b4f35cb43d
fix(ml): Round OCR max resolution up to a multiple of 32 ( #32263 )
...
fix(ml): round ocr max resolution up to a multiple of 32
2026-10-09 11:05:02 -04:00
655bfdb34f
fix(ml): move logging config to code ( #32220 )
...
This lets it be applied earlier on, so the initializing [version] log
line actually works.
2026-10-08 14:41:09 +00:00
3df38fe7f2
feat(ml): ppocr v6 ( #32147 )
2026-10-06 09:58:20 +02:00
6f356877f5
fix(ml): incorrect outputs on alchemist and battlemage ( #32110 )
...
* add alchemist and battlemage runners
* report the opencl devices on the intel runners
* pad unaligned attention heads for openvino
* build onnxruntime-openvino against an openvino nightly
* note the onnxruntime-openvino the intel runners should test
* update the intel gpu runtime
2026-10-06 06:37:05 +00:00
e89e034515
chore(ml): e2e gpu testing ( #31998 )
2026-10-03 17:15:09 -04:00
524cbb3167
fix(ml): upgrade to openvino 2026.4.1 ( #31996 )
2026-10-02 10:25:39 -04:00
2e7e50ca98
chore(deps): update machine-learning ( #31868 )
...
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
2026-09-30 20:10:43 -04:00
a3c8b359f2
fix(ml): serialize trt-rtx sessions ( #31951 )
2026-09-30 18:56:36 -04:00
b11e400a36
fix(ml): widen empty half-precision arrays ( #31950 )
2026-09-30 18:56:26 -04:00
a9d102234b
chore: version v3.3.0-rc.0
2026-09-28 21:54:03 +00:00
76c239c345
feat(ml): trt-rtx ( #31863 )
...
* trt-rtx
* update dependencies
2026-09-28 21:39:14 +00:00
ab159b8118
fix(ml): stricter typing and request validation ( #31856 )
2026-09-28 14:14:01 -04:00
23f1f87dd0
feat(ml): optimized model inference ( #31750 )
...
* add immich_model
* graph per shape
* decouple ocr from ort
* prepared graphs
* use f16 for accelerators
* preload all shapes
* warm-up pass after preload
* keep precision for fp32 graphs
* ensure padding for text recognition
* optimize postprocessing for fp16 outputs
* handle raw ctc logits
* use f16 output for rknn
* mmap token embedding
* read device metadata
2026-09-25 14:56:44 -04:00
de13aab713
feat(ml): improve model loading ( #31742 )
...
* use recent mimalloc
* jit caching
* bump onnxruntime-gpu
* explicit memory release
* refactor in-memory cache
* configure hf org and revision
* refactor model load
* move orjson response
* disable direct submission for i915
* flush denormals
* update tests and lock
* disable suboptimal fusions
* respect resize_mode
* increase threads for clip text
2026-09-25 14:56:43 -04:00
202015ed95
chore(deps): update machine-learning ( #29123 )
...
* chore(deps): update machine-learning
* bump to 3.13, increase minimum to 3.12
* cleanup
* target 3.12
* fix conditional
* fix builder images
* also update prod rknn
* unused parsers
* drop gold
* bump armnn
---------
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
Co-authored-by: mertalev <101130780+mertalev@users.noreply.github.com >
2026-09-19 02:21:57 +00:00
8ab8ce8cf7
chore(deps): update dependency onnxruntime to v1.29.0 ( #31563 )
2026-09-15 09:35:39 -04:00
cab12ecb51
chore(deps): update dependency onnxruntime-migraphx to v1.27.1 ( #31568 )
2026-09-15 09:34:23 -04:00
c0eed4902a
chore: version v3.2.0 ( #31441 )
...
chore: v3.2.0
2026-09-11 11:02:34 -04:00
26a25f0c3a
fix(ml): race when submitting to rknn execution queue ( #31143 )
2026-09-03 02:13:35 +00:00
c1a88dd9a3
fix(ml): read CLIP model configs as UTF-8 ( #31075 )
2026-08-27 23:04:40 -04:00
e61312084b
chore: version v3.2.0-rc.0
2026-08-27 19:43:45 +00:00
093f5c070a
chore(deps): update dependency pydantic-settings to v2.14.2 [security] ( #29252 )
2026-08-26 10:33:23 -04:00
ef72f1ec5b
chore(deps): update dependency pillow to v12.3.0 [security] ( #30084 )
2026-08-26 10:32:47 -04:00
dc688edb6f
refactor(ml): direct ocr processing ( #31016 )
2026-08-25 23:13:47 -04:00
6acd870371
chore(deps): lock file maintenance (mise) ( #30464 )
...
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
2026-08-25 01:50:02 -04:00
161e15ee2f
fix(deployment): Add huggingface cache directory to Dockerfile ( #30357 )
...
* Add huggingface cache directory to Dockerfile
Updated Dockerfile to include huggingface cache directory.
* Update Dockerfile environment variables and comments
Removed MPLCONFIGDIR environment variable and adjusted comments.
2026-08-08 16:48:33 +00:00
79e61c47c7
refactor(ml): remove insightface dependency ( #30631 )
...
* remove insightface dependency
* update tests
2026-08-08 12:07:30 -04:00
1094b69946
fix(deployment): matplotlib in rootless deployments ( #30328 )
2026-07-28 10:57:21 -04:00
8aa95c6747
chore: version v3.1.0
2026-07-27 18:20:33 +00:00
cd308ad930
chore: version v3.0.3
2026-07-14 21:10:19 +00:00
443b856ac3
chore: version v3.0.2
2026-07-09 17:33:21 +00:00
628de417e9
chore(deps): lock file maintenance (mise) ( #29637 )
2026-07-06 14:52:48 +02:00
f77c8a4699
chore: version v3.0.1
2026-07-02 19:24:13 +00:00
05d838b560
chore: version v3.0.0
2026-06-30 20:03:47 +00:00
ac74bca18b
chore: version v3.0.0-rc.4
2026-06-27 13:14:47 +00:00
09d4a6815f
chore: version v3.0.0-rc.3
2026-06-26 16:49:02 +00:00
38920fc4ca
chore: version v3.0.0-rc.2
2026-06-18 23:20:14 +00:00
9a3071ae5c
chore(deps): lock file maintenance machine-learning (mise) ( #29152 )
...
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
2026-06-17 19:34:45 -04:00
8d30cfa2ce
chore(ml): update intel graphics compiler and compute runtime ( #29163 )
...
chore(ml): update Intel graphics compiler and compute runtime to latest versions
2026-06-16 23:09:53 +00:00
c9aa9ba711
chore: version v3.0.0-rc.1
2026-06-16 19:50:41 +00:00
e4dbe777a0
chore: version v3.0.0-rc.0
2026-06-12 14:55:39 +00:00
563cff26bf
chore(deps): update machine-learning ( #28934 )
...
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
2026-06-11 14:40:20 -04:00
a665cec920
feat(ml): update Intel graphics compiler and compute runtime ( #28924 )
...
feat(ml): update Intel graphics compiler and compute runtime to latest versions
2026-06-09 11:08:03 -04:00
7f611d9031
test: fix tests when OpenVINO provider is available ( #28802 )
...
mocking `onnxruntime.get_available_providers()` to always use the CPU EP.
2026-06-03 20:52:08 -04:00
408e1180ca
chore(deps): update machine-learning ( #28239 )
...
* chore(deps): update machine-learning
* fix typing
* fix deprecation log
* no control socket
---------
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
Co-authored-by: mertalev <101130780+mertalev@users.noreply.github.com >
2026-06-02 16:44:50 +00:00
a838167f11
fix(ml): pass model_root_dir to OcrOptions for RapidOCR compatibility ( #28610 )
...
* fix(ml): pass model_root_dir to OcrOptions for RapidOCR compatibility
Fix a TypeError (Path(None)) when the OCR model is invoked, caused by an upstream change in RapidOCR v3.8.1 (RapidAI/RapidOCR@8ea9626 ).
RapidOCR now internally calls `Path(cfg.get("model_root_dir"))`. Since `model_root_dir` was missing from `OcrOptions`, it evaluated to `None` and triggered a `TypeError: argument should be a str or an os.PathLike`.
This fix adds the missing `model_root_dir` argument to prevent the error.
Ref: #28331
* fix(ml-test): update OCR tests for RapidOCR schema change
* chore(ml-test): remove unused `cache_dir` parameter from `TextRecognizer`
* Revert "chore(ml-test): remove unused `cache_dir` parameter from `TextRecognizer`"
This reverts commit 007ad7b3f2
2026-05-28 22:54:04 -04:00
53a24783f5
fix(ml): stabilize MIGraphX inference ( #28444 )
...
* fix: stabilize ROCm MIGraphX inference
Serialize MIGraphX session runs so lazy compiles cannot overlap within a worker.
Use a fixed face-recognition batch size for MIGraphX to avoid compiling a new program for each detected face count.
* fix(ml): increase ROCm worker timeout
* fix(ml): narrow MIGraphX compile locking
* docs: format environment variables table
* docs: apply prettier to environment variables table
2026-05-26 18:41:56 +00:00
5386b62dc4
chore(ml): allow insightface 1.x ( #28595 )
...
* chore(ml): allow insightface 1.x
The new insightface 1.0 release appears to have no breaking code changes nor relevant license changes ([before](https://github.com/deepinsight/insightface/blob/2a78baec428354883e0cda39c54b555a5ed8358a/README.md ), [after](https://github.com/deepinsight/insightface/blob/70f3269ea628d0658c5723976944c9de414e96f8/README.md ), c.f. https://github.com/immich-app/immich/blob/fd7ddfef54cdf2b6256c4fc08bc5ff3f86176775/machine-learning/README.md ), and it works on my machine.
* Update uv.lock
* please excuse my incompetence
2026-05-25 12:32:50 -04:00
a5a7380a26
feat: use lockfile for mise tools ( #28503 )
2026-05-20 11:37:33 +00:00
8c8dc9d32f
chore(ml)!: remove deprecated envs ( #28326 )
...
remove deprecated envs
2026-05-09 22:40:05 +00:00
Ștefan Muraru