merge from main

NexaAI · Aug 23, 2024 · 9637c55 · 9637c55
2 parents 59a0169 + 5a5a59e
commit 9637c55
Show file tree

Hide file tree

Showing 36 changed files with 673 additions and 398 deletions.
diff --git a/.github/workflows/build-wheels-cpu-macos.yaml b/.github/workflows/build-wheels-cpu-macos.yaml
@@ -36,6 +36,7 @@ jobs:
         env:
           # disable repair
           CIBW_REPAIR_WHEEL_COMMAND: ""
+          CIBW_BUILD_FRONTEND: "build"
           CIBW_ENVIRONMENT: CMAKE_ARGS="-DCMAKE_OSX_ARCHITECTURES=arm64;x86_64 "
           CIBW_BUILD: "cp310-* cp311-* cp312-*"
           MACOSX_DEPLOYMENT_TARGET: ${{ matrix.os }}

diff --git a/.github/workflows/build-wheels-cpu.yaml b/.github/workflows/build-wheels-cpu.yaml
@@ -35,6 +35,7 @@ jobs:
         env:
           # disable repair
           CIBW_REPAIR_WHEEL_COMMAND: ""
+          CIBW_BUILD_FRONTEND: "build"
           CIBW_SKIP: "*musllinux*"
           CIBW_BUILD: "cp310-* cp311-* cp312-*"
           CMAKE_BUILD_PARALLEL_LEVEL: 4
@@ -77,6 +78,7 @@ jobs:
         env:
           # disable repair
           CIBW_REPAIR_WHEEL_COMMAND: ""
+          CIBW_BUILD_FRONTEND: "build"
           CIBW_BUILD: "cp310-* cp311-* cp312-*"
           CMAKE_BUILD_PARALLEL_LEVEL: 16
         with:
@@ -104,8 +106,9 @@ jobs:
       - name: Build wheels
         uses: pypa/[email protected]
         env:
-          CIBW_SKIP: "*musllinux* pp*"
           CIBW_REPAIR_WHEEL_COMMAND: ""
+          CIBW_BUILD_FRONTEND: "build"
+          CIBW_SKIP: "*musllinux* pp*"
           CIBW_ARCHS: "aarch64"
           CIBW_BUILD: "cp310-* cp311-* cp312-*"
           CMAKE_BUILD_PARALLEL_LEVEL: $(nproc)

diff --git a/.github/workflows/build-wheels-metal.yaml b/.github/workflows/build-wheels-metal.yaml
@@ -35,6 +35,7 @@ jobs:
         env:
           # disable repair
           CIBW_REPAIR_WHEEL_COMMAND: ""
+          CIBW_BUILD_FRONTEND: "build"
           CIBW_ARCHS: "arm64"
           CIBW_BUILD: "cp310-* cp311-* cp312-*"
           CIBW_ENVIRONMENT: CMAKE_ARGS="-DCMAKE_OSX_ARCHITECTURES=arm64 -DCMAKE_APPLE_SILICON_PROCESSOR=arm64 -DGGML_METAL=ON -DSD_METAL=ON"

diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
@@ -25,7 +25,7 @@ jobs:
         fetch-depth: 0  # This fetches all history so you can access any version of the submodules
 
     - name: Set up Python
-      uses: actions/setup-python@v4
+      uses: actions/setup-python@v5
       with:
         python-version: ${{ matrix.python-version }}
 

diff --git a/README.md b/README.md
@@ -2,7 +2,7 @@
 
 <h1>Nexa SDK</h1>
 
-  <img src="assets/banner.png" alt="icon"/>
+  <img src="./assets/banner.png" alt="icon"/>
 
 [![MacOS][MacOS-image]][release-url] [![Linux][Linux-image]][release-url] [![Windows][Windows-image]][release-url]
 

diff --git a/nexa/__init__.py b/nexa/__init__.py
@@ -1 +1 @@
-__version__ = "0.0.1"
+__version__ = "0.0.6"
diff --git a/nexa/cli/entry.py b/nexa/cli/entry.py
diff --git a/nexa/constants.py b/nexa/constants.py
@@ -61,7 +61,12 @@
     "phi3v": "Phi-3-vision-128k-instruct:onnx-cpu-int4",
     "sd1-5": "stable-diffusion-v1-5:onnx-cpu-fp32",
     "lcm-dreamshaper": "lcm-dreamshaper-v7:onnx-cpu-fp32",
-    "whisper": "whisper-tiny:onnx-cpu-fp32",
+    "whisper-tiny.en": "whisper-tiny.en:onnx-cpu-fp32",
+    "whisper-tiny": "whisper-tiny:onnx-cpu-fp32",
+    "whisper-small.en": "whisper-small.en:onnx-cpu-fp32",
+    "whisper-small": "whisper-small:onnx-cpu-fp32",
+    "whisper-base.en": "whisper-base.en:onnx-cpu-fp32",
+    "whisper-base": "whisper-base:onnx-cpu-fp32",
     "ljspeech": "ljspeech-jets:onnx-cpu-fp32",
 }
 
@@ -86,10 +91,14 @@
     "whisper-large": "whisper-large:bin-large-v3",
     "whisper-tiny": "whisper-tiny:bin-tiny",
     "faster-whisper-tiny": "faster-whisper-tiny:bin-cpu-fp16",
+    "faster-whisper-tiny.en": "faster-whisper-tiny.en:bin-cpu-fp16",
     "faster-whisper-small": "faster-whisper-small:bin-cpu-fp16",
+    "faster-whisper-small.en": "faster-whisper-small.en:bin-cpu-fp16",
     "faster-whisper-medium": "faster-whisper-medium:bin-cpu-fp16",
+    "faster-whisper-medium.en": "faster-whisper-medium.en:bin-cpu-fp16",
     "faster-whisper-base": "faster-whisper-base:bin-cpu-fp16",
-    "faster-whisper-large": "faster-whisper-large:bin-cpu-fp16",
+    "faster-whisper-base.en": "faster-whisper-base.en:bin-cpu-fp16",
+    "faster-whisper-large": "faster-whisper-large-v3:bin-cpu-fp16",
 }
 
 NEXA_RUN_MODEL_MAP_FUNCTION_CALLING = {
@@ -129,6 +138,8 @@
     **NEXA_RUN_MODEL_MAP_TEXT,
     **NEXA_RUN_MODEL_MAP_VLM,
     **NEXA_RUN_MODEL_MAP_IMAGE,
+    **NEXA_RUN_MODEL_MAP_VOICE,
+    **NEXA_RUN_MODEL_MAP_FUNCTION_CALLING,
 }
 
 NEXA_RUN_CHAT_TEMPLATE_MAP = {
@@ -205,4 +216,79 @@
     "task": "transcribe",
     "temperature": 0.0,
     "compute_type": "default"
-}
+}
+
+NEXA_OFFICIAL_MODELS_TYPE = {
+  'gemma-2b': 'NLP',
+  'Llama-2-7b-chat': 'NLP',
+  'Llama-2-7b': 'NLP',
+  'Meta-Llama-3-8B-Instruct': 'NLP',
+  'Meta-Llama-3.1-8B-Instruct': 'NLP',
+  'Mistral-7B-Instruct-v0.3': 'NLP',
+  'Mistral-7B-Instruct-v0.2': 'NLP',
+  'Phi-3-mini-128k-instruct': 'NLP',
+  'Phi-3-mini-4k-instruct': 'NLP',
+  'CodeQwen1.5-7B-Instruct': 'NLP',
+  'Qwen2-0.5B-Instruct': 'NLP',
+  'Qwen2-1.5B-Instruct': 'NLP',
+  'Qwen2-7B-Instruct': 'NLP',
+  'codegemma-2b': 'NLP',
+  'gemma-1.1-2b-instruct': 'NLP',
+  'gemma-2b-instruct': 'NLP',
+  'gemma-2-9b-instruct': 'NLP',
+  'gemma-1.1-7b-instruct': 'NLP',
+  'gemma-7b-instruct': 'NLP',
+  'gemma-7b': 'NLP',
+  'Qwen2-1.5B': 'NLP',
+  'codegemma-7b': 'NLP',
+  'TinyLlama-1.1B-Chat-v1.0': 'NLP',
+  'CodeLlama-7b-Instruct': 'NLP',
+  'gpt2': 'NLP',
+  'CodeLlama-7b': 'NLP',
+  'CodeLlama-7b-Python': 'NLP',
+  'Qwen1.5-7B-Instruct': 'NLP',
+  'Qwen1.5-7B': 'NLP',
+  'Phi-2': 'NLP',
+  'deepseek-coder-1.3b-instruct': 'NLP',
+  'deepseek-coder-1.3b-base': 'NLP',
+  'deepseek-coder-6.7b-instruct': 'NLP',
+  'dolphin-2.8-mistral-7b': 'NLP',
+  'gemma-2-2b-instruct': 'NLP',
+  'Octopus-v2': 'NLP',
+  'Octopus-v4': 'NLP',
+  'Octo-planner': 'NLP',
+  'deepseek-coder-6.7b-base': 'NLP',
+  'Llama2-7b-chat-uncensored': 'NLP',
+  'Llama3-8B-Lexi-Uncensored': 'NLP',
+  'Llama2-7b-function-calling': 'NLP',
+  'lcm-dreamshaper-v7': 'Computer Vision',
+  'stable-diffusion-v1-5': 'Computer Vision',
+  'stable-diffusion-v1-4': 'Computer Vision',
+  'stable-diffusion-v2-1': 'Computer Vision',
+  'sdxl-turbo': 'Computer Vision',
+  'hassaku-hentai-model-v13-LCM': 'Computer Vision',
+  'anything-v30-LCM': 'Computer Vision',
+  'Phi-3-vision-128k-instruct': 'Multimodal',
+  'nanoLLaVA': 'Multimodal',
+  'llava-v1.6-mistral-7b': 'Multimodal',
+  'llava-v1.6-vicuna-7b': 'Multimodal',
+  'llava-phi-3-mini': 'Multimodal',
+  'llava-llama-3-8b-v1.1': 'Multimodal',
+  'faster-whisper-tiny.en': 'Audio',
+  'faster-whisper-tiny': 'Audio',
+  'faster-whisper-small.en': 'Audio',
+  'faster-whisper-small': 'Audio',
+  'faster-whisper-medium.en': 'Audio',
+  'faster-whisper-medium': 'Audio',
+  'faster-whisper-base.en': 'Audio',
+  'faster-whisper-base': 'Audio',
+  'faster-whisper-large-v3': 'Audio',
+  'whisper-tiny.en': 'Audio',
+  'whisper-tiny': 'Audio',
+  'whisper-small.en': 'Audio',
+  'whisper-small': 'Audio',
+  'whisper-base.en': 'Audio',
+  'whisper-base': 'Audio',
+}
+
+