Merge pull request #142 from acon96/release/v0.2.17

Release v0.2.17
2026-01-09 13:48:05 -05:00 · 2024-05-08 21:04:36 -04:00
parent 0cb361c305 105f09ba2c
commit d64f3a25f6
8 changed files with 103 additions and 28 deletions
--- a/.github/workflows/create-release.yml
+++ b/.github/workflows/create-release.yml
@@ -18,26 +18,50 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        home_assistant_version: ["2023.12.4", "2024.2.1"]
-        arch: [aarch64, armhf, amd64, i386]
-        suffix: [""]
        include:
+        # ARM variants
+        - home_assistant_version: "2023.12.4"
+          arch: "aarch64"
        - home_assistant_version: "2024.2.1"
+          arch: "aarch64"
+        - home_assistant_version: "2023.12.4"
+          arch: "armhf"
+        - home_assistant_version: "2024.2.1"
+          arch: "armhf"
+
+        # Base x86
+        - home_assistant_version: "2024.2.1"
+          suffix: "-noavx"
+          arch: "amd64"
+          extra_defines: "-DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF"
+        - home_assistant_version: "2023.12.4"
          arch: "amd64"
          suffix: "-noavx"
-          extra_defines: "-DLLAMA_NATIVE=OFF -DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF"
-        - home_assistant_version: "2024.2.1"
-          arch: "amd64"
-          suffix: "-avx512"
-          extra_defines: "-DLLAMA_NATIVE=OFF -DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_AVX512=ON"
+          extra_defines: "-DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF"
        - home_assistant_version: "2024.2.1"
          arch: "i386"
          suffix: "-noavx"
-          extra_defines: "-DLLAMA_NATIVE=OFF -DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF"
+          extra_defines: "-DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF"
+        - home_assistant_version: "2023.12.4"
+          arch: "i386"
+          suffix: "-noavx"
+          extra_defines: "-DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF"
+
+        # AVX2 and AVX512
+        - home_assistant_version: "2024.2.1"
+          arch: "amd64"
+          extra_defines: "-DLLAMA_AVX=ON -DLLAMA_AVX2=ON -DLLAMA_FMA=ON -DLLAMA_F16C=ON"
+        - home_assistant_version: "2024.2.1"
+          arch: "amd64"
+          suffix: "-avx512"
+          extra_defines: "-DLLAMA_AVX512=ON -DLLAMA_FMA=ON -DLLAMA_F16C=ON"
+        - home_assistant_version: "2024.2.1"
+          arch: "i386"
+          extra_defines: "-DLLAMA_AVX=ON -DLLAMA_AVX2=ON -DLLAMA_FMA=ON -DLLAMA_F16C=ON"
        - home_assistant_version: "2024.2.1"
          arch: "i386"
          suffix: "-avx512"
-          extra_defines: "-DLLAMA_NATIVE=OFF -DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_AVX512=ON"
+          extra_defines: "-DLLAMA_AVX512=ON -DLLAMA_FMA=ON -DLLAMA_F16C=ON"
        
    steps:
      - name: Checkout code
@@ -75,7 +99,7 @@ jobs:
            git clone --quiet --recurse-submodules https://github.com/abetlen/llama-cpp-python --branch "v${{ env.EMBEDDED_LLAMA_CPP_PYTHON_VERSION }}"
            cd llama-cpp-python

-            export CMAKE_ARGS="-DLLAVA_BUILD=OFF ${{ matrix.extra_defines }}"
+            export CMAKE_ARGS="-DLLAVA_BUILD=OFF -DLLAMA_NATIVE=OFF ${{ matrix.extra_defines }}"
            python3 -m build --wheel

            ls -la ./dist/
--- a/README.md
+++ b/README.md
@@ -126,7 +126,8 @@ In order to facilitate running the project entirely on the system where Home Ass
 ## Version History
 | Version | Description                                                                                                                                                                                                          |
 |---------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-| v0.2.16 | Fix for missing huggingface_hub package preventing startup                                                                                                                                                                   |
+| v0.2.17 | Disable native llama.cpp wheel optimizations, add Command R prompt format                                                                                                                                            |
+| v0.2.16 | Fix for missing huggingface_hub package preventing startup                                                                                                                                                           |
 | v0.2.15 | Fix startup error when using llama.cpp backend and add flash attention to llama.cpp backend                                                                                                                          |
 | v0.2.14 | Fix llama.cpp wheels + AVX detection                                                                                                                                                                                 |
 | v0.2.13 | Add support for Llama 3, build llama.cpp wheels that are compatible with non-AVX systems, fix an error with exposing script entities, fix multiple small Ollama backend issues, and add basic multi-language support |
--- a/custom_components/llama_conversation/const.py
+++ b/custom_components/llama_conversation/const.py
@@ -63,6 +63,7 @@ CONF_ALLOWED_SERVICE_CALL_ARGUMENTS = "allowed_service_call_arguments"
 DEFAULT_ALLOWED_SERVICE_CALL_ARGUMENTS = ["rgb_color", "brightness", "temperature", "humidity", "fan_mode", "hvac_mode", "preset_mode", "item", "duration"]
 CONF_PROMPT_TEMPLATE = "prompt_template"
 PROMPT_TEMPLATE_CHATML = "chatml"
+PROMPT_TEMPLATE_COMMAND_R = "command-r"
 PROMPT_TEMPLATE_ALPACA = "alpaca"
 PROMPT_TEMPLATE_VICUNA = "vicuna"
 PROMPT_TEMPLATE_MISTRAL = "mistral"
@@ -79,6 +80,12 @@ PROMPT_TEMPLATE_DESCRIPTIONS = {
        "assistant": { "prefix": "<|im_start|>assistant\n", "suffix": "<|im_end|>" },
        "generation_prompt": "<|im_start|>assistant"
    },
+    PROMPT_TEMPLATE_COMMAND_R: {
+        "system": { "prefix": "<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>", "suffix": "<|END_OF_TURN_TOKEN|>" },
+        "user": { "prefix": "<|START_OF_TURN_TOKEN|><|USER_TOKEN|>", "suffix": "<|END_OF_TURN_TOKEN|>" },
+        "assistant": { "prefix": "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>", "suffix": "<|END_OF_TURN_TOKEN|>" },
+        "generation_prompt": "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>"
+    },
    PROMPT_TEMPLATE_ALPACA: {
        "system": { "prefix": "", "suffix": "\n" },
        "user": { "prefix": "### Instruction:\n", "suffix": "\n" },
@@ -280,9 +287,14 @@ OPTIONS_OVERRIDES = {
        CONF_PROMPT_TEMPLATE: PROMPT_TEMPLATE_ZEPHYR,
    },
    "phi-3": {
-        CONF_PROMPT_TEMPLATE: PROMPT_TEMPLATE_ZEPHYR3
+        CONF_PROMPT: DEFAULT_PROMPT_BASE + ICL_EXTRAS,
+        CONF_PROMPT_TEMPLATE: PROMPT_TEMPLATE_ZEPHYR3,
+    },
+    "command-r": {
+        CONF_PROMPT: DEFAULT_PROMPT_BASE + ICL_EXTRAS,
+        CONF_PROMPT_TEMPLATE: PROMPT_TEMPLATE_COMMAND_R,
    }
 }

-INTEGRATION_VERSION = "0.2.16"
-EMBEDDED_LLAMA_CPP_PYTHON_VERSION = "0.2.69"
+INTEGRATION_VERSION = "0.2.17"
+EMBEDDED_LLAMA_CPP_PYTHON_VERSION = "0.2.70"
--- a/custom_components/llama_conversation/manifest.json
+++ b/custom_components/llama_conversation/manifest.json
@@ -1,7 +1,7 @@
 {
  "domain": "llama_conversation",
  "name": "LLaMA Conversation",
-  "version": "0.2.16",
+  "version": "0.2.17",
  "codeowners": ["@acon96"],
  "config_flow": true,
  "dependencies": ["conversation"],
--- a/custom_components/llama_conversation/translations/en.json
+++ b/custom_components/llama_conversation/translations/en.json
@@ -166,6 +166,7 @@
                "zephyr2": "Zephyr ('</s>')",
                "zephyr3": "Zephyr (<|end|>)",
                "llama3": "Llama 3",
+                "command-r": "Command R",
                "no_prompt_template": "None"
            }
        },
--- a/custom_components/llama_conversation/utils.py
+++ b/custom_components/llama_conversation/utils.py
@@ -114,13 +114,21 @@ def install_llama_cpp_python(config_dir: str):

    instruction_extensions_suffix = ""
    if platform_suffix == "amd64" or platform_suffix == "i386":
+        instruction_extensions_suffix = "-noavx"
+
        try:
            with open("/proc/cpuinfo") as f:
                cpu_features = [ line for line in f.readlines() if line.startswith("Features") or line.startswith("flags")][0]
-            if "avx512f" in cpu_features and "avx512bw" in cpu_features:
+
+            _LOGGER.debug(cpu_features)
+            if " avx512f " in cpu_features and " avx512bw " in cpu_features:
                instruction_extensions_suffix = "-avx512"
-            elif "avx2" not in cpu_features or "avx" not in cpu_features or "f16c" not in cpu_features or "fma" not in cpu_features or not ("sse3" in cpu_features or "ssse3" in cpu_features):
-                instruction_extensions_suffix = "-noavx"
+            elif " avx2 " in cpu_features and \
+                 " avx " in cpu_features and \
+                 " f16c " in cpu_features and \
+                 " fma " in cpu_features and \
+                 (" sse3 " in cpu_features or " ssse3 " in cpu_features):
+                instruction_extensions_suffix = ""
        except Exception as ex:
            _LOGGER.debug(f"Couldn't detect CPU features: {ex}")
            # default to the noavx build to avoid crashing home assistant
--- a/data/translate_data.py
+++ b/data/translate_data.py
@@ -1,14 +1,15 @@
 """Original script by @BramNH on GitHub"""
+import argparse
 import csv
 import os
 import time
 import re

-from deep_translator import GoogleTranslator
+from deep_translator import GoogleTranslator, DeeplTranslator
 from deep_translator.base import BaseTranslator
 from deep_translator.exceptions import TooManyRequests
 from tqdm import tqdm
-from transformers import pipeline
+import langcodes

 SUPPORTED_DEVICES = [
    "light",
@@ -32,20 +33,22 @@ class Seq2SeqTranslator(BaseTranslator):
    def __init__(self, model_name: str, **kwargs):
        super().__init__(**kwargs)

+        from transformers import pipeline
+
        self.translator = pipeline("translation", model=model_name, tokenizer=model_name, device=0)

    def translate(self, text: str, **kwargs):
        return self.translator(text)[0]["translation_text"]

 class DatasetTranslator:
-    translator: Seq2SeqTranslator
+    translator: BaseTranslator
    source_language: str
    target_language: str

-    def __init__(self, source_language, target_language, model_name):
+    def __init__(self, source_language, target_language, translator):
        self.source_language = source_language
        self.target_language = target_language
-        self.translator = Seq2SeqTranslator(model_name=model_name)
+        self.translator = translator

    def translate_all_piles(self):
        os.makedirs(f"./piles/{self.target_language}", exist_ok=True)
@@ -334,7 +337,32 @@ class DatasetTranslator:
            f.writelines(pile_of_todo_items_target)


-# TODO: cmd line args
-DatasetTranslator("english", "german", "Helsinki-NLP/opus-mt-en-de").translate_all_piles()
-# DatasetTranslator("english", "spanish", "Helsinki-NLP/opus-mt-en-es").translate_all_piles()
-# DatasetTranslator("english", "french", "Helsinki-NLP/opus-mt-en-fr").translate_all_piles()
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("destination_language", help="Destination language for translation")
+    parser.add_argument("--source_language", default="english", help="Source language for translation (default: english)")
+    parser.add_argument("--translator_type", choices=["google", "transformers", "deepl"], required=True, help="Translator type (choose from: google, transformers, deepl)")
+    parser.add_argument("--model_name", help="Model name (optional)")
+    parser.add_argument("--api_key", help="API key (optional)")
+
+    args = parser.parse_args()
+
+    source_code = langcodes.find(args.source_language).language
+    dest_code = langcodes.find(args.destination_language).language
+
+    if args.translator_type == "google":
+        translator = GoogleTranslator(source=source_code, target=dest_code)
+    elif args.translator_type == "transformers":
+        if not args.model_name:
+            print("No model name was provided!")
+            parser.print_usage()
+            exit(-1)
+        translator = Seq2SeqTranslator(model_name=args.model_name)
+    elif args.translator_type == "deepl":
+        if not args.api_key and os.getenv():
+            print("No api key was provided!")
+            parser.print_usage()
+            exit(-1)
+        translator = DeeplTranslator(source=source_code, target=dest_code, api_key=args.api_key)
+
+    DatasetTranslator(args.source_language, args.destination_language, translator).translate_all_piles()
--- a/requirements.txt
+++ b/requirements.txt
@@ -9,6 +9,7 @@ pandas
 # flash-attn
 sentencepiece
 deep-translator
+langcodes

 homeassistant
 hassil