multiple models and parallel
requests to a single model.  Users can still override these by the same
env var settings as before.  Parallel has a direct impact on
num_ctx, which in turn can have a significant impact on small VRAM GPUs
so this change also refines the algorithm so that when parallel is not
explicitly set by the user, we try to find a reasonable default that fits
the model on their GPU(s).  As before, multiple models will only load
concurrently if they fully fit in VRAM.
 2024-06-21 15:45:05 -07:00
Michael Yang
e835ef1836 fix: quantization with template 2024-06-21 13:39:25 -07:00
royjhan
fedf71635e
Extend api/show and ollama show to return more model info (#4881) 
* API Show Extended

* Initial Draft of Information

Co-Authored-By: Patrick Devine <pdevine@sonic.net>

* Clean Up

* Descriptive arg error messages and other fixes

* Second Draft of Show with Projectors Included

* Remove Chat Template

* Touches

* Prevent wrapping from files

* Verbose functionality

* Docs

* Address Feedback

* Lint

* Resolve Conflicts

* Function Name

* Tests for api/show model info

* Show Test File

* Add Projector Test

* Clean routes

* Projector Check

* Move Show Test

* Touches

* Doc update

---------

Co-authored-by: Patrick Devine <pdevine@sonic.net>
 2024-06-19 14:19:02 -07:00
royjhan
89c79bec8c
Add ModifiedAt Field to /api/show (#5033) 
* Add Mod Time to Show

* Error Handling
 2024-06-15 20:53:56 -07:00
Daniel Hiltgen
45cacbaf05
Merge pull request #4517 from dhiltgen/gpu_incremental 
Enhanced GPU discovery and multi-gpu support with concurrency
 2024-06-14 15:35:00 -07:00
Daniel Hiltgen
6f351bf586 review comments and coverage 2024-06-14 14:55:50 -07:00
Daniel Hiltgen
ff4f0cbd1d Prevent multiple concurrent loads on the same gpus 
While models are loading, the VRAM metrics are dynamic, so try
to load on a GPU that doesn't have a model actively loading, or wait
to avoid races that lead to OOMs
 2024-06-14 14:51:40 -07:00
Daniel Hiltgen
fc37c192ae Refine CPU load behavior with system memory visibility 2024-06-14 14:51:40 -07:00
Daniel Hiltgen
434dfe30c5 Reintroduce nvidia nvml library for windows 
This library will give us the most reliable free VRAM reporting on windows
to enable concurrent model scheduling.
 2024-06-14 14:51:40 -07:00
Daniel Hiltgen
48702dd149 Harden unload for empty runners 2024-06-14 14:51:40 -07:00
Daniel Hiltgen
5e8ff556cb Support forced spreading for multi GPU 
Our default behavior today is to try to fit into a single GPU if possible.
Some users would prefer the old behavior of always spreading across
multiple GPUs even if the model can fit into one.  This exposes that
tunable behavior.
 2024-06-14 14:51:40 -07:00
Daniel Hiltgen
6fd04ca922 Improve multi-gpu handling at the limit 
Still not complete, needs some refinement to our prediction to understand the
discrete GPUs available space so we can see how many layers fit in each one
since we can't split one layer across multiple GPUs we can't treat free space
as one logical block
 2024-06-14 14:51:40 -07:00
Jeffrey Morgan
dd7c9ebeaf
server: longer timeout in TestRequests (#5046) 2024-06-14 09:48:25 -07:00
Patrick Devine
94618b2365
add OLLAMA_MODELS to envconfig (#5029) 2024-06-13 12:52:03 -07:00
Jeffrey Morgan
1fd236d177
server: remove jwt decoding error (#5027) 2024-06-13 11:21:15 -07:00
Michael Yang
c16f8af911 fix: multiple templates when creating from model 
multiple templates may appear in a model if a model is created from
another model that 1) has an autodetected template and 2) defines a
custom template
 2024-06-12 13:35:49 -07:00
Michael Yang
515f497e6d fix: skip removing layers that no longer exist 2024-06-10 11:32:19 -07:00
Michael Yang
b27268aaef add test 2024-06-10 11:32:15 -07:00
Michael Yang
030e765e76 fix create model when template detection errors 2024-06-07 10:51:35 -07:00
Michael Yang
9b6c2e6eb6 detect chat template from KV 2024-06-06 16:03:47 -07:00
royjhan
1a29e9a879
API app/browser access (#4879) 
* API app/browser access

* Add tauri (resolves #2291, #4791, #3799, #4388)
 2024-06-06 15:19:03 -07:00
royjhan
4bf1da4944
Separate ListResponse and ModelResponse for api/tags vs api/ps (#4842) 
* Remove false time fields

* Struct Separation for List and Process

* Remove Marshaler
 2024-06-06 10:11:45 -07:00
Blake Mizerany
de5beb06b3 server: skip blob verification for already verified blobs 2024-06-05 16:39:11 -07:00
Michael Yang
d61ef8b954 update create handler to use model.Name 2024-06-04 13:28:25 -07:00
Michael Yang
6297f85606 gofmt, goimports 2024-06-04 13:20:24 -07:00
Michael Yang
8ce4032e72 more lint 2024-06-04 11:13:30 -07:00
Michael Yang
e40145a39d lint 2024-06-04 11:13:30 -07:00
Michael Yang
c895a7d13f some gocritic 2024-06-04 11:13:30 -07:00
Michael Yang
8ffb51749f nolintlint 2024-06-04 11:13:30 -07:00
Michael Yang
04f3c12bb7 replace x/exp/slices with slices 2024-06-04 11:13:30 -07:00
Michael Yang
96bc232b43
Merge pull request #4413 from ollama/mxyng/name-check 
check if name exists before create/pull/copy
 2024-05-29 12:06:58 -07:00
Michael Yang
bca7b12284
Merge pull request #3718 from ollama/mxyng/modelname-3 
update delete handler to use model.Name
 2024-05-29 12:02:07 -07:00
Michael Yang
6adca97f37
Merge pull request #4619 from noxer/patch-1 
Fix download retry issue
 2024-05-24 17:21:57 -07:00
Patrick Devine
4cc3be3035
Move envconfig and consolidate env vars (#4608) 2024-05-24 14:57:15 -07:00
Tim Scheuermann
db2ffa79f1
Fix download retry issue 2024-05-24 20:30:42 +02:00
Jeffrey Morgan
38255d2af1
Use flash attention flag for now (#4580) 
* put flash attention behind flag for now

* add test

* remove print

* up timeout for sheduler tests
 2024-05-22 21:52:09 -07:00
Sang Park
4434d7f447
Correct typo in error message (#4535) 
The spelling of the term "request" has been corrected, which was previously mistakenly written as "requeset" in the error log message.
 2024-05-21 13:39:01 -07:00
Michael Yang
807d092761 fix quantize file types 2024-05-20 15:22:11 -07:00
Michael Yang
f36f1d6be9 tidy intermediate blobs 2024-05-20 15:15:06 -07:00
Michael Yang
3520c0e4d5 cache and reuse intermediate blobs 
particularly useful for zipfiles and f16s
 2024-05-20 13:25:10 -07:00
Patrick Devine
ccdf0b2a44
Move the parser back + handle utf16 files (#4533) 2024-05-20 11:26:45 -07:00
Daniel Hiltgen
02b31c9dc8 Don't return error on signal exit 2024-05-16 16:25:38 -07:00
Michael Yang
84ed77cbd8
Merge pull request #4436 from ollama/mxyng/done-part 
return on part done
 2024-05-15 17:16:24 -07:00
Patrick Devine
d1692fd3e0
fix the cpu estimatedTotal memory + get the expiry time for loading models (#4461) 2024-05-15 15:43:16 -07:00
Patrick Devine
f2cf97d6f1
fix typo in modelfile generation (#4439) 2024-05-14 15:34:29 -07:00
Michael Yang
85a57006d1 check if name exists before create/pull/copy 2024-05-14 14:58:58 -07:00
Michael Yang
c5e892cb3e update tests 2024-05-14 14:56:31 -07:00
Michael Yang
81fb06f530 more resilient Manifests 2024-05-14 14:08:24 -07:00
Michael Yang
a385382ff5 filepath.Join 2024-05-14 14:08:24 -07:00
Michael Yang
b8772a353f remove DeleteModel 2024-05-14 14:08:24 -07:00
Michael Yang
c2714fcbfd routes: use Manifests for ListHandler 2024-05-14 14:08:24 -07:00
Michael Yang
a2fc933fed update delete handler to use model.Name 2024-05-14 14:08:24 -07:00
Michael Yang
ac145f75ca return on part done 2024-05-14 13:04:30 -07:00
Ryo Machida
798b107f19
Fixed the API endpoint /api/tags when the model list is empty. (#4424) 
* Fixed the API endpoint /api/tags to return {models: []} instead of {models: null} when the model list is empty.

* Update server/routes.go

---------

Co-authored-by: Jeffrey Morgan <jmorganca@gmail.com>
 2024-05-14 11:18:10 -07:00
Daniel Hiltgen
ec231a7923 Remove VRAM convergence check for windows 
The APIs we query are optimistic on free space, and windows pages
VRAM, so we don't have to wait to see reported usage recover on unload
 2024-05-14 09:53:46 -07:00
Patrick Devine
7ca71a6b0f
don't abort when an invalid model name is used in /save (#4416) 2024-05-13 18:48:28 -07:00
Patrick Devine
6845988807
Ollama ps command for showing currently loaded models (#4327) 2024-05-13 17:17:36 -07:00
jmorganca
4ec7445a6f Revert "use post token" 
This reverts commit 0fec3525ad.
 2024-05-11 22:19:14 -07:00
Michael Yang
0fec3525ad use post token 2024-05-11 19:13:16 -07:00
Daniel Hiltgen
824ee5446f Fix envconfig unit test 2024-05-10 16:49:48 -07:00
Daniel Hiltgen
4142c3ef7c Always use the sorted list of GPUs 
Make sure the first GPU has the most free space
 2024-05-10 13:53:21 -07:00
Jeffrey Morgan
6602e793c0
Use --quantize flag and quantize api parameter (#4321) 
* rename `--quantization` to `--quantize`

* backwards

* Update api/types.go

Co-authored-by: Michael Yang <mxyng@pm.me>

---------

Co-authored-by: Michael Yang <mxyng@pm.me>
 2024-05-10 13:06:13 -07:00
Jeffrey Morgan
bb6fd02298
Don't clamp ctx size in PredictServerFit (#4317) 
* dont clamp ctx size in `PredictServerFit`

* minimum 4 context

* remove context warning
 2024-05-10 10:17:12 -07:00
Michael Yang
e03637176d fix(routes): skip bad manifests 2024-05-10 08:46:11 -07:00
Jeffrey Morgan
302d7fdbf3
prune partial downloads (#4272) 2024-05-09 16:35:20 -07:00
Daniel Hiltgen
3ae2f441e0 Fix race in shutdown logic 
Ensure the runners are terminated
 2024-05-09 15:54:02 -07:00
Daniel Hiltgen
354ad9254e Wait for GPU free memory reporting to converge 
The GPU drivers take a while to update their free memory reporting, so we need
to wait until the values converge with what we're expecting before proceeding
to start another runner in order to get an accurate picture.
 2024-05-09 14:56:01 -07:00
Daniel Hiltgen
8727a9c140 Record more GPU information 
This cleans up the logging for GPU discovery a bit, and can
serve as a foundation to report GPU information in a future UX.
 2024-05-09 14:18:14 -07:00
Bruce MacDonald
cfa84b8470
add done_reason to the api (#4235) 2024-05-09 13:30:14 -07:00
Michael Yang
a7ee84fc31 routes: skip invalid filepaths 2024-05-09 11:23:22 -07:00
Jeffrey Morgan
d5eec16d23
use model defaults for num_gqa, rope_frequency_base and rope_frequency_scale (#1983) 2024-05-09 09:06:13 -07:00
Bruce MacDonald
cef45feaa4
Add preflight OPTIONS handling and update CORS config (#4086) 
* Add preflight OPTIONS handling and update CORS config

- Implement early return with HTTP 204 (No Content) for OPTIONS requests in allowedHostsMiddleware to optimize preflight handling.

- Extend CORS configuration to explicitly allow 'Authorization' headers and 'OPTIONS' method when OLLAMA_ORIGINS environment variable is set.

* allow auth, content-type, and user-agent headers

* Update routes.go
 2024-05-08 13:14:00 -07:00
Michael Yang
b25976aeb8 routes: fix show llava models 2024-05-08 12:43:36 -07:00
Michael Yang
88cf154483
Merge pull request #4244 from ollama/mxyng/skip-if-same 
skip if same quantization
 2024-05-07 19:03:37 -07:00
Bruce MacDonald
8cbd3e7510
skip hidden files in list models handler (#4247) 2024-05-07 19:01:45 -07:00
Michael Yang
eeb695261f skip if same quantization 2024-05-07 17:44:19 -07:00
Bruce MacDonald
dc9b1111e0 fix invalid destination error message 2024-05-07 17:35:52 -07:00
Michael Yang
ffbd3d173f
Merge pull request #3715 from ollama/mxyng/modelname-2 
update list handler to use model.Name
 2024-05-07 15:21:39 -07:00
Michael Yang
1e0a669f75
Merge pull request #3682 from ollama/mxyng/quantize-all-the-things 
quantize any fp16/fp32 model
 2024-05-07 15:20:49 -07:00
Michael Yang
548a7df014 update list handler to use model.Name 2024-05-07 09:38:45 -07:00
Jeffrey Morgan
39d9d22ca3
close server on receiving signal (#4213) 2024-05-06 16:01:37 -07:00
Michael Yang
b2f00aa977 close zip files 2024-05-06 15:27:19 -07:00
Michael Yang
f5e8b207fb s/DisplayLongest/String/ 2024-05-06 15:24:01 -07:00
Michael Yang
d245460362 only quantize language models 2024-05-06 15:24:01 -07:00
Michael Yang
4d0d0fa383 no iterator 2024-05-06 15:24:01 -07:00
Michael Yang
7ffe45734d rebase 2024-05-06 15:24:01 -07:00
Michael Yang
01811c176a comments 2024-05-06 15:24:01 -07:00
Michael Yang
a7248f6ea8 update tests 2024-05-06 15:24:01 -07:00
Michael Yang
9685c34509 quantize any fp16/fp32 model 
- FROM /path/to/{safetensors,pytorch}
- FROM /path/to/fp{16,32}.bin
- FROM model:fp{16,32}
 2024-05-06 15:24:01 -07:00
Daniel Hiltgen
0963c65027
Merge pull request #4208 from dhiltgen/fix_sched_test 
Fix stale test logic
 2024-05-06 14:23:12 -07:00
Jeffrey Morgan
c9f98622b1
Skip scheduling cancelled requests, always reload unloaded runners (#4189) 2024-05-06 14:22:24 -07:00
Daniel Hiltgen
0a954e5066 Fix stale test logic 
The model processing was recently changed to be deferred but
this test scenario hadn't been adjusted for that change in behavior.
 2024-05-06 14:15:37 -07:00
Jeffrey Morgan
dfa2f32ca0
unload in critical section (#4187) 2024-05-05 17:18:27 -07:00
Daniel Hiltgen
f56aa20014 Centralize server config handling 
This moves all the env var reading into one central module
and logs the loaded config once at startup which should
help in troubleshooting user server logs
 2024-05-05 16:49:50 -07:00
Jeffrey Morgan
942c979232
allocate a large enough kv cache for all parallel requests (#4162) 2024-05-05 15:59:32 -07:00
Patrick Devine
2a21363bb7
validate the format of the digest when getting the model path (#4175) 2024-05-05 11:46:12 -07:00
Daniel Hiltgen
20f6c06569 Make maximum pending request configurable 
This also bumps up the default to be 50 queued requests
instead of 10.
 2024-05-04 21:00:52 -07:00
Michael Yang
b7a87a22b6
Merge pull request #4059 from ollama/mxyng/parser-2 
rename parser to model/file
 2024-05-03 13:01:22 -07:00
Daniel Hiltgen
9a32c514cb Soften timeouts on sched unit tests 
This gives us more headroom on the scheduler tests to tamp
down some flakes.
 2024-05-03 09:08:33 -07:00
Michael Yang
e9ae607ece
Merge pull request #3892 from ollama/mxyng/parser 
refactor modelfile parser
 2024-05-02 17:04:47 -07:00
Michael Yang
5b806d8d24
Merge pull request #4089 from ollama/mxyng/target-invalid 
server: destination invalid
 2024-05-01 12:46:35 -07:00
Michael Yang
45b6a12e45 server: target invalid 2024-05-01 12:40:45 -07:00
Mark Ward
63c763685f log when the waiting for the process to stop to help debug when other tasks execute during this wait. 
expire timer clear the timer reference because it will not be reused.
close will clean up expireTimer if calling code has not already done this.
 2024-05-01 18:51:10 +00:00
Mark Ward
f4a73d57a4 fix runner expire during active use. Clearing the expire timer as it is used. Allowing the finish to assign an expire timer so that the runner will expire after no use. 2024-05-01 18:51:10 +00:00
Michael Yang
119589fcb3 rename parser to model/file 2024-05-01 09:53:50 -07:00
Michael Yang
9cf0f2e973 use parser.Format instead of templating modelfile 2024-05-01 09:52:54 -07:00
Michael Yang
c0a00f68ae refactor modelfile parser 2024-05-01 09:52:54 -07:00
Bruce MacDonald
0a7fdbe533
prompt to display and add local ollama keys to account (#3717) 
- return descriptive error messages when unauthorized to create blob or push a model
- display the local public key associated with the request that was denied
 2024-04-30 11:02:08 -07:00
Jeffrey Morgan
586672f490
fix copying model to itself (#4019) 2024-04-28 23:47:49 -04:00
Daniel Hiltgen
d6e3b64582 Fix concurrency for CPU mode 
Prior refactoring passes accidentally removed the logic to bypass VRAM
checks for CPU loads.  This adds that back, along with test coverage.

This also fixes loaded map access in the unit test to be behind the mutex which was
likely the cause of various flakes in the tests.
 2024-04-28 13:42:39 -07:00
Jeffrey Morgan
bb31def011
return code 499 when user cancels request while a model is loading (#3955) 2024-04-26 17:38:29 -04:00
Blake Mizerany
37f9c8ad99
types/model: overhaul Name and Digest types (#3924) 2024-04-26 13:08:32 -07:00
Daniel Hiltgen
9b5a3c5991
Merge pull request #3914 from dhiltgen/mac_perf 
Improve mac parallel performance
 2024-04-25 16:28:31 -07:00
Jeffrey Morgan
00b0699c75
Reload model if num_gpu changes (#3920) 
* reload model if `num_gpu` changes

* dont reload on -1

* fix tests
 2024-04-25 19:02:40 -04:00
Daniel Hiltgen
b123be5b71 Adjust context size for parallelism 2024-04-25 13:58:54 -07:00
Daniel Hiltgen
f503a848c2
Merge pull request #3895 from brycereitano/shiftloading 
Move ggml loading to when attempting to fit
 2024-04-25 09:24:08 -07:00
Bryce Reitano
36a6daccab Restructure loading conditional chain 2024-04-24 17:37:03 -06:00
Bryce Reitano
ceb0e26e5e Provide variable ggml for TestLoad 2024-04-24 17:19:55 -06:00
Bryce Reitano
284e02bed0 Move ggml loading to when we attempt fitting 2024-04-24 17:17:24 -06:00
Michael Yang
592dae31c8 update copy to use model.Name 2024-04-24 15:54:54 -07:00
Daniel Hiltgen
d8851cb7a0 Harden sched TestLoad 
Give the go routine a moment to deliver the expired event
 2024-04-23 16:14:47 -07:00
Daniel Hiltgen
34b9db5afc Request and model concurrency 
This change adds support for multiple concurrent requests, as well as
loading multiple models by spawning multiple runners. The default
settings are currently set at 1 concurrent request per model and only 1
loaded model at a time, but these can be adjusted by setting
OLLAMA_NUM_PARALLEL and OLLAMA_MAX_LOADED_MODELS.
 2024-04-22 19:29:12 -07:00
Cheng
62be2050dd
chore: use errors.New to replace fmt.Errorf will much better (#3789) 2024-04-20 22:11:06 -04:00
Patrick Devine
9f8691c6c8
Add llama2 / torch models for ollama create (#3607) 2024-04-15 11:26:42 -07:00
Jeffrey Morgan
a0b8a32eb4
Terminate subprocess if receiving SIGINT or SIGTERM signals while model is loading (#3653) 
* terminate subprocess if receiving `SIGINT` or `SIGTERM` signals while model is loading

* use `unload` in signal handler
 2024-04-15 12:09:32 -04:00
Blake Mizerany
a7b431e743
server: provide helpful workaround hint when stalling on pull (#3584) 
This is a quick fix to help users who are stuck on the "pull" step at
99%.

In the near future we're introducing a new registry client that
should/will hopefully be smarter. In the meantime, this should unblock
the users hitting issue #1736.
 2024-04-10 16:24:37 -07:00
Michael Yang
9502e5661f cgo quantize 2024-04-08 15:31:08 -07:00
Michael Yang
e1c9a2a00f no blob create if already exists 2024-04-08 15:09:48 -07:00
Daniel Hiltgen
6589eb8a8c Revert options as a ref in the server 2024-04-02 16:44:10 -07:00
Daniel Hiltgen
58d95cc9bd Switch back to subprocessing for llama.cpp 
This should resolve a number of memory leak and stability defects by allowing
us to isolate llama.cpp in a separate process and shutdown when idle, and
gracefully restart if it has problems.  This also serves as a first step to be
able to run multiple copies to support multiple models concurrently.
 2024-04-01 16:48:18 -07:00
Patrick Devine
3b6a9154dd
Simplify model conversion (#3422) 2024-04-01 16:14:53 -07:00
Michael Yang
91b3e4d282 update memory calcualtions 
count each layer independently when deciding gpu offloading
 2024-04-01 13:16:32 -07:00
Michael Yang
d338d70492 refactor model parsing 2024-04-01 13:16:15 -07:00
Patrick Devine
5a5efee46b
Add gemma safetensors conversion (#3250) 
Co-authored-by: Michael Yang <mxyng@pm.me>
 2024-03-28 18:54:01 -07:00
Michael Yang
af8a8a6b59 fix: trim quotes on OLLAMA_ORIGINS 2024-03-27 15:24:29 -07:00
Patrick Devine
1b272d5bcd
change github.com/jmorganca/ollama to github.com/ollama/ollama (#3347) 2024-03-26 13:04:17 -07:00
Daniel Hiltgen
949b6c01e0 Revamp go based integration tests 
This uplevels the integration tests to run the server which can allow
testing an existing server, or a remote server.
 2024-03-23 14:24:18 +01:00
Blake Mizerany
703684a82a
server: replace blob prefix separator from ':' to '-' (#3146) 
This fixes issues with blob file names that contain ':' characters to be rejected by file systems that do not support them.
 2024-03-14 20:18:06 -07:00
Patrick Devine
47cfe58af5
Default Keep Alive environment variable (#3094) 
---------

Co-authored-by: Chris-AS1 <8493773+Chris-AS1@users.noreply.github.com>
 2024-03-13 13:29:40 -07:00
Daniel Hiltgen
4a5c9b8035 Finish unwinding idempotent payload logic 
The recent ROCm change partially removed idempotent
payloads, but the ggml-metal.metal file for mac was still
idempotent.  This finishes switching to always extract
the payloads, and now that idempotentcy is gone, the
version directory is no longer useful.
 2024-03-09 08:34:39 -08:00
Jeffrey Morgan
5b3fad9636 separate out isLocalIP 2024-03-09 00:22:08 -08:00
Jeffrey Morgan
bfec2c6e10 simplify host checks 2024-03-08 23:29:53 -08:00
Jeffrey Morgan
5c143af726 add additional allowed hosts 2024-03-08 23:23:59 -08:00
Jeffrey Morgan
fc8c044584
add allowed host middleware and remove workDir middleware (#3018) 2024-03-08 22:23:47 -08:00
Michael Yang
76bdebbadf decode ggla 2024-03-08 15:46:25 -08:00
Bruce MacDonald
0cebc79cba
fix: allow importing a model from name reference (#3005) 2024-03-08 12:27:47 -05:00
Jeffrey Morgan
fc06205971
Revert "adjust download and upload concurrency based on available bandwidth" (#2995) 2024-03-07 18:10:16 -08:00
Daniel Hiltgen
6c5ccb11f9 Revamp ROCm support 
This refines where we extract the LLM libraries to by adding a new
OLLAMA_HOME env var, that defaults to `~/.ollama` The logic was already
idempotenent, so this should speed up startups after the first time a
new release is deployed.  It also cleans up after itself.

We now build only a single ROCm version (latest major) on both windows
and linux.  Given the large size of ROCms tensor files, we split the
dependency out.  It's bundled into the installer on windows, and a
separate download on windows.  The linux install script is now smart and
detects the presence of AMD GPUs and looks to see if rocm v6 is already
present, and if not, then downloads our dependency tar file.

For Linux discovery, we now use sysfs and check each GPU against what
ROCm supports so we can degrade to CPU gracefully instead of having
llama.cpp+rocm assert/crash on us.  For Windows, we now use go's windows
dynamic library loading logic to access the amdhip64.dll APIs to query
the GPU information.
 2024-03-07 10:36:50 -08:00
Michael Yang
2e20110e50
Merge pull request #2221 from ollama/mxyng/up-down-ccy 
adjust download and upload concurrency based on available bandwidth
 2024-03-07 09:27:33 -08:00
Patrick Devine
2c017ca441
Convert Safetensors to an Ollama model (#2824) 2024-03-06 21:01:51 -08:00
Jeffrey Morgan
3b4bab3dc5
Fix embeddings load model behavior (#2848) 2024-02-29 17:40:56 -08:00
Michael Yang
0e19476b56
prepend image tags (#2789) 
instead of appending image tags, prepend them - this generally produces better results
 2024-02-29 11:30:14 -08:00
Michael Yang
084d846621 refactor 2024-02-21 13:42:48 -08:00
Michael Yang
6a4b994433 lint 2024-02-21 13:42:48 -08:00
Michael Yang
bea007deb7 use LimitGroup for uploads 2024-02-21 13:42:48 -08:00
Michael Yang
074934be03 adjust group limit based on download speed 2024-02-21 13:42:48 -08:00
Michael Yang
0de12368a0 add new LimitGroup for dynamic concurrency 2024-02-21 13:42:48 -08:00
Michael Yang
917bd61084 refactor download run 2024-02-21 13:42:46 -08:00
Jeffrey Morgan
287ba11500 better error message when calling /api/generate or /api/chat with embedding models 2024-02-20 21:53:45 -05:00
Jeffrey Morgan
63861f58cc
Support for bert and nomic-bert embedding models 2024-02-20 21:37:29 -05:00
Michael Yang
210b65268e
replace strings buffer with hasher (#2437) 
the buffered value is going into the hasher eventually so write directly
to the hasher instead
 2024-02-20 19:07:50 -05:00
Michael Yang
897b213468
use http.DefaultClient (#2530) 
default client already handles proxy
 2024-02-20 18:34:47 -05:00
Bruce MacDonald
88622847c6
fix: chat system prompting overrides (#2542) 2024-02-16 14:42:43 -05:00
Michael Yang
e43648afe5 rerefactor 2024-02-15 05:56:45 +00:00
Daniel Hiltgen
f397e0e988 Move hub auth out to new package 2024-02-15 05:56:45 +00:00
Jeffrey Morgan
48a273f80b
Fix issues with templating prompt in chat mode (#2460) 2024-02-12 15:06:57 -08:00
Jeffrey Morgan
1f9078d6ae
Check image filetype in api handlers (#2467) 2024-02-12 11:16:20 -08:00
Jeffrey Morgan
a0a199b108
Fix hanging issue when sending empty content (#2399) 2024-02-07 19:30:33 -05:00
Jeffrey Morgan
453f572f83
Initial OpenAI /v1/chat/completions API compatibility (#2376) 2024-02-07 17:24:29 -05:00
Michael Yang
e805ac1d59 fix response on token error 2024-02-07 11:05:49 -08:00
Michael Yang
bfbf2f7cf7
Merge pull request #2296 from ollama/mxyng/img-tags 
append image tags to user content
 2024-02-01 13:16:59 -08:00
Michael Yang
3d6f48507a structured debug prompt 2024-02-01 11:56:28 -08:00
Michael Yang
f3761405c8 use image id 2024-02-01 11:52:42 -08:00
Michael Yang
e49dc9f3d8 fix tests 2024-02-01 11:48:11 -08:00
Michael Yang
d125510b4b remove image tags 2024-02-01 11:32:51 -08:00
Michael Yang
fb56988014 account for image projection in token count 2024-02-01 09:50:48 -08:00
Michael Yang
d046bee790 use llm.ImageData for chat 2024-01-31 19:18:25 -08:00
Jeffrey Morgan
f11bf0740b use llm.ImageData 2024-01-31 19:13:48 -08:00
Michael Yang
8450bf66e6 trim images 2024-01-31 19:13:47 -08:00
Michael Yang
b4e11be8ef append image tags to user content 2024-01-31 19:13:10 -08:00
Bruce MacDonald
a896079705
preserve last system message from modelfile (#2289) 2024-01-31 21:45:01 -05:00
Michael Yang
8ac08a0eec update slog handler options 
- consistent format by using text handler for debug and non-debug
- truncate source file to just the file name
 2024-01-31 15:15:00 -08:00
Michael Yang
c8b1f2369e remove unnecessary parse raw 2024-01-30 17:00:53 -08:00
Bruce MacDonald
0632dff3f8
trim chat prompt based on llm context size (#1963) 2024-01-30 15:59:29 -05:00
Jeffrey Morgan
f2245c7c77
print prompt with OLLAMA_DEBUG=1 (#2245) 2024-01-28 15:22:35 -08:00
Jeffrey Morgan
e4b9b72f2a
Do not repeat system prompt for chat templating (#2241) 2024-01-28 14:15:56 -08:00
Patrick Devine
b5cf31b460
add keep_alive to generate/chat/embedding api endpoints (#2146) 2024-01-26 14:28:02 -08:00
Michael Yang
9d3dcfd0ec fix logging 2024-01-26 11:04:27 -08:00
Michael Yang
6e0ea5ecc8
Merge pull request #1916 from ollama/mxyng/inactivity-monitor 
download: add inactivity monitor
 2024-01-26 10:56:00 -08:00
Patrick Devine
7c40a67841
Save and load sessions (#2063) 2024-01-25 12:12:36 -08:00
Michael Yang
c08dfaa23d fix: remove overwritten model layers 
if create overrides a manifest, first add the older manifest's layers to
the delete map so they can be cleaned up
 2024-01-19 14:58:37 -08:00
Michael Yang
aac9ab4db7 fix show handler 2024-01-18 15:36:50 -08:00
Michael Yang
745b5934fa add model to ModelResponse 2024-01-18 14:32:55 -08:00
Michael Yang
a38d88d828 api: add model for all requests 
prefer using req.Model and fallback to req.Name
 2024-01-18 14:31:37 -08:00
Daniel Hiltgen
fedd705aea Mechanical switch from log to slog 
A few obvious levels were adjusted, but generally everything mapped to "info" level.
 2024-01-18 14:12:57 -08:00