nexaai 1.0.29__cp310-cp310-macosx_14_0_universal2.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- nexaai/__init__.py +99 -0
- nexaai/_stub.cpython-310-darwin.so +0 -0
- nexaai/_version.py +4 -0
- nexaai/asr.py +68 -0
- nexaai/asr_impl/__init__.py +0 -0
- nexaai/asr_impl/mlx_asr_impl.py +93 -0
- nexaai/asr_impl/pybind_asr_impl.py +127 -0
- nexaai/base.py +39 -0
- nexaai/binds/__init__.py +7 -0
- nexaai/binds/asr_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/common_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/cpu_gpu/libggml-base.dylib +0 -0
- nexaai/binds/cpu_gpu/libggml-cpu.so +0 -0
- nexaai/binds/cpu_gpu/libggml-metal.so +0 -0
- nexaai/binds/cpu_gpu/libggml.dylib +0 -0
- nexaai/binds/cpu_gpu/libmtmd.dylib +0 -0
- nexaai/binds/cpu_gpu/libnexa_cpu_gpu.dylib +0 -0
- nexaai/binds/cpu_gpu/libnexa_plugin.dylib +0 -0
- nexaai/binds/cv_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/diarize_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/embedder_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/libnexa_bridge.dylib +0 -0
- nexaai/binds/llm_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/metal/libnexa_plugin.dylib +0 -0
- nexaai/binds/metal/py-lib/ml.py +888 -0
- nexaai/binds/metal/py-lib/mlx_audio/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/__init__.py +5 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/bigvgan/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/bigvgan/activation.py +51 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/bigvgan/amp.py +96 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/bigvgan/bigvgan.py +149 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/bigvgan/conv.py +114 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/bigvgan/resample.py +177 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/descript/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/descript/base.py +228 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/descript/dac.py +285 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/descript/nn/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/descript/nn/layers.py +129 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/descript/nn/quantize.py +149 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/encodec/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/encodec/encodec.py +777 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/mimi.py +286 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/modules/__init__.py +20 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/modules/conv.py +398 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/modules/kv_cache.py +199 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/modules/quantization.py +179 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/modules/seanet.py +314 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/mimi/modules/transformer.py +256 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/s3/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/s3/model.py +260 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/s3/model_v2.py +383 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/s3/utils.py +122 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/snac/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/snac/attention.py +97 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/snac/layers.py +306 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/snac/snac.py +154 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/snac/vq.py +135 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/vocos/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/vocos/mel.py +33 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/models/vocos/vocos.py +359 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_bigvgan.py +54 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_descript.py +109 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_encodec.py +58 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_mimi.py +22 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_s3.py +25 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_snac.py +40 -0
- nexaai/binds/metal/py-lib/mlx_audio/codec/tests/test_vocos.py +93 -0
- nexaai/binds/metal/py-lib/mlx_audio/server.py +525 -0
- nexaai/binds/metal/py-lib/mlx_audio/sts/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/sts/tests/test_voice_pipeline.py +156 -0
- nexaai/binds/metal/py-lib/mlx_audio/sts/voice_pipeline.py +327 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/generate.py +174 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/alignment.py +248 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/attention.py +187 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/audio.py +76 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/conformer.py +331 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/ctc.py +34 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/parakeet.py +604 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/rnnt.py +157 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/parakeet/tokenizer.py +2 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/wav2vec/feature_extractor.py +757 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/wav2vec/wav2vec.py +738 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/audio.py +82 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/decoding.py +742 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/timing.py +329 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/tokenizer.py +398 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/whisper.py +862 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/models/whisper/writers.py +268 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/tests/test_models.py +381 -0
- nexaai/binds/metal/py-lib/mlx_audio/stt/utils.py +195 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/audio_player.py +120 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/convert.py +71 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/generate.py +449 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/bark/__init__.py +4 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/bark/bark.py +528 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/bark/isftnet.py +12 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/bark/pipeline.py +442 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/base.py +84 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/dia/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/dia/audio.py +287 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/dia/config.py +256 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/dia/dia.py +592 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/dia/layers.py +870 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/__init__.py +3 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/attention.py +180 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/bigvgan.py +124 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/conformer.py +247 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/ecapa_tdnn/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/ecapa_tdnn/asp.py +59 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/ecapa_tdnn/ecapa_tdnn.py +91 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/ecapa_tdnn/se_res2net.py +132 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/ecapa_tdnn/tdnn.py +42 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/gpt2.py +38 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/indextts.py +412 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/mel.py +37 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/normalize.py +294 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/indextts/perceiver.py +62 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/interpolate.py +108 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/kokoro/__init__.py +4 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/kokoro/istftnet.py +979 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/kokoro/kokoro.py +331 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/kokoro/modules.py +659 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/kokoro/pipeline.py +453 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/kokoro/voice.py +113 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/llama/__init__.py +3 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/llama/llama.py +324 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/outetts/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/outetts/audio_processor.py +351 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/outetts/dac_interface.py +162 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/outetts/outetts.py +255 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/outetts/prompt_processor.py +181 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/outetts/tokens.py +36 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/sesame/__init__.py +3 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/sesame/attention.py +195 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/sesame/sesame.py +633 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/sesame/watermarking.py +105 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/audio_tokenizer.py +138 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/bicodec.py +269 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/blocks/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/blocks/sampler.py +111 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/encoder_decoder/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/encoder_decoder/feat_decoder.py +120 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/encoder_decoder/feat_encoder.py +136 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/encoder_decoder/wave_generator.py +113 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/finite_scalar_quantization.py +238 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/residual.py +209 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/residual_fsq.py +309 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/speaker/__init__.py +1 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/speaker/ecapa_tdnn.py +283 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/speaker/perceiver_encoder.py +326 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/speaker/pooling_layers.py +297 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/modules/speaker/speaker_encoder.py +155 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/spark.py +382 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/utils/audio.py +220 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/utils/file.py +221 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/models/spark/utils/token_parser.py +181 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/tests/__init__.py +0 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/tests/test_base.py +66 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/tests/test_convert.py +173 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/tests/test_interpolate.py +88 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/tests/test_models.py +974 -0
- nexaai/binds/metal/py-lib/mlx_audio/tts/utils.py +337 -0
- nexaai/binds/metal/py-lib/mlx_audio/utils.py +237 -0
- nexaai/binds/metal/py-lib/mlx_audio/version.py +1 -0
- nexaai/binds/metal/py-lib/profiling.py +239 -0
- nexaai/binds/nexaml/libfftw3.3.dylib +0 -0
- nexaai/binds/nexaml/libfftw3f.3.dylib +0 -0
- nexaai/binds/nexaml/libggml-base.dylib +0 -0
- nexaai/binds/nexaml/libggml-cpu.so +0 -0
- nexaai/binds/nexaml/libggml-metal.so +0 -0
- nexaai/binds/nexaml/libggml.dylib +0 -0
- nexaai/binds/nexaml/libmp3lame.0.dylib +0 -0
- nexaai/binds/nexaml/libmpg123.0.dylib +0 -0
- nexaai/binds/nexaml/libnexa-mm-process.dylib +0 -0
- nexaai/binds/nexaml/libnexa-sampling.dylib +0 -0
- nexaai/binds/nexaml/libnexa_plugin.dylib +0 -0
- nexaai/binds/nexaml/libnexaproc.dylib +0 -0
- nexaai/binds/nexaml/libomp.dylib +0 -0
- nexaai/binds/nexaml/libqwen3-vl.dylib +0 -0
- nexaai/binds/nexaml/libqwen3vl-vision.dylib +0 -0
- nexaai/binds/rerank_bind.cpython-310-darwin.so +0 -0
- nexaai/binds/vlm_bind.cpython-310-darwin.so +0 -0
- nexaai/common.py +106 -0
- nexaai/cv.py +95 -0
- nexaai/cv_impl/__init__.py +0 -0
- nexaai/cv_impl/mlx_cv_impl.py +91 -0
- nexaai/cv_impl/pybind_cv_impl.py +124 -0
- nexaai/diarize.py +80 -0
- nexaai/diarize_impl/__init__.py +1 -0
- nexaai/diarize_impl/pybind_diarize_impl.py +125 -0
- nexaai/embedder.py +73 -0
- nexaai/embedder_impl/__init__.py +0 -0
- nexaai/embedder_impl/mlx_embedder_impl.py +118 -0
- nexaai/embedder_impl/pybind_embedder_impl.py +96 -0
- nexaai/image_gen.py +141 -0
- nexaai/image_gen_impl/__init__.py +0 -0
- nexaai/image_gen_impl/mlx_image_gen_impl.py +292 -0
- nexaai/image_gen_impl/pybind_image_gen_impl.py +85 -0
- nexaai/llm.py +98 -0
- nexaai/llm_impl/__init__.py +0 -0
- nexaai/llm_impl/mlx_llm_impl.py +271 -0
- nexaai/llm_impl/pybind_llm_impl.py +238 -0
- nexaai/log.py +92 -0
- nexaai/mlx_backend/asr/__init__.py +12 -0
- nexaai/mlx_backend/asr/interface.py +122 -0
- nexaai/mlx_backend/common/__init__.py +0 -0
- nexaai/mlx_backend/common/utils.py +25 -0
- nexaai/mlx_backend/cv/__init__.py +0 -0
- nexaai/mlx_backend/cv/generate.py +195 -0
- nexaai/mlx_backend/cv/interface.py +162 -0
- nexaai/mlx_backend/cv/main.py +81 -0
- nexaai/mlx_backend/cv/modeling/pp_ocr_v4.py +1736 -0
- nexaai/mlx_backend/embedding/__init__.py +0 -0
- nexaai/mlx_backend/embedding/generate.py +333 -0
- nexaai/mlx_backend/embedding/interface.py +617 -0
- nexaai/mlx_backend/embedding/main.py +173 -0
- nexaai/mlx_backend/embedding/modeling/__init__.py +0 -0
- nexaai/mlx_backend/embedding/modeling/nexa_jina_v2.py +399 -0
- nexaai/mlx_backend/image_gen/__init__.py +1 -0
- nexaai/mlx_backend/image_gen/generate_sd.py +244 -0
- nexaai/mlx_backend/image_gen/interface.py +82 -0
- nexaai/mlx_backend/image_gen/main.py +281 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/__init__.py +306 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/clip.py +116 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/config.py +65 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/model_io.py +386 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/sampler.py +105 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/tokenizer.py +100 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/unet.py +460 -0
- nexaai/mlx_backend/image_gen/stable_diffusion/vae.py +274 -0
- nexaai/mlx_backend/llm/__init__.py +0 -0
- nexaai/mlx_backend/llm/generate.py +149 -0
- nexaai/mlx_backend/llm/interface.py +764 -0
- nexaai/mlx_backend/llm/main.py +68 -0
- nexaai/mlx_backend/ml.py +888 -0
- nexaai/mlx_backend/mlx_audio/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/codec/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/__init__.py +5 -0
- nexaai/mlx_backend/mlx_audio/codec/models/bigvgan/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/bigvgan/activation.py +51 -0
- nexaai/mlx_backend/mlx_audio/codec/models/bigvgan/amp.py +96 -0
- nexaai/mlx_backend/mlx_audio/codec/models/bigvgan/bigvgan.py +149 -0
- nexaai/mlx_backend/mlx_audio/codec/models/bigvgan/conv.py +114 -0
- nexaai/mlx_backend/mlx_audio/codec/models/bigvgan/resample.py +177 -0
- nexaai/mlx_backend/mlx_audio/codec/models/descript/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/descript/base.py +228 -0
- nexaai/mlx_backend/mlx_audio/codec/models/descript/dac.py +285 -0
- nexaai/mlx_backend/mlx_audio/codec/models/descript/nn/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/descript/nn/layers.py +129 -0
- nexaai/mlx_backend/mlx_audio/codec/models/descript/nn/quantize.py +149 -0
- nexaai/mlx_backend/mlx_audio/codec/models/encodec/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/encodec/encodec.py +777 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/mimi.py +286 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/modules/__init__.py +20 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/modules/conv.py +398 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/modules/kv_cache.py +199 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/modules/quantization.py +179 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/modules/seanet.py +314 -0
- nexaai/mlx_backend/mlx_audio/codec/models/mimi/modules/transformer.py +256 -0
- nexaai/mlx_backend/mlx_audio/codec/models/s3/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/s3/model.py +260 -0
- nexaai/mlx_backend/mlx_audio/codec/models/s3/model_v2.py +383 -0
- nexaai/mlx_backend/mlx_audio/codec/models/s3/utils.py +122 -0
- nexaai/mlx_backend/mlx_audio/codec/models/snac/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/snac/attention.py +97 -0
- nexaai/mlx_backend/mlx_audio/codec/models/snac/layers.py +306 -0
- nexaai/mlx_backend/mlx_audio/codec/models/snac/snac.py +154 -0
- nexaai/mlx_backend/mlx_audio/codec/models/snac/vq.py +135 -0
- nexaai/mlx_backend/mlx_audio/codec/models/vocos/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/codec/models/vocos/mel.py +33 -0
- nexaai/mlx_backend/mlx_audio/codec/models/vocos/vocos.py +359 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_bigvgan.py +54 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_descript.py +109 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_encodec.py +58 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_mimi.py +22 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_s3.py +25 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_snac.py +40 -0
- nexaai/mlx_backend/mlx_audio/codec/tests/test_vocos.py +93 -0
- nexaai/mlx_backend/mlx_audio/server.py +525 -0
- nexaai/mlx_backend/mlx_audio/sts/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/sts/tests/test_voice_pipeline.py +156 -0
- nexaai/mlx_backend/mlx_audio/sts/voice_pipeline.py +327 -0
- nexaai/mlx_backend/mlx_audio/stt/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/stt/generate.py +174 -0
- nexaai/mlx_backend/mlx_audio/stt/models/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/alignment.py +248 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/attention.py +187 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/audio.py +76 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/conformer.py +331 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/ctc.py +34 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/parakeet.py +604 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/rnnt.py +157 -0
- nexaai/mlx_backend/mlx_audio/stt/models/parakeet/tokenizer.py +2 -0
- nexaai/mlx_backend/mlx_audio/stt/models/wav2vec/feature_extractor.py +757 -0
- nexaai/mlx_backend/mlx_audio/stt/models/wav2vec/wav2vec.py +738 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/audio.py +82 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/decoding.py +742 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/timing.py +329 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/tokenizer.py +398 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/whisper.py +862 -0
- nexaai/mlx_backend/mlx_audio/stt/models/whisper/writers.py +268 -0
- nexaai/mlx_backend/mlx_audio/stt/tests/test_models.py +381 -0
- nexaai/mlx_backend/mlx_audio/stt/utils.py +195 -0
- nexaai/mlx_backend/mlx_audio/tts/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/tts/audio_player.py +120 -0
- nexaai/mlx_backend/mlx_audio/tts/convert.py +71 -0
- nexaai/mlx_backend/mlx_audio/tts/generate.py +449 -0
- nexaai/mlx_backend/mlx_audio/tts/models/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/tts/models/bark/__init__.py +4 -0
- nexaai/mlx_backend/mlx_audio/tts/models/bark/bark.py +528 -0
- nexaai/mlx_backend/mlx_audio/tts/models/bark/isftnet.py +12 -0
- nexaai/mlx_backend/mlx_audio/tts/models/bark/pipeline.py +442 -0
- nexaai/mlx_backend/mlx_audio/tts/models/base.py +84 -0
- nexaai/mlx_backend/mlx_audio/tts/models/dia/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/tts/models/dia/audio.py +287 -0
- nexaai/mlx_backend/mlx_audio/tts/models/dia/config.py +256 -0
- nexaai/mlx_backend/mlx_audio/tts/models/dia/dia.py +592 -0
- nexaai/mlx_backend/mlx_audio/tts/models/dia/layers.py +870 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/__init__.py +3 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/attention.py +180 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/bigvgan.py +124 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/conformer.py +247 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/ecapa_tdnn/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/ecapa_tdnn/asp.py +59 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/ecapa_tdnn/ecapa_tdnn.py +91 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/ecapa_tdnn/se_res2net.py +132 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/ecapa_tdnn/tdnn.py +42 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/gpt2.py +38 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/indextts.py +412 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/mel.py +37 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/normalize.py +294 -0
- nexaai/mlx_backend/mlx_audio/tts/models/indextts/perceiver.py +62 -0
- nexaai/mlx_backend/mlx_audio/tts/models/interpolate.py +108 -0
- nexaai/mlx_backend/mlx_audio/tts/models/kokoro/__init__.py +4 -0
- nexaai/mlx_backend/mlx_audio/tts/models/kokoro/istftnet.py +979 -0
- nexaai/mlx_backend/mlx_audio/tts/models/kokoro/kokoro.py +331 -0
- nexaai/mlx_backend/mlx_audio/tts/models/kokoro/modules.py +659 -0
- nexaai/mlx_backend/mlx_audio/tts/models/kokoro/pipeline.py +453 -0
- nexaai/mlx_backend/mlx_audio/tts/models/kokoro/voice.py +113 -0
- nexaai/mlx_backend/mlx_audio/tts/models/llama/__init__.py +3 -0
- nexaai/mlx_backend/mlx_audio/tts/models/llama/llama.py +324 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/audio_processor.py +351 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/dac_interface.py +162 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/default_speaker.json +461 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/outetts.py +255 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/prompt_processor.py +181 -0
- nexaai/mlx_backend/mlx_audio/tts/models/outetts/tokens.py +36 -0
- nexaai/mlx_backend/mlx_audio/tts/models/sesame/__init__.py +3 -0
- nexaai/mlx_backend/mlx_audio/tts/models/sesame/attention.py +195 -0
- nexaai/mlx_backend/mlx_audio/tts/models/sesame/sesame.py +633 -0
- nexaai/mlx_backend/mlx_audio/tts/models/sesame/watermarking.py +105 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/audio_tokenizer.py +138 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/bicodec.py +269 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/blocks/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/blocks/sampler.py +111 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/encoder_decoder/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/encoder_decoder/feat_decoder.py +120 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/encoder_decoder/feat_encoder.py +136 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/encoder_decoder/wave_generator.py +113 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/finite_scalar_quantization.py +238 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/residual.py +209 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/residual_fsq.py +309 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/speaker/__init__.py +1 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/speaker/ecapa_tdnn.py +283 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/speaker/perceiver_encoder.py +326 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/speaker/pooling_layers.py +297 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/modules/speaker/speaker_encoder.py +155 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/spark.py +382 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/utils/audio.py +220 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/utils/file.py +221 -0
- nexaai/mlx_backend/mlx_audio/tts/models/spark/utils/token_parser.py +181 -0
- nexaai/mlx_backend/mlx_audio/tts/tests/__init__.py +0 -0
- nexaai/mlx_backend/mlx_audio/tts/tests/test_base.py +66 -0
- nexaai/mlx_backend/mlx_audio/tts/tests/test_convert.py +173 -0
- nexaai/mlx_backend/mlx_audio/tts/tests/test_interpolate.py +88 -0
- nexaai/mlx_backend/mlx_audio/tts/tests/test_models.py +974 -0
- nexaai/mlx_backend/mlx_audio/tts/utils.py +337 -0
- nexaai/mlx_backend/mlx_audio/utils.py +237 -0
- nexaai/mlx_backend/mlx_audio/version.py +1 -0
- nexaai/mlx_backend/profiling.py +239 -0
- nexaai/mlx_backend/rerank/__init__.py +0 -0
- nexaai/mlx_backend/rerank/generate.py +174 -0
- nexaai/mlx_backend/rerank/interface.py +287 -0
- nexaai/mlx_backend/rerank/main.py +127 -0
- nexaai/mlx_backend/rerank/modeling/__init__.py +0 -0
- nexaai/mlx_backend/rerank/modeling/nexa_jina_rerank.py +330 -0
- nexaai/mlx_backend/sd/__init__.py +1 -0
- nexaai/mlx_backend/sd/interface.py +362 -0
- nexaai/mlx_backend/sd/main.py +286 -0
- nexaai/mlx_backend/sd/modeling/__init__.py +306 -0
- nexaai/mlx_backend/sd/modeling/clip.py +116 -0
- nexaai/mlx_backend/sd/modeling/config.py +65 -0
- nexaai/mlx_backend/sd/modeling/model_io.py +385 -0
- nexaai/mlx_backend/sd/modeling/sampler.py +105 -0
- nexaai/mlx_backend/sd/modeling/tokenizer.py +100 -0
- nexaai/mlx_backend/sd/modeling/unet.py +460 -0
- nexaai/mlx_backend/sd/modeling/vae.py +274 -0
- nexaai/mlx_backend/tts/__init__.py +12 -0
- nexaai/mlx_backend/tts/interface.py +276 -0
- nexaai/mlx_backend/vlm/__init__.py +3 -0
- nexaai/mlx_backend/vlm/generate.py +572 -0
- nexaai/mlx_backend/vlm/generate_qwen3_vl.py +374 -0
- nexaai/mlx_backend/vlm/generate_qwen3_vl_moe.py +259 -0
- nexaai/mlx_backend/vlm/interface.py +559 -0
- nexaai/mlx_backend/vlm/main.py +365 -0
- nexaai/mlx_backend/vlm/modeling/__init__.py +0 -0
- nexaai/mlx_backend/vlm/modeling/convert.py +68 -0
- nexaai/mlx_backend/vlm/modeling/models/__init__.py +0 -0
- nexaai/mlx_backend/vlm/modeling/models/aya_vision/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/aya_vision/aya_vision.py +193 -0
- nexaai/mlx_backend/vlm/modeling/models/aya_vision/interpolate.py +186 -0
- nexaai/mlx_backend/vlm/modeling/models/aya_vision/language.py +233 -0
- nexaai/mlx_backend/vlm/modeling/models/aya_vision/vision.py +503 -0
- nexaai/mlx_backend/vlm/modeling/models/base.py +202 -0
- nexaai/mlx_backend/vlm/modeling/models/cache.py +230 -0
- nexaai/mlx_backend/vlm/modeling/models/deepseek_vl_v2/__init__.py +10 -0
- nexaai/mlx_backend/vlm/modeling/models/deepseek_vl_v2/conversation.py +264 -0
- nexaai/mlx_backend/vlm/modeling/models/deepseek_vl_v2/deepseek_vl_v2.py +472 -0
- nexaai/mlx_backend/vlm/modeling/models/deepseek_vl_v2/language.py +591 -0
- nexaai/mlx_backend/vlm/modeling/models/deepseek_vl_v2/processing_deepsek_vl_v2.py +526 -0
- nexaai/mlx_backend/vlm/modeling/models/deepseek_vl_v2/vision.py +356 -0
- nexaai/mlx_backend/vlm/modeling/models/florence2/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/florence2/florence2.py +366 -0
- nexaai/mlx_backend/vlm/modeling/models/florence2/language.py +488 -0
- nexaai/mlx_backend/vlm/modeling/models/florence2/vision.py +591 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3/gemma3.py +213 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3/language.py +315 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3/vision.py +238 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3n/__init__.py +2 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3n/audio.py +1038 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3n/config.py +139 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3n/gemma3n.py +322 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3n/language.py +629 -0
- nexaai/mlx_backend/vlm/modeling/models/gemma3n/vision.py +1022 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics2/__init__.py +9 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics2/idefics2.py +294 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics2/language.py +191 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics2/vision.py +267 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics3/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics3/idefics3.py +175 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics3/language.py +192 -0
- nexaai/mlx_backend/vlm/modeling/models/idefics3/vision.py +233 -0
- nexaai/mlx_backend/vlm/modeling/models/internvl_chat/__init__.py +9 -0
- nexaai/mlx_backend/vlm/modeling/models/internvl_chat/internvl_chat.py +140 -0
- nexaai/mlx_backend/vlm/modeling/models/internvl_chat/language.py +220 -0
- nexaai/mlx_backend/vlm/modeling/models/internvl_chat/processor.py +393 -0
- nexaai/mlx_backend/vlm/modeling/models/internvl_chat/vision.py +293 -0
- nexaai/mlx_backend/vlm/modeling/models/kernels.py +307 -0
- nexaai/mlx_backend/vlm/modeling/models/kimi_vl/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/kimi_vl/kimi_vl.py +143 -0
- nexaai/mlx_backend/vlm/modeling/models/kimi_vl/language.py +509 -0
- nexaai/mlx_backend/vlm/modeling/models/kimi_vl/vision.py +522 -0
- nexaai/mlx_backend/vlm/modeling/models/llama4/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/llama4/language.py +386 -0
- nexaai/mlx_backend/vlm/modeling/models/llama4/llama4.py +138 -0
- nexaai/mlx_backend/vlm/modeling/models/llama4/vision.py +560 -0
- nexaai/mlx_backend/vlm/modeling/models/llava/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/llava/language.py +240 -0
- nexaai/mlx_backend/vlm/modeling/models/llava/llava.py +153 -0
- nexaai/mlx_backend/vlm/modeling/models/llava/vision.py +259 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_bunny/__init__.py +9 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_bunny/language.py +236 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_bunny/llava_bunny.py +256 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_bunny/vision.py +303 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_next/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_next/language.py +230 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_next/llava_next.py +160 -0
- nexaai/mlx_backend/vlm/modeling/models/llava_next/vision.py +243 -0
- nexaai/mlx_backend/vlm/modeling/models/mistral3/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/mistral3/mistral3.py +283 -0
- nexaai/mlx_backend/vlm/modeling/models/mllama/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/mllama/language.py +416 -0
- nexaai/mlx_backend/vlm/modeling/models/mllama/mllama.py +172 -0
- nexaai/mlx_backend/vlm/modeling/models/mllama/vision.py +499 -0
- nexaai/mlx_backend/vlm/modeling/models/molmo/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/molmo/language.py +243 -0
- nexaai/mlx_backend/vlm/modeling/models/molmo/molmo.py +133 -0
- nexaai/mlx_backend/vlm/modeling/models/molmo/vision.py +465 -0
- nexaai/mlx_backend/vlm/modeling/models/multi_modality/__init__.py +10 -0
- nexaai/mlx_backend/vlm/modeling/models/multi_modality/language.py +230 -0
- nexaai/mlx_backend/vlm/modeling/models/multi_modality/multi_modality.py +385 -0
- nexaai/mlx_backend/vlm/modeling/models/multi_modality/sam.py +557 -0
- nexaai/mlx_backend/vlm/modeling/models/multi_modality/vision.py +526 -0
- nexaai/mlx_backend/vlm/modeling/models/paligemma/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/paligemma/language.py +282 -0
- nexaai/mlx_backend/vlm/modeling/models/paligemma/paligemma.py +160 -0
- nexaai/mlx_backend/vlm/modeling/models/paligemma/vision.py +242 -0
- nexaai/mlx_backend/vlm/modeling/models/phi3_v/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/phi3_v/language.py +21 -0
- nexaai/mlx_backend/vlm/modeling/models/phi3_v/phi3_v.py +243 -0
- nexaai/mlx_backend/vlm/modeling/models/phi3_v/su_rope.py +71 -0
- nexaai/mlx_backend/vlm/modeling/models/phi3_v/vision.py +324 -0
- nexaai/mlx_backend/vlm/modeling/models/pixtral/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/pixtral/language.py +229 -0
- nexaai/mlx_backend/vlm/modeling/models/pixtral/pixtral.py +161 -0
- nexaai/mlx_backend/vlm/modeling/models/pixtral/vision.py +320 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_5_vl/__init__.py +2 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_5_vl/config.py +108 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_5_vl/language.py +490 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_5_vl/qwen2_5_vl.py +168 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_5_vl/vision.py +414 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/__init__.py +2 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/config.py +104 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/language.py +490 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/qwen2_vl.py +167 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/vision.py +312 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/__init__.py +0 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/base.py +117 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/cache.py +531 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/generate.py +701 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/rope_utils.py +255 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/sample_utils.py +303 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/tokenizer_utils.py +407 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/processor.py +476 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/qwen3vl.py +1262 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/__init__.py +0 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/base.py +117 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/cache.py +531 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/generate.py +701 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/rope_utils.py +255 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/sample_utils.py +303 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/tokenizer_utils.py +407 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/processor.py +476 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/qwen3vl_moe.py +1308 -0
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/switch_layers.py +210 -0
- nexaai/mlx_backend/vlm/modeling/models/smolvlm/__init__.py +8 -0
- nexaai/mlx_backend/vlm/modeling/models/smolvlm/smolvlm.py +62 -0
- nexaai/mlx_backend/vlm/modeling/processing_qwen2_5_vl.py +209 -0
- nexaai/mlx_backend/vlm/modeling/processing_qwen2_vl.py +215 -0
- nexaai/mlx_backend/vlm/modeling/prompt_utils.py +474 -0
- nexaai/mlx_backend/vlm/modeling/sample_utils.py +39 -0
- nexaai/mlx_backend/vlm/modeling/tokenizer_utils.py +344 -0
- nexaai/mlx_backend/vlm/modeling/trainer/__init__.py +9 -0
- nexaai/mlx_backend/vlm/modeling/trainer/lora.py +70 -0
- nexaai/mlx_backend/vlm/modeling/trainer/trainer.py +296 -0
- nexaai/mlx_backend/vlm/modeling/trainer/utils.py +160 -0
- nexaai/mlx_backend/vlm/modeling/utils.py +928 -0
- nexaai/rerank.py +57 -0
- nexaai/rerank_impl/__init__.py +0 -0
- nexaai/rerank_impl/mlx_rerank_impl.py +94 -0
- nexaai/rerank_impl/pybind_rerank_impl.py +136 -0
- nexaai/runtime.py +68 -0
- nexaai/runtime_error.py +24 -0
- nexaai/tts.py +75 -0
- nexaai/tts_impl/__init__.py +0 -0
- nexaai/tts_impl/mlx_tts_impl.py +94 -0
- nexaai/tts_impl/pybind_tts_impl.py +43 -0
- nexaai/utils/decode.py +18 -0
- nexaai/utils/manifest_utils.py +531 -0
- nexaai/utils/model_manager.py +1745 -0
- nexaai/utils/model_types.py +49 -0
- nexaai/utils/progress_tracker.py +389 -0
- nexaai/utils/quantization_utils.py +245 -0
- nexaai/vlm.py +130 -0
- nexaai/vlm_impl/__init__.py +0 -0
- nexaai/vlm_impl/mlx_vlm_impl.py +259 -0
- nexaai/vlm_impl/pybind_vlm_impl.py +275 -0
- nexaai-1.0.29.dist-info/METADATA +35 -0
- nexaai-1.0.29.dist-info/RECORD +580 -0
- nexaai-1.0.29.dist-info/WHEEL +5 -0
- nexaai-1.0.29.dist-info/top_level.txt +1 -0
nexaai/image_gen.py
ADDED
|
@@ -0,0 +1,141 @@
|
|
|
1
|
+
from typing import List, Optional, Union
|
|
2
|
+
from abc import abstractmethod
|
|
3
|
+
from dataclasses import dataclass
|
|
4
|
+
|
|
5
|
+
from nexaai.base import BaseModel
|
|
6
|
+
from nexaai.common import PluginID
|
|
7
|
+
|
|
8
|
+
|
|
9
|
+
@dataclass
|
|
10
|
+
class Image:
|
|
11
|
+
"""Image data structure."""
|
|
12
|
+
data: List[float] # width × height × channels
|
|
13
|
+
width: int
|
|
14
|
+
height: int
|
|
15
|
+
channels: int # 3 = RGB, 4 = RGBA
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
@dataclass
|
|
19
|
+
class ImageSamplerConfig:
|
|
20
|
+
"""Configuration for image sampling."""
|
|
21
|
+
method: str = "ddim"
|
|
22
|
+
steps: int = 20
|
|
23
|
+
guidance_scale: float = 7.5
|
|
24
|
+
eta: float = 0.0
|
|
25
|
+
seed: int = -1 # –1 for random
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
@dataclass
|
|
29
|
+
class ImageGenerationConfig:
|
|
30
|
+
"""Configuration for image generation."""
|
|
31
|
+
prompts: Union[str, List[str]]
|
|
32
|
+
negative_prompts: Optional[Union[str, List[str]]] = None
|
|
33
|
+
height: int = 512
|
|
34
|
+
width: int = 512
|
|
35
|
+
sampler_config: Optional[ImageSamplerConfig] = None
|
|
36
|
+
lora_id: int = -1 # –1 for none
|
|
37
|
+
init_image: Optional[Image] = None
|
|
38
|
+
strength: float = 1.0
|
|
39
|
+
n_images: int = 1
|
|
40
|
+
n_rows: int = 1
|
|
41
|
+
decoding_batch_size: int = 1
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
@dataclass
|
|
45
|
+
class SchedulerConfig:
|
|
46
|
+
"""Configuration for diffusion scheduler."""
|
|
47
|
+
type: str = "ddim"
|
|
48
|
+
num_train_timesteps: int = 1000
|
|
49
|
+
steps_offset: int = 0 # An offset added to the inference steps
|
|
50
|
+
beta_start: float = 0.00085
|
|
51
|
+
beta_end: float = 0.012
|
|
52
|
+
beta_schedule: str = "scaled_linear"
|
|
53
|
+
prediction_type: str = "epsilon"
|
|
54
|
+
timestep_type: str = "discrete"
|
|
55
|
+
timestep_spacing: str = "linspace"
|
|
56
|
+
interpolation_type: str = "linear"
|
|
57
|
+
config_path: Optional[str] = None
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
class ImageGen(BaseModel):
|
|
61
|
+
"""Abstract base class for image generation models."""
|
|
62
|
+
|
|
63
|
+
def __init__(self):
|
|
64
|
+
"""Initialize base image generation class."""
|
|
65
|
+
pass
|
|
66
|
+
|
|
67
|
+
@classmethod
|
|
68
|
+
def _load_from(cls,
|
|
69
|
+
model_path: str,
|
|
70
|
+
scheduler_config_path: str = "",
|
|
71
|
+
plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
|
|
72
|
+
device_id: Optional[str] = None,
|
|
73
|
+
float16: bool = True,
|
|
74
|
+
quantize: bool = False,
|
|
75
|
+
**kwargs
|
|
76
|
+
) -> 'ImageGen':
|
|
77
|
+
"""Load image generation model from local path, routing to appropriate implementation."""
|
|
78
|
+
# Check plugin_id value for routing - handle both enum and string
|
|
79
|
+
plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
|
|
80
|
+
|
|
81
|
+
if plugin_value == "mlx":
|
|
82
|
+
from nexaai.image_gen_impl.mlx_image_gen_impl import MLXImageGenImpl
|
|
83
|
+
return MLXImageGenImpl._load_from(model_path, scheduler_config_path, plugin_id, device_id, float16, quantize)
|
|
84
|
+
else:
|
|
85
|
+
from nexaai.image_gen_impl.pybind_image_gen_impl import PyBindImageGenImpl
|
|
86
|
+
return PyBindImageGenImpl._load_from(model_path, scheduler_config_path, plugin_id, device_id, float16, quantize)
|
|
87
|
+
|
|
88
|
+
@abstractmethod
|
|
89
|
+
def load_model(self, model_path: str, extra_data: Optional[str] = None) -> bool:
|
|
90
|
+
"""Load model from path."""
|
|
91
|
+
pass
|
|
92
|
+
|
|
93
|
+
@abstractmethod
|
|
94
|
+
def set_scheduler(self, config: SchedulerConfig) -> None:
|
|
95
|
+
"""Set scheduler configuration."""
|
|
96
|
+
pass
|
|
97
|
+
|
|
98
|
+
@abstractmethod
|
|
99
|
+
def set_sampler(self, config: ImageSamplerConfig) -> None:
|
|
100
|
+
"""Set sampler configuration."""
|
|
101
|
+
pass
|
|
102
|
+
|
|
103
|
+
@abstractmethod
|
|
104
|
+
def reset_sampler(self) -> None:
|
|
105
|
+
"""Reset sampler to default configuration."""
|
|
106
|
+
pass
|
|
107
|
+
|
|
108
|
+
@abstractmethod
|
|
109
|
+
def txt2img(self, prompt: str, config: ImageGenerationConfig) -> Image:
|
|
110
|
+
"""Generate image from text prompt."""
|
|
111
|
+
pass
|
|
112
|
+
|
|
113
|
+
@abstractmethod
|
|
114
|
+
def img2img(self, init_image: Image, prompt: str, config: ImageGenerationConfig) -> Image:
|
|
115
|
+
"""Generate image from initial image and text prompt."""
|
|
116
|
+
pass
|
|
117
|
+
|
|
118
|
+
@abstractmethod
|
|
119
|
+
def generate(self, config: ImageGenerationConfig) -> Image:
|
|
120
|
+
"""Generate image from configuration."""
|
|
121
|
+
pass
|
|
122
|
+
|
|
123
|
+
@abstractmethod
|
|
124
|
+
def set_lora(self, lora_id: int) -> None:
|
|
125
|
+
"""Set active LoRA adapter."""
|
|
126
|
+
pass
|
|
127
|
+
|
|
128
|
+
@abstractmethod
|
|
129
|
+
def add_lora(self, lora_path: str) -> int:
|
|
130
|
+
"""Add LoRA adapter and return its ID."""
|
|
131
|
+
pass
|
|
132
|
+
|
|
133
|
+
@abstractmethod
|
|
134
|
+
def remove_lora(self, lora_id: int) -> None:
|
|
135
|
+
"""Remove LoRA adapter."""
|
|
136
|
+
pass
|
|
137
|
+
|
|
138
|
+
@abstractmethod
|
|
139
|
+
def list_loras(self) -> List[int]:
|
|
140
|
+
"""List available LoRA adapters."""
|
|
141
|
+
pass
|
|
File without changes
|
|
@@ -0,0 +1,292 @@
|
|
|
1
|
+
# Note: This code is generated by Cursor, not tested yet.
|
|
2
|
+
|
|
3
|
+
from typing import List, Optional, Union
|
|
4
|
+
import os
|
|
5
|
+
|
|
6
|
+
from nexaai.common import PluginID
|
|
7
|
+
from nexaai.image_gen import ImageGen, ImageGenerationConfig, ImageSamplerConfig, SchedulerConfig, Image
|
|
8
|
+
from nexaai.mlx_backend.sd.interface import ImageGen as MLXImageGenInterface
|
|
9
|
+
|
|
10
|
+
|
|
11
|
+
class MLXImageGenImpl(ImageGen):
|
|
12
|
+
def __init__(self):
|
|
13
|
+
"""Initialize MLX Image Generation implementation."""
|
|
14
|
+
super().__init__()
|
|
15
|
+
self._mlx_image_gen = None
|
|
16
|
+
|
|
17
|
+
@classmethod
|
|
18
|
+
def _load_from(cls,
|
|
19
|
+
model_path: str,
|
|
20
|
+
scheduler_config_path: str = "",
|
|
21
|
+
plugin_id: Union[PluginID, str] = PluginID.MLX,
|
|
22
|
+
device_id: Optional[str] = None,
|
|
23
|
+
float16: bool = True,
|
|
24
|
+
quantize: bool = False
|
|
25
|
+
) -> 'MLXImageGenImpl':
|
|
26
|
+
"""Load image generation model from local path using MLX backend."""
|
|
27
|
+
try:
|
|
28
|
+
# MLX Image Generation interface is already imported
|
|
29
|
+
|
|
30
|
+
# Create instance and load MLX image generation model
|
|
31
|
+
instance = cls()
|
|
32
|
+
instance._mlx_image_gen = MLXImageGenInterface(
|
|
33
|
+
model_path=model_path,
|
|
34
|
+
scheduler_config_path=scheduler_config_path,
|
|
35
|
+
device=device_id,
|
|
36
|
+
float16=float16,
|
|
37
|
+
quantize=quantize
|
|
38
|
+
)
|
|
39
|
+
|
|
40
|
+
return instance
|
|
41
|
+
except Exception as e:
|
|
42
|
+
raise RuntimeError(f"Failed to load MLX Image Generation: {str(e)}")
|
|
43
|
+
|
|
44
|
+
def eject(self):
|
|
45
|
+
"""Destroy the model and free resources."""
|
|
46
|
+
if self._mlx_image_gen:
|
|
47
|
+
self._mlx_image_gen.destroy()
|
|
48
|
+
self._mlx_image_gen = None
|
|
49
|
+
|
|
50
|
+
def load_model(self, model_path: str, extra_data: Optional[str] = None) -> bool:
|
|
51
|
+
"""Load model from path."""
|
|
52
|
+
if not self._mlx_image_gen:
|
|
53
|
+
raise RuntimeError("MLX Image Generator not initialized")
|
|
54
|
+
|
|
55
|
+
try:
|
|
56
|
+
return self._mlx_image_gen.load_model(model_path, extra_data)
|
|
57
|
+
except Exception as e:
|
|
58
|
+
raise RuntimeError(f"Failed to load image generation model: {str(e)}")
|
|
59
|
+
|
|
60
|
+
def set_scheduler(self, config: SchedulerConfig) -> None:
|
|
61
|
+
"""Set scheduler configuration."""
|
|
62
|
+
if not self._mlx_image_gen:
|
|
63
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
64
|
+
|
|
65
|
+
try:
|
|
66
|
+
# Convert our config to MLX format
|
|
67
|
+
import sys
|
|
68
|
+
mlx_path = self._get_mlx_path()
|
|
69
|
+
if mlx_path not in sys.path:
|
|
70
|
+
sys.path.insert(0, mlx_path)
|
|
71
|
+
|
|
72
|
+
# MLX ImageGen uses a simpler interface, we'll store the config for use in generation
|
|
73
|
+
self._scheduler_config = config
|
|
74
|
+
self._mlx_image_gen.set_scheduler(config)
|
|
75
|
+
except Exception as e:
|
|
76
|
+
raise RuntimeError(f"Failed to set scheduler: {str(e)}")
|
|
77
|
+
|
|
78
|
+
def set_sampler(self, config: ImageSamplerConfig) -> None:
|
|
79
|
+
"""Set sampler configuration."""
|
|
80
|
+
if not self._mlx_image_gen:
|
|
81
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
82
|
+
|
|
83
|
+
try:
|
|
84
|
+
# Convert our config to MLX format
|
|
85
|
+
import sys
|
|
86
|
+
mlx_path = self._get_mlx_path()
|
|
87
|
+
if mlx_path not in sys.path:
|
|
88
|
+
sys.path.insert(0, mlx_path)
|
|
89
|
+
from sd.interface import ImageSamplerConfig as MLXImageSamplerConfig
|
|
90
|
+
|
|
91
|
+
mlx_config = MLXImageSamplerConfig(
|
|
92
|
+
method=config.method,
|
|
93
|
+
steps=config.steps,
|
|
94
|
+
guidance_scale=config.guidance_scale,
|
|
95
|
+
eta=config.eta,
|
|
96
|
+
seed=config.seed
|
|
97
|
+
)
|
|
98
|
+
|
|
99
|
+
self._mlx_image_gen.set_sampler(mlx_config)
|
|
100
|
+
except Exception as e:
|
|
101
|
+
raise RuntimeError(f"Failed to set sampler: {str(e)}")
|
|
102
|
+
|
|
103
|
+
def reset_sampler(self) -> None:
|
|
104
|
+
"""Reset sampler to default configuration."""
|
|
105
|
+
if not self._mlx_image_gen:
|
|
106
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
107
|
+
|
|
108
|
+
try:
|
|
109
|
+
self._mlx_image_gen.reset_sampler()
|
|
110
|
+
except Exception as e:
|
|
111
|
+
raise RuntimeError(f"Failed to reset sampler: {str(e)}")
|
|
112
|
+
|
|
113
|
+
def txt2img(self, prompt: str, config: ImageGenerationConfig) -> Image:
|
|
114
|
+
"""Generate image from text prompt."""
|
|
115
|
+
if not self._mlx_image_gen:
|
|
116
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
117
|
+
|
|
118
|
+
try:
|
|
119
|
+
# Convert our config to MLX format
|
|
120
|
+
import sys
|
|
121
|
+
mlx_path = self._get_mlx_path()
|
|
122
|
+
if mlx_path not in sys.path:
|
|
123
|
+
sys.path.insert(0, mlx_path)
|
|
124
|
+
from sd.interface import ImageGenerationConfig as MLXImageGenerationConfig
|
|
125
|
+
|
|
126
|
+
mlx_config = self._convert_generation_config(config)
|
|
127
|
+
|
|
128
|
+
# Use MLX text-to-image generation
|
|
129
|
+
result = self._mlx_image_gen.txt2img(prompt, mlx_config)
|
|
130
|
+
|
|
131
|
+
# Convert MLX result to our format
|
|
132
|
+
return self._convert_image_result(result)
|
|
133
|
+
|
|
134
|
+
except Exception as e:
|
|
135
|
+
raise RuntimeError(f"Failed to generate image from text: {str(e)}")
|
|
136
|
+
|
|
137
|
+
def img2img(self, init_image: Image, prompt: str, config: ImageGenerationConfig) -> Image:
|
|
138
|
+
"""Generate image from initial image and text prompt."""
|
|
139
|
+
if not self._mlx_image_gen:
|
|
140
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
141
|
+
|
|
142
|
+
try:
|
|
143
|
+
# Convert our config to MLX format
|
|
144
|
+
mlx_config = self._convert_generation_config(config)
|
|
145
|
+
|
|
146
|
+
# Convert our image to MLX format
|
|
147
|
+
import sys
|
|
148
|
+
mlx_path = self._get_mlx_path()
|
|
149
|
+
if mlx_path not in sys.path:
|
|
150
|
+
sys.path.insert(0, mlx_path)
|
|
151
|
+
from sd.interface import Image as MLXImage
|
|
152
|
+
|
|
153
|
+
mlx_init_image = MLXImage(
|
|
154
|
+
data=init_image.data,
|
|
155
|
+
width=init_image.width,
|
|
156
|
+
height=init_image.height,
|
|
157
|
+
channels=init_image.channels
|
|
158
|
+
)
|
|
159
|
+
|
|
160
|
+
# Use MLX image-to-image generation
|
|
161
|
+
result = self._mlx_image_gen.img2img(mlx_init_image, prompt, mlx_config)
|
|
162
|
+
|
|
163
|
+
# Convert MLX result to our format
|
|
164
|
+
return self._convert_image_result(result)
|
|
165
|
+
|
|
166
|
+
except Exception as e:
|
|
167
|
+
raise RuntimeError(f"Failed to generate image from image: {str(e)}")
|
|
168
|
+
|
|
169
|
+
def generate(self, config: ImageGenerationConfig) -> Image:
|
|
170
|
+
"""Generate image from configuration."""
|
|
171
|
+
if not self._mlx_image_gen:
|
|
172
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
173
|
+
|
|
174
|
+
try:
|
|
175
|
+
# Convert our config to MLX format
|
|
176
|
+
mlx_config = self._convert_generation_config(config)
|
|
177
|
+
|
|
178
|
+
# Use MLX generation
|
|
179
|
+
result = self._mlx_image_gen.generate(mlx_config)
|
|
180
|
+
|
|
181
|
+
# Convert MLX result to our format
|
|
182
|
+
return self._convert_image_result(result)
|
|
183
|
+
|
|
184
|
+
except Exception as e:
|
|
185
|
+
raise RuntimeError(f"Failed to generate image: {str(e)}")
|
|
186
|
+
|
|
187
|
+
def set_lora(self, lora_id: int) -> None:
|
|
188
|
+
"""Set active LoRA adapter."""
|
|
189
|
+
if not self._mlx_image_gen:
|
|
190
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
191
|
+
|
|
192
|
+
try:
|
|
193
|
+
self._mlx_image_gen.set_lora(lora_id)
|
|
194
|
+
except Exception as e:
|
|
195
|
+
raise RuntimeError(f"Failed to set LoRA: {str(e)}")
|
|
196
|
+
|
|
197
|
+
def add_lora(self, lora_path: str) -> int:
|
|
198
|
+
"""Add LoRA adapter and return its ID."""
|
|
199
|
+
if not self._mlx_image_gen:
|
|
200
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
201
|
+
|
|
202
|
+
try:
|
|
203
|
+
return self._mlx_image_gen.add_lora(lora_path)
|
|
204
|
+
except Exception as e:
|
|
205
|
+
raise RuntimeError(f"Failed to add LoRA: {str(e)}")
|
|
206
|
+
|
|
207
|
+
def remove_lora(self, lora_id: int) -> None:
|
|
208
|
+
"""Remove LoRA adapter."""
|
|
209
|
+
if not self._mlx_image_gen:
|
|
210
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
211
|
+
|
|
212
|
+
try:
|
|
213
|
+
self._mlx_image_gen.remove_lora(lora_id)
|
|
214
|
+
except Exception as e:
|
|
215
|
+
raise RuntimeError(f"Failed to remove LoRA: {str(e)}")
|
|
216
|
+
|
|
217
|
+
def list_loras(self) -> List[int]:
|
|
218
|
+
"""List available LoRA adapters."""
|
|
219
|
+
if not self._mlx_image_gen:
|
|
220
|
+
raise RuntimeError("MLX Image Generator not loaded")
|
|
221
|
+
|
|
222
|
+
try:
|
|
223
|
+
return self._mlx_image_gen.list_loras()
|
|
224
|
+
except Exception as e:
|
|
225
|
+
raise RuntimeError(f"Failed to list LoRAs: {str(e)}")
|
|
226
|
+
|
|
227
|
+
def _convert_generation_config(self, config: ImageGenerationConfig):
|
|
228
|
+
"""Convert our config to MLX format."""
|
|
229
|
+
import sys
|
|
230
|
+
mlx_path = self._get_mlx_path()
|
|
231
|
+
if mlx_path not in sys.path:
|
|
232
|
+
sys.path.insert(0, mlx_path)
|
|
233
|
+
from sd.interface import ImageGenerationConfig as MLXImageGenerationConfig, ImageSamplerConfig as MLXImageSamplerConfig
|
|
234
|
+
|
|
235
|
+
mlx_sampler_config = None
|
|
236
|
+
if config.sampler_config:
|
|
237
|
+
mlx_sampler_config = MLXImageSamplerConfig(
|
|
238
|
+
method=config.sampler_config.method,
|
|
239
|
+
steps=config.sampler_config.steps,
|
|
240
|
+
guidance_scale=config.sampler_config.guidance_scale,
|
|
241
|
+
eta=config.sampler_config.eta,
|
|
242
|
+
seed=config.sampler_config.seed
|
|
243
|
+
)
|
|
244
|
+
|
|
245
|
+
mlx_init_image = None
|
|
246
|
+
if config.init_image:
|
|
247
|
+
from sd.interface import Image as MLXImage
|
|
248
|
+
mlx_init_image = MLXImage(
|
|
249
|
+
data=config.init_image.data,
|
|
250
|
+
width=config.init_image.width,
|
|
251
|
+
height=config.init_image.height,
|
|
252
|
+
channels=config.init_image.channels
|
|
253
|
+
)
|
|
254
|
+
|
|
255
|
+
return MLXImageGenerationConfig(
|
|
256
|
+
prompts=config.prompts,
|
|
257
|
+
negative_prompts=config.negative_prompts,
|
|
258
|
+
height=config.height,
|
|
259
|
+
width=config.width,
|
|
260
|
+
sampler_config=mlx_sampler_config,
|
|
261
|
+
lora_id=config.lora_id,
|
|
262
|
+
init_image=mlx_init_image,
|
|
263
|
+
strength=config.strength,
|
|
264
|
+
n_images=config.n_images,
|
|
265
|
+
n_rows=config.n_rows,
|
|
266
|
+
decoding_batch_size=config.decoding_batch_size
|
|
267
|
+
)
|
|
268
|
+
|
|
269
|
+
def _convert_image_result(self, mlx_image):
|
|
270
|
+
"""Convert MLX image to our format."""
|
|
271
|
+
return Image(
|
|
272
|
+
data=mlx_image.data,
|
|
273
|
+
width=mlx_image.width,
|
|
274
|
+
height=mlx_image.height,
|
|
275
|
+
channels=mlx_image.channels
|
|
276
|
+
)
|
|
277
|
+
|
|
278
|
+
@staticmethod
|
|
279
|
+
def _get_mlx_path():
|
|
280
|
+
"""Get the MLX py-lib path."""
|
|
281
|
+
from pathlib import Path
|
|
282
|
+
current_dir = Path(__file__).parent.parent.parent.parent # Go back to nexa root
|
|
283
|
+
possible_paths = [
|
|
284
|
+
current_dir / "plugins" / "mlx" / "py-lib",
|
|
285
|
+
Path(os.environ.get("NEXA_MLX_PATH", "")) if os.environ.get("NEXA_MLX_PATH") else None,
|
|
286
|
+
]
|
|
287
|
+
|
|
288
|
+
for path in possible_paths:
|
|
289
|
+
if path and path.exists() and (path / "sd" / "interface.py").exists():
|
|
290
|
+
return str(path.absolute())
|
|
291
|
+
|
|
292
|
+
raise ImportError("Could not find MLX py-lib directory")
|
|
@@ -0,0 +1,85 @@
|
|
|
1
|
+
from typing import List, Optional, Union
|
|
2
|
+
|
|
3
|
+
from nexaai.common import PluginID
|
|
4
|
+
from nexaai.image_gen import ImageGen, ImageGenerationConfig, ImageSamplerConfig, SchedulerConfig, Image
|
|
5
|
+
|
|
6
|
+
|
|
7
|
+
class PyBindImageGenImpl(ImageGen):
|
|
8
|
+
def __init__(self):
|
|
9
|
+
"""Initialize PyBind Image Generation implementation."""
|
|
10
|
+
super().__init__()
|
|
11
|
+
# TODO: Add PyBind-specific initialization
|
|
12
|
+
|
|
13
|
+
@classmethod
|
|
14
|
+
def _load_from(cls,
|
|
15
|
+
model_path: str,
|
|
16
|
+
scheduler_config_path: str = "",
|
|
17
|
+
plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
|
|
18
|
+
device_id: Optional[str] = None,
|
|
19
|
+
float16: bool = True,
|
|
20
|
+
quantize: bool = False
|
|
21
|
+
) -> 'PyBindImageGenImpl':
|
|
22
|
+
"""Load image generation model from local path using PyBind backend."""
|
|
23
|
+
# TODO: Implement PyBind image generation loading
|
|
24
|
+
instance = cls()
|
|
25
|
+
return instance
|
|
26
|
+
|
|
27
|
+
def eject(self):
|
|
28
|
+
"""Destroy the model and free resources."""
|
|
29
|
+
# TODO: Implement PyBind image generation cleanup
|
|
30
|
+
pass
|
|
31
|
+
|
|
32
|
+
def load_model(self, model_path: str, extra_data: Optional[str] = None) -> bool:
|
|
33
|
+
"""Load model from path."""
|
|
34
|
+
# TODO: Implement PyBind image generation model loading
|
|
35
|
+
raise NotImplementedError("PyBind image generation model loading not yet implemented")
|
|
36
|
+
|
|
37
|
+
def set_scheduler(self, config: SchedulerConfig) -> None:
|
|
38
|
+
"""Set scheduler configuration."""
|
|
39
|
+
# TODO: Implement PyBind scheduler setting
|
|
40
|
+
raise NotImplementedError("PyBind scheduler setting not yet implemented")
|
|
41
|
+
|
|
42
|
+
def set_sampler(self, config: ImageSamplerConfig) -> None:
|
|
43
|
+
"""Set sampler configuration."""
|
|
44
|
+
# TODO: Implement PyBind sampler setting
|
|
45
|
+
raise NotImplementedError("PyBind sampler setting not yet implemented")
|
|
46
|
+
|
|
47
|
+
def reset_sampler(self) -> None:
|
|
48
|
+
"""Reset sampler to default configuration."""
|
|
49
|
+
# TODO: Implement PyBind sampler reset
|
|
50
|
+
raise NotImplementedError("PyBind sampler reset not yet implemented")
|
|
51
|
+
|
|
52
|
+
def txt2img(self, prompt: str, config: ImageGenerationConfig) -> Image:
|
|
53
|
+
"""Generate image from text prompt."""
|
|
54
|
+
# TODO: Implement PyBind text-to-image
|
|
55
|
+
raise NotImplementedError("PyBind text-to-image not yet implemented")
|
|
56
|
+
|
|
57
|
+
def img2img(self, init_image: Image, prompt: str, config: ImageGenerationConfig) -> Image:
|
|
58
|
+
"""Generate image from initial image and text prompt."""
|
|
59
|
+
# TODO: Implement PyBind image-to-image
|
|
60
|
+
raise NotImplementedError("PyBind image-to-image not yet implemented")
|
|
61
|
+
|
|
62
|
+
def generate(self, config: ImageGenerationConfig) -> Image:
|
|
63
|
+
"""Generate image from configuration."""
|
|
64
|
+
# TODO: Implement PyBind image generation
|
|
65
|
+
raise NotImplementedError("PyBind image generation not yet implemented")
|
|
66
|
+
|
|
67
|
+
def set_lora(self, lora_id: int) -> None:
|
|
68
|
+
"""Set active LoRA adapter."""
|
|
69
|
+
# TODO: Implement PyBind LoRA setting
|
|
70
|
+
raise NotImplementedError("PyBind LoRA setting not yet implemented")
|
|
71
|
+
|
|
72
|
+
def add_lora(self, lora_path: str) -> int:
|
|
73
|
+
"""Add LoRA adapter and return its ID."""
|
|
74
|
+
# TODO: Implement PyBind LoRA addition
|
|
75
|
+
raise NotImplementedError("PyBind LoRA addition not yet implemented")
|
|
76
|
+
|
|
77
|
+
def remove_lora(self, lora_id: int) -> None:
|
|
78
|
+
"""Remove LoRA adapter."""
|
|
79
|
+
# TODO: Implement PyBind LoRA removal
|
|
80
|
+
raise NotImplementedError("PyBind LoRA removal not yet implemented")
|
|
81
|
+
|
|
82
|
+
def list_loras(self) -> List[int]:
|
|
83
|
+
"""List available LoRA adapters."""
|
|
84
|
+
# TODO: Implement PyBind LoRA listing
|
|
85
|
+
raise NotImplementedError("PyBind LoRA listing not yet implemented")
|
nexaai/llm.py
ADDED
|
@@ -0,0 +1,98 @@
|
|
|
1
|
+
from typing import Generator, Optional, Union
|
|
2
|
+
from abc import abstractmethod
|
|
3
|
+
import queue
|
|
4
|
+
import threading
|
|
5
|
+
|
|
6
|
+
from nexaai.common import ModelConfig, GenerationConfig, ChatMessage, PluginID
|
|
7
|
+
from nexaai.base import BaseModel, ProfilingData
|
|
8
|
+
|
|
9
|
+
class LLM(BaseModel):
|
|
10
|
+
def __init__(self, m_cfg: ModelConfig = ModelConfig()):
|
|
11
|
+
"""Initialize base LLM class."""
|
|
12
|
+
self._m_cfg = m_cfg
|
|
13
|
+
self._cancel_event = threading.Event() # New attribute to control cancellation
|
|
14
|
+
|
|
15
|
+
@classmethod
|
|
16
|
+
def _load_from(cls,
|
|
17
|
+
local_path: str,
|
|
18
|
+
model_name: Optional[str] = None,
|
|
19
|
+
tokenizer_path: Optional[str] = None,
|
|
20
|
+
m_cfg: ModelConfig = ModelConfig(),
|
|
21
|
+
plugin_id: Union[PluginID, str] = PluginID.LLAMA_CPP,
|
|
22
|
+
device_id: Optional[str] = None,
|
|
23
|
+
**kwargs
|
|
24
|
+
) -> 'LLM':
|
|
25
|
+
"""Load model from local path, routing to appropriate implementation."""
|
|
26
|
+
# Check plugin_id value for routing - handle both enum and string
|
|
27
|
+
plugin_value = plugin_id.value if isinstance(plugin_id, PluginID) else plugin_id
|
|
28
|
+
|
|
29
|
+
if plugin_value == "mlx":
|
|
30
|
+
from nexaai.llm_impl.mlx_llm_impl import MLXLLMImpl
|
|
31
|
+
return MLXLLMImpl._load_from(local_path, model_name, tokenizer_path, m_cfg, plugin_id, device_id)
|
|
32
|
+
else:
|
|
33
|
+
from nexaai.llm_impl.pybind_llm_impl import PyBindLLMImpl
|
|
34
|
+
return PyBindLLMImpl._load_from(local_path, model_name, tokenizer_path, m_cfg, plugin_id, device_id)
|
|
35
|
+
|
|
36
|
+
def cancel_generation(self):
|
|
37
|
+
"""Signal to cancel any ongoing stream generation."""
|
|
38
|
+
self._cancel_event.set()
|
|
39
|
+
|
|
40
|
+
def reset_cancel(self):
|
|
41
|
+
"""Reset the cancel event. Call before starting a new generation if needed."""
|
|
42
|
+
self._cancel_event.clear()
|
|
43
|
+
|
|
44
|
+
@abstractmethod
|
|
45
|
+
def apply_chat_template(self, messages: list[ChatMessage], tools: Optional[str] = None, enable_thinking: bool = True, add_generation_prompt: bool = True) -> str:
|
|
46
|
+
"""Apply the chat template to messages."""
|
|
47
|
+
pass
|
|
48
|
+
|
|
49
|
+
@abstractmethod
|
|
50
|
+
def generate_stream(self, prompt: str, g_cfg: GenerationConfig = GenerationConfig()) -> Generator[str, None, None]:
|
|
51
|
+
"""Generate text with streaming."""
|
|
52
|
+
pass
|
|
53
|
+
|
|
54
|
+
@abstractmethod
|
|
55
|
+
def generate(self, prompt: str, g_cfg: GenerationConfig = GenerationConfig()) -> str:
|
|
56
|
+
"""
|
|
57
|
+
Generate text without streaming.
|
|
58
|
+
|
|
59
|
+
Args:
|
|
60
|
+
prompt (str): The prompt to generate text from. For chat models, this is the chat messages after chat template is applied.
|
|
61
|
+
g_cfg (GenerationConfig): Generation configuration.
|
|
62
|
+
|
|
63
|
+
Returns:
|
|
64
|
+
str: The generated text.
|
|
65
|
+
"""
|
|
66
|
+
pass
|
|
67
|
+
|
|
68
|
+
def get_profiling_data(self) -> Optional[ProfilingData]:
|
|
69
|
+
"""Get profiling data from the last generation."""
|
|
70
|
+
pass
|
|
71
|
+
|
|
72
|
+
@abstractmethod
|
|
73
|
+
def save_kv_cache(self, path: str):
|
|
74
|
+
"""
|
|
75
|
+
Save the key-value cache to the file.
|
|
76
|
+
|
|
77
|
+
Args:
|
|
78
|
+
path (str): The path to the file.
|
|
79
|
+
"""
|
|
80
|
+
pass
|
|
81
|
+
|
|
82
|
+
@abstractmethod
|
|
83
|
+
def load_kv_cache(self, path: str):
|
|
84
|
+
"""
|
|
85
|
+
Load the key-value cache from the file.
|
|
86
|
+
|
|
87
|
+
Args:
|
|
88
|
+
path (str): The path to the file.
|
|
89
|
+
"""
|
|
90
|
+
pass
|
|
91
|
+
|
|
92
|
+
@abstractmethod
|
|
93
|
+
def reset(self):
|
|
94
|
+
"""
|
|
95
|
+
Reset the LLM model context and KV cache. If not reset, the model will skip the number of evaluated tokens and treat tokens after those as the new incremental tokens.
|
|
96
|
+
If your past chat history changed, or you are starting a new chat, you should always reset the model before running generate.
|
|
97
|
+
"""
|
|
98
|
+
pass
|
|
File without changes
|