Merge pull request #275 from acon96/release/v0.3.9

Release/v0.3.9
2026-01-09 13:48:05 -05:00 · 2025-05-26 21:58:23 -04:00
parent b9e16c5ed2 55f377d8cb
commit 05927ad693
9 changed files with 58 additions and 54 deletions
--- a/.github/workflows/create-release.yml
+++ b/.github/workflows/create-release.yml
@@ -13,43 +13,41 @@ permissions:

 jobs:
  build_wheels:
-    name: Build wheels on ${{ matrix.arch }}${{ matrix.suffix }} (HA ${{ matrix.home_assistant_version }})
+    name: Build wheels for ${{ matrix.arch }}${{ matrix.suffix }} (HA ${{ matrix.home_assistant_image }})
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        include:
        # ARM variants
-        - home_assistant_version: "2025.4.1"
+        - home_assistant_image: "aarch64-homeassistant:2025.4.1"
          arch: "aarch64"
-        - home_assistant_version: "2025.4.1"
+        - home_assistant_image: "armhf-homeassistant:2025.4.1"
          arch: "armhf"

        # Base x86
-        - home_assistant_version: "2025.4.1"
-          suffix: "-noavx"
-          arch: "amd64"
-          extra_defines: "-DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF"
-        - home_assistant_version: "2025.4.1"
-          arch: "i386"
-          suffix: "-noavx"
-          extra_defines: "-DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF"
+        - home_assistant_image: "amd64-homeassistant:2025.4.1"
+          suffix: "noavx"
+          arch: "x86_64"
+          extra_defines: "-DGGML_SSE42=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_BMI2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF"

        # AVX2 and AVX512
-        - home_assistant_version: "2025.4.1"
-          arch: "amd64"
-          extra_defines: "-DGGML_AVX=ON -DGGML_AVX2=ON -DGGML_FMA=ON -DGGML_F16C=ON"
-        - home_assistant_version: "2025.4.1"
-          arch: "amd64"
-          suffix: "-avx512"
-          extra_defines: "-DGGML_AVX512=ON -DGGML_FMA=ON -DGGML_F16C=ON"
-        - home_assistant_version: "2025.4.1"
+        - home_assistant_image: "amd64-homeassistant:2025.4.1"
+          arch: "x86_64"
+          extra_defines: "-DGGML_SSE42=ON -DGGML_AVX=ON -DGGML_AVX2=ON -DGGML_BMI2=ON -DGGML_FMA=ON -DGGML_F16C=ON"
+        - home_assistant_image: "amd64-homeassistant:2025.4.1"
+          arch: "x86_64"
+          suffix: "avx512"
+          extra_defines: "-DGGML_SSE42=ON -DGGML_AVX=ON -DGGML_AVX2=ON -DGGML_BMI2=ON -DGGML_AVX512=ON -DGGML_FMA=ON -DGGML_F16C=ON"
+
+        # 32 bit for older processors, with and without AVX enabled
+        - home_assistant_image: "i386-homeassistant:2025.4.1"
          arch: "i386"
-          extra_defines: "-DGGML_AVX=ON -DGGML_AVX2=ON -DGGML_FMA=ON -DGGML_F16C=ON"
-        - home_assistant_version: "2025.4.1"
+          suffix: "noavx"
+          extra_defines: "-DGGML_SSE42=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_BMI2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF"
+        - home_assistant_image: "i386-homeassistant:2025.4.1"
          arch: "i386"
-          suffix: "-avx512"
-          extra_defines: "-DGGML_AVX512=ON -DGGML_FMA=ON -DGGML_F16C=ON"
+          extra_defines: "-DGGML_SSE42=ON -DGGML_AVX=ON -DGGML_AVX2=OFF -DGGML_BMI2=OFF -DGGML_FMA=ON -DGGML_F16C=ON"
        
    steps:
      - name: Checkout code
@@ -83,7 +81,7 @@ jobs:
        with:
          arch: none
          distro: none
-          base_image: homeassistant/${{ matrix.arch }}-homeassistant:${{ matrix.home_assistant_version }}
+          base_image: homeassistant/${{ matrix.home_assistant_image }}

          # Create an artifacts directory
          setup: |
@@ -106,23 +104,20 @@ 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

+            tag="homellm${{ matrix.suffix }}"
+            sed -i -E "s/^(__version__ *= *\"[0-9]+\.[0-9]+\.[0-9]+)\"/\1+${tag}\"/" llama_cpp/__init__.py
+
            export CMAKE_ARGS="-DLLAVA_BUILD=OFF -DGGML_NATIVE=OFF ${{ matrix.extra_defines }}"
            python3 -m build --wheel

-            ls -la ./dist/
-            for filename in ./dist/*.whl; do
-              output_file=$(basename $filename .whl)${{ matrix.suffix }}.whl
-              echo "$filename -> $output_file"
-              mv "$filename" "/artifacts/${output_file}";
-            done;
-
+            mv ./dist/*.whl /artifacts
            ls -la /artifacts/

      - name: Upload artifacts
        uses: actions/upload-artifact@v4
        with:
          path: ./artifacts/*.whl
-          name: artifact_${{ matrix.arch }}${{ matrix.suffix }}_${{ matrix.home_assistant_version }}
+          name: artifact_${{ matrix.arch }}${{ matrix.suffix }}

  release:
    name: Create Release
--- a/README.md
+++ b/README.md
@@ -150,6 +150,7 @@ In order to facilitate running the project entirely on the system where Home Ass
 ## Version History
 | Version | Description                                                                                                                                                                                                          |
 |---------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| v0.3.9  | Update llama.cpp version, fix installation bugs, fix conversation history not working                                                                                                                                |
 | v0.3.8  | Update llama.cpp, remove think blocks from "thinking" models, fix wheel detection for some Intel CPUs, Fixes for compatibility with latest Home Assistant version (2025.4), other small bug fixes                    |
 | v0.3.7  | Update llama.cpp version to support newer models, Update minimum Home Assistant version to 2024.12.3, Add German In-Context Learning examples, Fix multi-turn use, Fix an issue with webcolors                       | 
 | v0.3.6  | Small llama.cpp backend fixes                                                                                                                                                                                        |
--- a/custom_components/llama_conversation/const.py
+++ b/custom_components/llama_conversation/const.py
@@ -403,5 +403,5 @@ OPTIONS_OVERRIDES = {
    },
 }

-INTEGRATION_VERSION = "0.3.8"
-EMBEDDED_LLAMA_CPP_PYTHON_VERSION = "0.3.5"
+INTEGRATION_VERSION = "0.3.9"
+EMBEDDED_LLAMA_CPP_PYTHON_VERSION = "0.3.9"
--- a/custom_components/llama_conversation/conversation.py
+++ b/custom_components/llama_conversation/conversation.py
@@ -25,7 +25,7 @@ from homeassistant.core import HomeAssistant, callback
 from homeassistant.exceptions import ConfigEntryNotReady, ConfigEntryError, TemplateError, HomeAssistantError
 from homeassistant.helpers import config_validation as cv, intent, template, entity_registry as er, llm, \
    area_registry as ar, device_registry as dr, chat_session
-from homeassistant.helpers.entity_platform import AddEntitiesCallback
+from homeassistant.helpers.entity_platform import AddConfigEntryEntitiesCallback
 from homeassistant.helpers.aiohttp_client import async_get_clientsession
 from homeassistant.helpers.event import async_track_state_change, async_call_later
 from homeassistant.components.sensor import SensorEntity
@@ -148,7 +148,7 @@ async def update_listener(hass: HomeAssistant, entry: ConfigEntry):

    return True

-async def async_setup_entry(hass: HomeAssistant, entry: ConfigEntry, async_add_entities: AddEntitiesCallback) -> bool:
+async def async_setup_entry(hass: HomeAssistant, entry: ConfigEntry, async_add_entities: AddConfigEntryEntitiesCallback) -> bool:
    """Set up Local LLM Conversation from a config entry."""

    # handle updates to the options
@@ -198,6 +198,7 @@ class LocalLLMAgent(ConversationEntity, AbstractConversationAgent):
    in_context_examples: list[dict]

    _attr_has_entity_name = True
+    _attr_supports_streaming = False # TODO: add support for backends that can stream

    def __init__(self, hass: HomeAssistant, entry: ConfigEntry) -> None:
        """Initialize the agent."""
@@ -423,7 +424,7 @@ class LocalLLMAgent(ConversationEntity, AbstractConversationAgent):
            if remember_num_interactions and len(message_history) > (remember_num_interactions * 2) + 1:
                for i in range(0,2):
                    message_history.pop(1)
-            # chat_log.content = [_convert_content_back(user_input.agent_id, message_history_entry) for message_history_entry in message_history ]
+            chat_log.content = [_convert_content_back(user_input.agent_id, message_history_entry) for message_history_entry in message_history ]

        if llm_api is None:
            # return the output without messing with it if there is no API exposed to the model
--- a/custom_components/llama_conversation/manifest.json
+++ b/custom_components/llama_conversation/manifest.json
@@ -1,7 +1,7 @@
 {
  "domain": "llama_conversation",
  "name": "Local LLM Conversation",
-  "version": "0.3.8",
+  "version": "0.3.9",
  "codeowners": ["@acon96"],
  "config_flow": true,
  "dependencies": ["conversation"],
--- a/custom_components/llama_conversation/utils.py
+++ b/custom_components/llama_conversation/utils.py
@@ -169,14 +169,14 @@ def install_llama_cpp_python(config_dir: str):
    # remap other names for architectures to the names we use
    if platform_suffix == "arm64":
        platform_suffix = "aarch64"
-    if platform_suffix == "i386" or platform_suffix == "x86_64":
-        platform_suffix = "amd64"
+    if platform_suffix == "i386" or platform_suffix == "amd64":
+        platform_suffix = "x86_64"

    runtime_version = f"cp{sys.version_info.major}{sys.version_info.minor}"

    instruction_extensions_suffix = ""
-    if platform_suffix == "amd64":
-        instruction_extensions_suffix = "-noavx"
+    if platform_suffix == "x86_64":
+        instruction_extensions_suffix = "noavx"

        try:
            with open("/proc/cpuinfo") as f:
@@ -184,7 +184,7 @@ def install_llama_cpp_python(config_dir: str):

            _LOGGER.debug(cpu_features)
            if " avx512f " in cpu_features and " avx512bw " in cpu_features:
-                instruction_extensions_suffix = "-avx512"
+                instruction_extensions_suffix = "avx512"
            elif " avx2 " in cpu_features and \
                 " avx " in cpu_features and \
                 " f16c " in cpu_features and \
@@ -194,11 +194,15 @@ def install_llama_cpp_python(config_dir: str):
        except Exception as ex:
            _LOGGER.debug(f"Couldn't detect CPU features: {ex}")
            # default to the noavx build to avoid crashing home assistant
-            instruction_extensions_suffix = "-noavx"
+            instruction_extensions_suffix = "noavx"
    
    folder = os.path.dirname(__file__)
-    potential_wheels = sorted([ path for path in os.listdir(folder) if path.endswith(f"{platform_suffix}{instruction_extensions_suffix}.whl") ], reverse=True)
+    potential_wheels = sorted([ path for path in os.listdir(folder) if path.endswith(f"{platform_suffix}.whl") ], reverse=True)
    potential_wheels = [ wheel for wheel in potential_wheels if runtime_version in wheel ]
+    if instruction_extensions_suffix:
+        potential_wheels = [ wheel for wheel in potential_wheels if f"+homellm{instruction_extensions_suffix}" in wheel ]
+
+    _LOGGER.debug(f"{potential_wheels=}")
    if len(potential_wheels) > 0:

        latest_wheel = potential_wheels[0]
@@ -208,8 +212,8 @@ def install_llama_cpp_python(config_dir: str):
        return install_package(os.path.join(folder, latest_wheel), pip_kwargs(config_dir))
        
    # scikit-build-core v0.9.7+ doesn't recognize these builds as musllinux, and just tags them as generic linux
-    # github_release_url = f"https://github.com/acon96/home-llm/releases/download/v{INTEGRATION_VERSION}/llama_cpp_python-{EMBEDDED_LLAMA_CPP_PYTHON_VERSION}-{runtime_version}-{runtime_version}-musllinux_1_2_{platform_suffix}{instruction_extensions_suffix}.whl"
-    github_release_url = f"https://github.com/acon96/home-llm/releases/download/v{INTEGRATION_VERSION}/llama_cpp_python-{EMBEDDED_LLAMA_CPP_PYTHON_VERSION}-{runtime_version}-{runtime_version}-linux_{platform_suffix}{instruction_extensions_suffix}.whl"
+    # github_release_url = f"https://github.com/acon96/home-llm/releases/download/v{INTEGRATION_VERSION}/llama_cpp_python-{EMBEDDED_LLAMA_CPP_PYTHON_VERSION}+homellm{instruction_extensions_suffix}-{runtime_version}-{runtime_version}-musllinux_1_2_{platform_suffix}.whl"
+    github_release_url = f"https://github.com/acon96/home-llm/releases/download/v{INTEGRATION_VERSION}/llama_cpp_python-{EMBEDDED_LLAMA_CPP_PYTHON_VERSION}+homellm{instruction_extensions_suffix}-{runtime_version}-{runtime_version}-linux_{platform_suffix}.whl"
    if install_package(github_release_url, pip_kwargs(config_dir)):
        _LOGGER.info("llama-cpp-python successfully installed from GitHub release")
        return True
--- a/docs/Training.md
+++ b/docs/Training.md
@@ -232,18 +232,18 @@ python3 train.py \

 #### Llama 3.2 3B Instruct
 ```
-python3 generate_home_assistant_data.py --train --test --large --sharegpt --language english german french spanish
+python3 generate_home_assistant_data.py --train --test --large --sharegpt --language english german french spanish polish

 python3 train.py \
-    --run_name Home-Llama-3.2-3B-rev1 \
+    --run_name Home-Llama-3.2-3B-rev2 \
    --base_model meta-llama/Llama-3.2-3B-Instruct \
    --bf16 \
-    --train_dataset data/home_assistant_train.jsonl \
+    --train_dataset data/home_assistant_train_large.jsonl \
    --test_dataset data/home_assistant_test.jsonl \
    --learning_rate 1e-5 --learning_rate_warmup 0.03 --batch_size 64 --epochs 1 \
-    --micro_batch_size 2 \
+    --micro_batch_size 1 \
    --ctx_size 2048 \
-    --save_steps 200 --save_total_limit 3 --eval_steps 100 --logging_steps 2
+    --save_steps 200 --save_total_limit 1 --eval_steps 100 --logging_steps 2
 ```

 ### Problems
--- a/scripts/make_wheel.sh
+++ b/scripts/make_wheel.sh
@@ -9,6 +9,9 @@ git clone --recurse-submodules https://github.com/abetlen/llama-cpp-python --bra
 cd llama-cpp-python
 pip3 install build

+tag="homellm"
+sed -i -E "s/^(__version__ *= *\"[0-9]+\.[0-9]+\.[0-9]+)\"/\1+${tag}\"/" llama_cpp/__init__.py
+
 export CMAKE_ARGS="-DLLAVA_BUILD=OFF -DGGML_NATIVE=ON"
 python3 -m build --wheel
 cp -f ./dist/*.whl /tmp/dist/
--- a/scripts/run_docker_to_make_wheels.sh
+++ b/scripts/run_docker_to_make_wheels.sh
@@ -1,6 +1,6 @@
 #!/bin/bash

-VERSION_TO_BUILD="v0.3.5"
+VERSION_TO_BUILD="v0.3.9"

 # make python 11 wheels
 # docker run -it --rm \
@@ -18,4 +18,4 @@ VERSION_TO_BUILD="v0.3.5"
 docker run -it --rm \
    --entrypoint bash \
    -v $(pwd):/tmp/dist \
-    homeassistant/home-assistant:2024.12.3 /tmp/dist/make_wheel.sh $VERSION_TO_BUILD
+    homeassistant/home-assistant:2025.4.1 /tmp/dist/make_wheel.sh $VERSION_TO_BUILD