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
Michael Yang
171eb040fc
simplify safetensors reading
2024-05-21 11:28:22 -07:00
Michael Yang
bbbd9f20f3
cleanup
2024-05-20 16:13:57 -07:00
Michael Yang
547132e820
bpe pretokenizer
2024-05-20 16:13:57 -07:00
Patrick Devine
c8cf0d94ed
llama3 conversion
2024-05-20 16:13:57 -07:00
jmorganca
5cab13739e
set llama.cpp submodule commit to 614d3b9
2024-05-20 15:28:17 -07:00
Josh Yan
8aadad9c72
updated updateURL
2024-05-20 15:24:32 -07:00
Sam
e15307fdf4
feat: add support for flash_attn ( #4120 )
...
* feat: enable flash attention if supported
* feat: enable flash attention if supported
* feat: enable flash attention if supported
* feat: add flash_attn support
2024-05-20 13:36:03 -07:00
Jeffrey Morgan
583c1f472c
update llama.cpp submodule to 614d3b9 ( #4414 )
2024-05-16 13:53:09 -07:00
Daniel Hiltgen
c48c1d7c46
Port cuda/rocm skip build vars to linux
...
Windows already implements these, carry over to linux.
2024-05-15 15:56:43 -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
Daniel Hiltgen
853ae490e1
Sanitize the env var debug log
...
Only dump env vars we care about in the logs
2024-05-15 14:42:57 -07:00
Michael Yang
0e331c7168
Merge pull request #4328 from ollama/mxyng/mem
...
count memory up to NumGPU if set by user
2024-05-14 13:47:44 -07:00
Patrick Devine
6845988807
Ollama ps command for showing currently loaded models ( #4327 )
2024-05-13 17:17:36 -07:00
Michael Yang
1d359e737e
typo
2024-05-13 14:18:34 -07:00
Michael Yang
50b9056e09
count memory up to NumGPU
2024-05-13 14:13:10 -07:00
jmorganca
92ca2cca95
Revert "only forward some env vars"
...
This reverts commit ce3b212d12
2024-05-10 22:53:21 -07:00
Daniel Hiltgen
c4014e73a2
Fall back to CPU runner with zero layers
2024-05-10 15:09:48 -07:00
Michael Yang
1eb382da5a
add phi2 mem
2024-05-10 12:13:28 -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
cf442cd57e
fix typo
2024-05-09 16:23:37 -07:00
Michael Yang
ce3b212d12
only forward some env vars
2024-05-09 15:16:09 -07:00
Michael Yang
58876091f7
log clean up
2024-05-09 14:55:36 -07:00
Daniel Hiltgen
d0425f26cf
Merge pull request #4294 from dhiltgen/harden_subprocess_reaping
...
Harden subprocess reaping
2024-05-09 14:02:16 -07:00
Bruce MacDonald
cfa84b8470
add done_reason to the api ( #4235 )
2024-05-09 13:30:14 -07:00
Daniel Hiltgen
84ac7ce139
Refine subprocess reaping
2024-05-09 11:21:31 -07:00
Daniel Hiltgen
920a4b0794
Merge remote-tracking branch 'upstream/main' into pr3702
2024-05-08 16:44:35 -07:00
Daniel Hiltgen
ee49844d09
Merge pull request #4153 from dhiltgen/gpu_verbose_response
...
Add GPU usage
2024-05-08 16:39:11 -07:00
Daniel Hiltgen
8a516ac862
Merge pull request #4241 from dhiltgen/fix_tmp_override
...
Detect noexec and report a better error
2024-05-08 15:34:22 -07:00
Daniel Hiltgen
bee2f4a3b0
Record GPU usage information
...
This records more GPU usage information for eventual UX inclusion.
2024-05-08 14:45:39 -07:00
Michael Yang
eeb695261f
skip if same quantization
2024-05-07 17:44:19 -07:00
Daniel Hiltgen
72700279e2
Detect noexec and report a better error
...
This will bubble up a much more informative error message if noexec
is preventing us from running the subprocess
2024-05-07 16:46:15 -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
4736391bfb
llm: add minimum based on layer size
2024-05-06 17:04:19 -07:00
Michael Yang
01811c176a
comments
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
380378cc80
Use our libraries first
...
Trying to live off the land for cuda libraries was not the right strategy. We need to use the version we compiled against to ensure things work properly
2024-05-06 14:23:29 -07:00
Jeffrey Morgan
ed740a2504
Fix no slots available error with concurrent requests ( #4160 )
2024-05-06 14:22:53 -07:00
Jeffrey Morgan
1b0e6c9c0e
Fix llava models not working after first request ( #4164 )
...
* fix llava models not working after first request
* individual requests only for llava models
2024-05-05 20:50:31 -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
Michael Yang
44869c59d6
omit prompt and generate settings from final response
2024-05-03 17:00:02 -07:00
Mark Ward
321d57e1a0
Removing go routine calling .wait from load.
2024-05-01 18:51:10 +00:00
Mark Ward
ba26c7aa00
it will always return an error due to Kill() discarding Wait() errors
2024-05-01 18:51:10 +00:00
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
948114e3e3
fix sched to wait for the runner to terminate to ensure following vram check will be more accurate
2024-05-01 18:51:10 +00:00
Jeffrey Morgan
f0c454ab57
gpu: add 512MiB to darwin minimum, metal doesn't have partial offloading overhead ( #4068 )
2024-05-01 11:46:03 -04:00
jmorganca
fcf4d60eee
llm: add back check for empty token cache
2024-04-30 17:38:44 -04:00
jmorganca
e33d5c2dbc
update llama.cpp commit to 952d03d
2024-04-30 17:31:20 -04:00
Jeffrey Morgan
18d9a7e1f1
update llama.cpp submodule to f364eb6 ( #4060 )
2024-04-30 17:25:39 -04:00
Daniel Hiltgen
23d23409a0
Update llama.cpp ( #4036 )
...
* Bump llama.cpp to b2761
* Adjust types for bump
2024-04-29 23:18:48 -04:00
Jeffrey Morgan
7aa08a77ca
llm: dont cap context window limit to training context window ( #3988 )
2024-04-29 10:07:30 -04:00
Hernan Martinez
8a65717f55
Do not build AVX runners on ARM64
2024-04-26 23:55:32 -06:00
Hernan Martinez
b438d485f1
Use architecture specific folders in the generate script
2024-04-26 23:34:12 -06:00
Hernan Martinez
86e67fc4a9
Add import declaration for windows,arm64 to llm.go
2024-04-26 23:23:53 -06:00
Daniel Hiltgen
e4859c4563
Fine grain control over windows generate steps
...
This will speed up CI which already tries to only build static for unit tests
2024-04-26 15:49:46 -07:00
Daniel Hiltgen
0b5c589ca2
Merge pull request #3966 from dhiltgen/bump
...
Fix target in gen_windows.ps1
2024-04-26 15:36:53 -07:00
Michael Yang
65fadddc85
Merge pull request #3964 from ollama/mxyng/weights
...
fix gemma, command-r layer weights
2024-04-26 15:23:33 -07:00
Daniel Hiltgen
ed5fb088c4
Fix target in gen_windows.ps1
2024-04-26 15:10:42 -07:00
Michael Yang
f81f308118
fix gemma, command-r layer weights
2024-04-26 15:00:55 -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
Daniel Hiltgen
5c0c2d1d09
Merge pull request #3954 from dhiltgen/ci_fixes
...
Put back non-avx CPU build for windows
2024-04-26 13:09:03 -07:00
Daniel Hiltgen
421c878a2d
Put back non-avx CPU build for windows
2024-04-26 12:44:07 -07:00
Daniel Hiltgen
85801317d1
Fix clip log import
2024-04-26 09:43:46 -07:00
Daniel Hiltgen
2ed0d65948
Bump llama.cpp to b2737
2024-04-26 09:43:28 -07:00
Daniel Hiltgen
8671fdeda6
Refactor windows generate for more modular usage
2024-04-26 08:35:50 -07:00
Daniel Hiltgen
8feb97dc0d
Move cuda/rocm dependency gathering into generate script
...
This will make it simpler for CI to accumulate artifacts from prior steps
2024-04-25 22:38:44 -07:00
Michael Yang
de4ded68b0
Merge pull request #3923 from ollama/mxyng/mem
...
only count output tensors
2024-04-25 16:34:17 -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
993cf8bf55
llm: limit generation to 10x context size to avoid run on generations ( #3918 )
...
* llm: limit generation to 10x context size to avoid run on generations
* add comment
* simplify condition statement
2024-04-25 19:02:30 -04:00
Michael Yang
7bb7cb8a60
only count output tensors
2024-04-25 15:24:08 -07:00
jmorganca
ddf5c09a9b
use matrix multiplcation kernels in more cases
2024-04-25 13:58:54 -07:00
Roy Yang
5f73c08729
Remove trailing spaces ( #3889 )
2024-04-25 14:32:26 -04:00
Daniel Hiltgen
6e76348df7
Merge pull request #3834 from dhiltgen/not_found_in_path
...
Report errors on server lookup instead of path lookup failure
2024-04-24 10:50:48 -07:00
Patrick Devine
14476d48cc
fixes for gguf ( #3863 )
2024-04-23 20:57:20 -07:00
Daniel Hiltgen
5445aaa94e
Add back memory escape valve
...
If we get our predictions wrong, this can be used to
set a lower memory limit as a workaround. Recent multi-gpu
refactoring accidentally removed it, so this adds it back.
2024-04-23 17:09:02 -07:00
Daniel Hiltgen
058f6cd2cc
Move nested payloads to installer and zip file on windows
...
Now that the llm runner is an executable and not just a dll, more users are facing
problems with security policy configurations on windows that prevent users
writing to directories and then executing binaries from the same location.
This change removes payloads from the main executable on windows and shifts them
over to be packaged in the installer and discovered based on the executables location.
This also adds a new zip file for people who want to "roll their own" installation model.
2024-04-23 16:14:47 -07:00
Daniel Hiltgen
58888a74bc
Detect and recover if runner removed
...
Tmp cleaners can nuke the file out from underneath us. This detects the missing
runner, and re-initializes the payloads.
2024-04-23 10:05:26 -07:00
Daniel Hiltgen
cc5a71e0e3
Merge pull request #3709 from remy415/custom-gpu-defs
...
Adds support for customizing GPU build flags in llama.cpp
2024-04-23 09:28:34 -07:00
Michael Yang
e83bcf7f9a
Merge pull request #3836 from ollama/mxyng/mixtral
...
fix: mixtral graph
2024-04-23 09:15:10 -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
Daniel Hiltgen
8711d03df7
Report errors on server lookup instead of path lookup failure
2024-04-22 19:08:47 -07:00
Michael Yang
435cc866a3
fix: mixtral graph
2024-04-22 17:19:44 -07:00
Daniel Hiltgen
aa72281eae
Trim spaces and quotes from llm lib override
2024-04-22 17:11:14 -07:00
Jeremy
9c0db4cc83
Update gen_windows.ps1
...
Fixed improper env references
2024-04-21 16:13:41 -04:00
Cheng
62be2050dd
chore: use errors.New to replace fmt.Errorf will much better ( #3789 )
2024-04-20 22:11:06 -04:00
Jeremy
6f18297b3a
Update gen_windows.ps1
...
Forgot a " on the write-host
2024-04-18 19:47:44 -04:00
Jeremy
15016413de
Update gen_windows.ps1
...
Added OLLAMA_CUSTOM_CUDA_DEFS and OLLAMA_CUSTOM_ROCM_DEFS to customize GPU builds on Windows
2024-04-18 19:27:16 -04:00
Jeremy
440b7190ed
Update gen_linux.sh
...
Added OLLAMA_CUSTOM_CUDA_DEFS and OLLAMA_CUSTOM_ROCM_DEFS instead of OLLAMA_CUSTOM_GPU_DEFS
2024-04-18 19:18:10 -04:00
ManniX-ITA
c496967e56
Merge branch 'ollama:main' into mannix-server
2024-04-18 18:45:15 +02:00
Jeremy
3934c15895
Merge branch 'ollama:main' into custom-gpu-defs
2024-04-18 09:55:10 -04:00
Jeremy
fd048f1367
Merge branch 'ollama:main' into arm64static
2024-04-18 09:55:04 -04:00
Michael Yang
8645076a71
Merge pull request #3712 from ollama/mxyng/mem
...
add stablelm graph calculation
2024-04-17 15:57:51 -07:00
Michael Yang
05e9424824
Merge pull request #3664 from ollama/mxyng/fix-padding-2
...
fix padding to only return padding
2024-04-17 15:57:40 -07:00
Michael Yang
3cf483fe48
add stablelm graph calculation
2024-04-17 13:57:19 -07:00
Jeremy
52f5370c48
add support for custom gpu build flags for llama.cpp
2024-04-17 16:00:48 -04:00
Jeremy
7c000ec3ed
adds support for OLLAMA_CUSTOM_GPU_DEFS to customize GPU build flags
2024-04-17 15:21:05 -04:00
Jeremy
ea4c284a48
Merge branch 'ollama:main' into arm64static
2024-04-17 15:11:38 -04:00
Jeremy
8aec92fa6d
rearranged conditional logic for static build, dockerfile updated
2024-04-17 14:43:28 -04:00
Michael Yang
a8b9b930b4
account for all non-repeating layers
2024-04-17 11:21:21 -07:00
Jeremy
70261b9bb6
move static build to its own flag
2024-04-17 13:04:28 -04:00
ManniX-ITA
c942e4a07b
Fixed startup sequence to report model loading
2024-04-17 17:40:32 +02:00
ManniX-ITA
bd54b08261
Streamlined WaitUntilRunning
2024-04-17 17:39:52 +02:00
Michael Yang
e74163af4c
fix padding to only return padding
2024-04-16 15:43:26 -07:00
Michael Yang
26df674785
scale graph based on gpu count
2024-04-16 14:44:13 -07:00
Jeffrey Morgan
7c9792a6e0
Support unicode characters in model path ( #3681 )
...
* parse wide argv characters on windows
* cleanup
* move cleanup to end of `main`
2024-04-16 17:00:12 -04:00
Michael Yang
41a272de9f
darwin: no partial offloading if required memory greater than system
2024-04-16 11:22:38 -07:00
Jeffrey Morgan
f335722275
update llama.cpp submodule to 7593639 ( #3665 )
2024-04-15 23:04:43 -04:00
Michael Yang
6d53b67c2c
Merge pull request #3663 from ollama/mxyng/fix-padding
2024-04-15 17:44:54 -07:00
Michael Yang
969238b19e
fix padding in decode
...
TODO: update padding() to _only_ returning the padding
2024-04-15 17:27:06 -07:00
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
Jeffrey Morgan
309aef7fee
update llama.cpp submodule to 4bd0f93 ( #3627 )
2024-04-13 10:43:02 -07:00
Michael Yang
3397eff0cd
mixtral mem
2024-04-11 11:10:41 -07:00
Michael Yang
7e33a017c0
partial offloading
2024-04-10 11:37:20 -07:00
Michael Yang
8b2c10061c
refactor tensor query
2024-04-10 11:37:20 -07:00
Daniel Hiltgen
c5ff443b9f
Handle very slow model loads
...
During testing, we're seeing some models take over 3 minutes.
2024-04-09 16:35:10 -07:00
Blake Mizerany
1524f323a3
Revert "build.go: introduce a friendlier way to build Ollama ( #3548 )" ( #3564 )
2024-04-09 15:57:45 -07:00
Blake Mizerany
fccf3eecaa
build.go: introduce a friendlier way to build Ollama ( #3548 )
...
This commit introduces a more friendly way to build Ollama dependencies
and the binary without abusing `go generate` and removing the
unnecessary extra steps it brings with it.
This script also provides nicer feedback to the user about what is
happening during the build process.
At the end, it prints a helpful message to the user about what to do
next (e.g. run the new local Ollama).
2024-04-09 14:18:47 -07:00
Michael Yang
c77d45d836
Merge pull request #3506 from ollama/mxyng/quantize-redux
...
cgo quantize
2024-04-09 12:32:53 -07:00
Jeffrey Morgan
5ec12cec6c
update llama.cpp submodule to 1b67731 ( #3561 )
2024-04-09 15:10:17 -04:00
Michael Yang
9502e5661f
cgo quantize
2024-04-08 15:31:08 -07:00
Jeffrey Morgan
63efa075a0
update generate scripts with new LLAMA_CUDA variable, set HIP_PLATFORM to avoid compiler errors ( #3528 )
2024-04-07 19:29:51 -04:00
Michael Yang
be517e491c
no rope parameters
2024-04-05 18:05:27 -07:00
Michael Yang
fc8e108642
Merge pull request #3496 from ollama/mxyng/cmd-r-graph
...
add command-r graph estimate
2024-04-05 12:26:21 -07:00
Daniel Hiltgen
dfe330fa1c
Merge pull request #3488 from mofanke/fix-windows-dll-compress
...
fix dll compress in windows building
2024-04-04 16:12:13 -07:00
Michael Yang
01f77ae25d
add command-r graph estimate
2024-04-04 14:07:24 -07:00
Daniel Hiltgen
36bd967722
Fail fast if mingw missing on windows
2024-04-04 09:51:26 -07:00
mofanke
4de0126719
fix dll compress in windows building
2024-04-04 21:27:33 +08:00
Daniel Hiltgen
e4a7e5b2ca
Fix CI release glitches
...
The subprocess change moved the build directory
arm64 builds weren't setting cross-compilation flags when building on x86
2024-04-03 16:41:40 -07:00
Michael Yang
12e923e158
update graph size estimate
2024-04-03 13:34:12 -07:00
Jeffrey Morgan
cd135317d2
Fix macOS builds on older SDKs ( #3467 )
2024-04-03 10:45:54 -07:00
Michael Yang
4f895d633f
Merge pull request #3466 from ollama/mxyng/head-kv
...
default head_kv to 1
2024-04-03 10:41:00 -07:00
Daniel Hiltgen
464d817824
Merge pull request #3464 from dhiltgen/subprocess
...
Fix numgpu opt miscomparison
2024-04-02 20:10:17 -07:00
Daniel Hiltgen
6589eb8a8c
Revert options as a ref in the server
2024-04-02 16:44:10 -07:00
Michael Yang
90f071c658
default head_kv to 1
2024-04-02 16:37:59 -07:00
Michael Yang
80163ebcb5
fix metal gpu
2024-04-02 16:06:45 -07:00
Daniel Hiltgen
0035e31af8
Bump to b2581
2024-04-02 11:53:07 -07:00
Daniel Hiltgen
0a0e9f3e0f
Apply 01-cache.diff
2024-04-01 16:48:18 -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
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
Jeffrey Morgan
f5ca7f8c8e
add license in file header for vendored llama.cpp code ( #3351 )
2024-03-26 16:23:23 -04:00
Jeffrey Morgan
856b8ec131
remove need for $VSINSTALLDIR since build will fail if ninja cannot be found ( #3350 )
2024-03-26 16:23:16 -04:00
Patrick Devine
1b272d5bcd
change github.com/jmorganca/ollama to github.com/ollama/ollama ( #3347 )
2024-03-26 13:04:17 -07:00
Daniel Hiltgen
8091ef2eeb
Bump llama.cpp to b2527
2024-03-25 13:47:44 -07:00
Daniel Hiltgen
560be5e0b6
Merge pull request #3308 from dhiltgen/bump_more
...
Bump llama.cpp to b2510
2024-03-25 12:56:12 -07:00
Jeremy
dfc6721b20
add support for libcudart.so for CUDA devices (adds Jetson support)
2024-03-25 11:07:44 -04:00
Blake Mizerany
acfa2b9422
llm: prevent race appending to slice ( #3320 )
2024-03-24 11:35:54 -07:00
Daniel Hiltgen
3e30c75f3e
Bump llama.cpp to b2510
2024-03-23 19:55:56 +01:00
