granite-3.3-release-test.yml

name: granite-3.3-tagged-test-release
run-name: "[TEST] Granite 3.3 - CI/CD quantized GGUF models [${{github.ref_type}}: ${{github.ref_name}}]"

on:
  push:
    tags:
      - "test-v3.3*"
  workflow_dispatch:

permissions:
 contents: read
 packages: read

env:
  DEBUG: true
  TRACE: false
  ENABLE_LANGUAGE_JOBS: true
  ENABLE_VISION_JOBS: true
  ENABLE_GUARDIAN_JOBS: true
  ENABLE_EMBEDDING_JOBS: true
  ENABLE_DOCLING_JOBS: false

  # Enable appending the converted format (BF16/F16) to BVT test matrix
  ENABLE_BVT_TEST_CONVERTED_FORMAT: true

  # Enable RAG testing for embedding models (runs after successful BVT)
  ENABLE_RAG_TESTING: false

  TARGET_HF_REPO_NAME_EXT: -GGUF
  TARGET_HF_REPO_OWNER: mrutkows
  TARGET_HF_REPO_PRIVATE: false
  COLLECTION_CONFIG: "resources/json/latest/hf_collection_mapping_gguf.json"

  DEFAULT_ARTIFACTS_DIR: 'artifacts'
  ARTIFACT_LOGS_NAME: 'bvt-logs'
  ARTIFACT_RESPONSES_NAME: 'bvt-responses'
  EXT_LOG: '.log.txt'
  EXT_RESPONSE: '.response.txt'

  # HuggingFace Transformers versions
  # V4.52.1 is the last well-tested version we know the CI can support for Granite 3.3 & 3.2
  TRANSFORMERS_VERSION_LANGUAGE: "4.52.1"
  TRANSFORMERS_VERSION_VISION: "4.52.1" # granite-vision-3.3-2b w/ config.json model arch. rename hack
  TRANSFORMERS_VERSION_GUARDIAN: "4.52.1"
  TRANSFORMERS_VERSION_EMBEDDING: "4.52.1"
  TRANSFORMERS_VERSION_DOCLING: "4.52.1"

  # llama.cpp build tags per model type - determines which binary tools to use from ./bin subdirectory
  # Available tags: b6569, b6808, b8100 (corresponds to llama.cpp release tags)
  LLAMACPP_BUILD_TAG_LANGUAGE: b6569
  LLAMACPP_BUILD_TAG_VISION: b8100 # b7951 worked, b6808 did not... mtmd support for granite-vision-3.3-2b vision model
  LLAMACPP_BUILD_TAG_GUARDIAN: b6569
  LLAMACPP_BUILD_TAG_EMBEDDING: b8100 # ModernBert support: https://github.com/ggml-org/llama.cpp/pull/18330
  LLAMACPP_BUILD_TAG_DOCLING: b6569

  # 'ibm-granite/granite-3.3-2b-base',
  # 'ibm-granite/granite-3.3-8b-base',
  # 'ibm-granite/granite-3.3-2b-instruct',
  # 'ibm-granite/granite-3.3-8b-instruct'
  SOURCE_LANGUAGE_REPOS: "[
      'ibm-granite/granite-3.3-2b-instruct',
      'ibm-granite/granite-3.3-8b-instruct'
    ]"
  # 'Q2_K',
  # 'Q3_K_L', 'Q3_K_M', 'Q3_K_S',
  # 'Q4_0', 'Q4_1', 'Q4_K_M', 'Q4_K_S',
  # 'Q5_0', 'Q5_1', 'Q5_K_M', 'Q5_K_S',
  # 'Q6_K',
  # 'Q8_0'
  TARGET_LANGUAGE_QUANTIZATIONS: "[
      'Q4_K_M',
      'Q8_0'
    ]"
  # 'ibm-granite/granite-guardian-3.3-8b'
  SOURCE_GUARDIAN_REPOS: "[
      'ibm-granite/granite-guardian-3.3-8b'
    ]"
  # 'Q4_K_M',
  # 'Q5_K_M',
  # 'Q6_K',
  # 'Q8_0'
  TARGET_GUARDIAN_QUANTIZATIONS: "[
      'Q8_0'
    ]"
  # 'ibm-granite/granite-vision-3.3-2b',
  # 'ibm-granite/granite-vision-3.3-2b-chart2csv-preview'
  SOURCE_VISION_REPOS: "[
      'ibm-granite/granite-vision-3.3-2b'
    ]"
  # 'Q4_K_M',
  # 'Q5_K_M',
  # 'Q6_K',
  # 'Q8_0'
  TARGET_VISION_QUANTIZATIONS: "[
      'Q8_0'
    ]"
  # TODO: Sparse arch. not supp. by llama.cpp: 'ibm-granite/granite-embedding-30m-sparse',
  # Note: vis. embed. arch. is not supported in llama.cpp, transformers only with custom classes for ibm layers
  # 'ibm-granite/granite-embedding-30m-english',
  # 'ibm-granite/granite-embedding-125m-english',
  # 'ibm-granite/granite-embedding-107m-multilingual',
  # 'ibm-granite/granite-embedding-278m-multilingual'
  SOURCE_EMBEDDING_REPOS: "[
      'ibm-granite/granite-embedding-30m-english'
    ]"
  # 'Q4_K_M',
  # 'Q8_0'
  TARGET_EMBEDDING_QUANTIZATIONS: "[
      'Q8_0'
    ]"
  SOURCE_DOCLING_REPOS: "[
      'None'
    ]"
  TARGET_DOCLING_QUANTIZATIONS: "[
      'None'
    ]"

jobs:
  # ===========================================================================
  # Environment setup
  # ===========================================================================
  # TODO: use conditionals for skipping jobs: if: ${{ false }} such as TEST_ONLY
  # TODO: figure out why we cannot pass api_token to secrets.token on a workflow_call:
  environment-setup:
    runs-on: ubuntu-latest
    # Note: declaration of these outputs is essential to pass to the next job
    outputs:
      debug: ${{ steps.set-vars.outputs.debug }}
      trace: ${{ steps.set-vars.outputs.trace }}
      enable_language_jobs: ${{ steps.set-vars.outputs.enable_language_jobs }}
      enable_vision_jobs: ${{ steps.set-vars.outputs.enable_vision_jobs }}
      enable_guardian_jobs: ${{ steps.set-vars.outputs.enable_guardian_jobs }}
      enable_embedding_jobs: ${{ steps.set-vars.outputs.enable_embedding_jobs }}
      enable_docling_jobs: ${{ steps.set-vars.outputs.enable_docling_jobs }}
      enable_bvt_test_converted_format: ${{ steps.set-vars.outputs.enable_bvt_test_converted_format }}
      enable_rag_testing: ${{ steps.set-vars.outputs.enable_rag_testing }}
      hf_collection_config: ${{ steps.set-vars.outputs.hf_collection_config }}
      target_repo_private: ${{ steps.set-vars.outputs.target_repo_private }}
      target_repo_owner: ${{ steps.set-vars.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ steps.set-vars.outputs.target_repo_name_ext }}
      source_language_repos: "${{ steps.set-vars.outputs.source_language_repos }}"
      target_language_quantizations: "${{ steps.set-vars.outputs.target_language_quantizations }}"
      target_language_quantizations_bvt: "${{ steps.append-converted-format.outputs.target_language_quantizations_bvt }}"
      source_vision_repos: "${{ steps.set-vars.outputs.source_vision_repos }}"
      target_vision_quantizations: "${{ steps.set-vars.outputs.target_vision_quantizations }}"
      target_vision_quantizations_bvt: "${{ steps.append-converted-format.outputs.target_vision_quantizations_bvt }}"
      source_guardian_repos: "${{ steps.set-vars.outputs.source_guardian_repos }}"
      target_guardian_quantizations: "${{ steps.set-vars.outputs.target_guardian_quantizations }}"
      target_guardian_quantizations_bvt: "${{ steps.append-converted-format.outputs.target_guardian_quantizations_bvt }}"
      source_embedding_repos: "${{ steps.set-vars.outputs.source_embedding_repos }}"
      target_embedding_quantizations: "${{ steps.set-vars.outputs.target_embedding_quantizations }}"
      target_embedding_quantizations_bvt: "${{ steps.append-converted-format.outputs.target_embedding_quantizations_bvt }}"
      source_docling_repos: "${{ steps.set-vars.outputs.source_docling_repos }}"
      target_docling_quantizations: "${{ steps.set-vars.outputs.target_docling_quantizations }}"
      target_docling_quantizations_bvt: "${{ steps.append-converted-format.outputs.target_docling_quantizations_bvt }}"
      default_artifacts_dir: "${{ steps.set-vars.outputs.default_artifacts_dir }}"
      artifact_logs_name: "${{ steps.set-vars.outputs.artifact_logs_name }}"
      artifact_responses_name: "${{ steps.set-vars.outputs.artifact_responses_name }}"
      ext_log: "${{ steps.set-vars.outputs.ext_log }}"
      ext_response: "${{ steps.set-vars.outputs.ext_response }}"
      transformers_version_language: "${{ steps.set-vars.outputs.transformers_version_language }}"
      transformers_version_vision: "${{ steps.set-vars.outputs.transformers_version_vision }}"
      transformers_version_guardian: "${{ steps.set-vars.outputs.transformers_version_guardian }}"
      transformers_version_embedding: "${{ steps.set-vars.outputs.transformers_version_embedding }}"
      transformers_version_docling: "${{ steps.set-vars.outputs.transformers_version_docling }}"
      llamacpp_build_tag_language: "${{ steps.set-vars.outputs.llamacpp_build_tag_language }}"
      llamacpp_build_tag_vision: "${{ steps.set-vars.outputs.llamacpp_build_tag_vision }}"
      llamacpp_build_tag_guardian: "${{ steps.set-vars.outputs.llamacpp_build_tag_guardian }}"
      llamacpp_build_tag_embedding: "${{ steps.set-vars.outputs.llamacpp_build_tag_embedding }}"
      llamacpp_build_tag_docling: "${{ steps.set-vars.outputs.llamacpp_build_tag_docling }}"
      def_converted_quant_language: "${{ steps.set-vars.outputs.def_converted_quant_language }}"
      def_converted_quant_vision: "${{ steps.set-vars.outputs.def_converted_quant_vision }}"
      def_converted_quant_guardian: "${{ steps.set-vars.outputs.def_converted_quant_guardian }}"
      def_converted_quant_embedding: "${{ steps.set-vars.outputs.def_converted_quant_embedding }}"
      def_converted_quant_docling: "${{ steps.set-vars.outputs.def_converted_quant_docling }}"
    env:
      ARRAY_SEP: "','"
      DEF_CONVERTED_QUANT_LANGUAGE: "bf16"
      DEF_CONVERTED_QUANT_VISION: "bf16"
      DEF_CONVERTED_QUANT_GUARDIAN: "bf16"
      DEF_CONVERTED_QUANT_EMBEDDING: "bf16"
      DEF_CONVERTED_QUANT_DOCLING: "f16"
    steps:
      - name: Determine valid model build paths
        id: determine-valid-model-build-paths
        run: |
            NORMALIZED_SOURCE_LANGUAGE_REPOS="$(echo "$SOURCE_LANGUAGE_REPOS" | tr -d '[:space:]')"
            NORMALIZED_SOURCE_GUARDIAN_REPOS="$(echo "$SOURCE_GUARDIAN_REPOS" | tr -d '[:space:]')"
            NORMALIZED_SOURCE_VISION_REPOS="$(echo "$SOURCE_VISION_REPOS" | tr -d '[:space:]')"
            NORMALIZED_SOURCE_EMBEDDING_REPOS="$(echo "$SOURCE_EMBEDDING_REPOS" | tr -d '[:space:]')"
            NORMALIZED_SOURCE_DOCLING_REPOS="$(echo "$SOURCE_DOCLING_REPOS" | tr -d '[:space:]')"

            EFFECTIVE_ENABLE_LANGUAGE_JOBS="$ENABLE_LANGUAGE_JOBS"
            EFFECTIVE_ENABLE_GUARDIAN_JOBS="$ENABLE_GUARDIAN_JOBS"
            EFFECTIVE_ENABLE_VISION_JOBS="$ENABLE_VISION_JOBS"
            EFFECTIVE_ENABLE_EMBEDDING_JOBS="$ENABLE_EMBEDDING_JOBS"
            EFFECTIVE_ENABLE_DOCLING_JOBS="$ENABLE_DOCLING_JOBS"

            if [ "$ENABLE_LANGUAGE_JOBS" = "true" ] && { [ "$NORMALIZED_SOURCE_LANGUAGE_REPOS" = "['None']" ] || [ "$NORMALIZED_SOURCE_LANGUAGE_REPOS" = "[]" ]; }; then
              echo "⚠️ SOURCE_LANGUAGE_REPOS is empty/None; overriding ENABLE_LANGUAGE_JOBS=false"
              EFFECTIVE_ENABLE_LANGUAGE_JOBS=false
              SOURCE_LANGUAGE_REPOS="[]"
              TARGET_LANGUAGE_QUANTIZATIONS="[]"
            fi

            if [ "$ENABLE_GUARDIAN_JOBS" = "true" ] && { [ "$NORMALIZED_SOURCE_GUARDIAN_REPOS" = "['None']" ] || [ "$NORMALIZED_SOURCE_GUARDIAN_REPOS" = "[]" ]; }; then
              echo "⚠️ SOURCE_GUARDIAN_REPOS is empty/None; overriding ENABLE_GUARDIAN_JOBS=false"
              EFFECTIVE_ENABLE_GUARDIAN_JOBS=false
              SOURCE_GUARDIAN_REPOS="[]"
              TARGET_GUARDIAN_QUANTIZATIONS="[]"
            fi

            if [ "$ENABLE_VISION_JOBS" = "true" ] && { [ "$NORMALIZED_SOURCE_VISION_REPOS" = "['None']" ] || [ "$NORMALIZED_SOURCE_VISION_REPOS" = "[]" ]; }; then
              echo "⚠️ SOURCE_VISION_REPOS is empty/None; overriding ENABLE_VISION_JOBS=false"
              EFFECTIVE_ENABLE_VISION_JOBS=false
              SOURCE_VISION_REPOS="[]"
              TARGET_VISION_QUANTIZATIONS="[]"
            fi

            if [ "$ENABLE_EMBEDDING_JOBS" = "true" ] && { [ "$NORMALIZED_SOURCE_EMBEDDING_REPOS" = "['None']" ] || [ "$NORMALIZED_SOURCE_EMBEDDING_REPOS" = "[]" ]; }; then
              echo "⚠️ SOURCE_EMBEDDING_REPOS is empty/None; overriding ENABLE_EMBEDDING_JOBS=false"
              EFFECTIVE_ENABLE_EMBEDDING_JOBS=false
              SOURCE_EMBEDDING_REPOS="[]"
              TARGET_EMBEDDING_QUANTIZATIONS="[]"
            fi

            if [ "$ENABLE_DOCLING_JOBS" = "true" ] && { [ "$NORMALIZED_SOURCE_DOCLING_REPOS" = "['None']" ] || [ "$NORMALIZED_SOURCE_DOCLING_REPOS" = "[]" ]; }; then
              echo "⚠️ SOURCE_DOCLING_REPOS is empty/None; overriding ENABLE_DOCLING_JOBS=false"
              EFFECTIVE_ENABLE_DOCLING_JOBS=false
              SOURCE_DOCLING_REPOS="[]"
              TARGET_DOCLING_QUANTIZATIONS="[]"
            fi

            echo "ENABLE_LANGUAGE_JOBS=$EFFECTIVE_ENABLE_LANGUAGE_JOBS" >> "$GITHUB_ENV"
            echo "SOURCE_LANGUAGE_REPOS=$SOURCE_LANGUAGE_REPOS" >> "$GITHUB_ENV"
            echo "TARGET_LANGUAGE_QUANTIZATIONS=$TARGET_LANGUAGE_QUANTIZATIONS" >> "$GITHUB_ENV"

            echo "ENABLE_GUARDIAN_JOBS=$EFFECTIVE_ENABLE_GUARDIAN_JOBS" >> "$GITHUB_ENV"
            echo "SOURCE_GUARDIAN_REPOS=$SOURCE_GUARDIAN_REPOS" >> "$GITHUB_ENV"
            echo "TARGET_GUARDIAN_QUANTIZATIONS=$TARGET_GUARDIAN_QUANTIZATIONS" >> "$GITHUB_ENV"

            echo "ENABLE_VISION_JOBS=$EFFECTIVE_ENABLE_VISION_JOBS" >> "$GITHUB_ENV"
            echo "SOURCE_VISION_REPOS=$SOURCE_VISION_REPOS" >> "$GITHUB_ENV"
            echo "TARGET_VISION_QUANTIZATIONS=$TARGET_VISION_QUANTIZATIONS" >> "$GITHUB_ENV"

            echo "ENABLE_EMBEDDING_JOBS=$EFFECTIVE_ENABLE_EMBEDDING_JOBS" >> "$GITHUB_ENV"
            echo "SOURCE_EMBEDDING_REPOS=$SOURCE_EMBEDDING_REPOS" >> "$GITHUB_ENV"
            echo "TARGET_EMBEDDING_QUANTIZATIONS=$TARGET_EMBEDDING_QUANTIZATIONS" >> "$GITHUB_ENV"

            echo "ENABLE_DOCLING_JOBS=$EFFECTIVE_ENABLE_DOCLING_JOBS" >> "$GITHUB_ENV"
            echo "SOURCE_DOCLING_REPOS=$SOURCE_DOCLING_REPOS" >> "$GITHUB_ENV"
            echo "TARGET_DOCLING_QUANTIZATIONS=$TARGET_DOCLING_QUANTIZATIONS" >> "$GITHUB_ENV"

      - name: Set environment variables in GitHub output
        id: set-vars
        run: |
            echo "DEBUG: $DEBUG"
            echo "TRACE: $TRACE"
            echo "LLAMACPP_BUILD_TAG: $LLAMACPP_BUILD_TAG"
            echo "COLLECTION_CONFIG: $COLLECTION_CONFIG"
            echo "TARGET_HF_REPO_OWNER: $TARGET_HF_REPO_OWNER"
            echo "TARGET_HF_REPO_NAME_EXT: $TARGET_HF_REPO_NAME_EXT"
            echo "debug=$DEBUG" >> "$GITHUB_OUTPUT"
            echo "trace=$TRACE" >> "$GITHUB_OUTPUT"
            echo "enable_language_jobs=$ENABLE_LANGUAGE_JOBS" >> "$GITHUB_OUTPUT"
            echo "enable_vision_jobs=$ENABLE_VISION_JOBS" >> "$GITHUB_OUTPUT"
            echo "enable_guardian_jobs=$ENABLE_GUARDIAN_JOBS" >> "$GITHUB_OUTPUT"
            echo "enable_embedding_jobs=$ENABLE_EMBEDDING_JOBS" >> "$GITHUB_OUTPUT"
            echo "enable_docling_jobs=$ENABLE_DOCLING_JOBS" >> "$GITHUB_OUTPUT"
            echo "enable_bvt_test_converted_format=$ENABLE_BVT_TEST_CONVERTED_FORMAT" >> "$GITHUB_OUTPUT"
            echo "enable_rag_testing=$ENABLE_RAG_TESTING" >> "$GITHUB_OUTPUT"
            echo "hf_collection_config=$COLLECTION_CONFIG" >> "$GITHUB_OUTPUT"
            echo "target_repo_owner=$TARGET_HF_REPO_OWNER" >> "$GITHUB_OUTPUT"
            echo "target_repo_name_ext=$TARGET_HF_REPO_NAME_EXT" >> "$GITHUB_OUTPUT"
            echo "source_language_repos=$SOURCE_LANGUAGE_REPOS" >> "$GITHUB_OUTPUT"
            echo "target_language_quantizations=$TARGET_LANGUAGE_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
            echo "target_repo_private=$TARGET_HF_REPO_PRIVATE" >> "$GITHUB_OUTPUT"
            echo "source_vision_repos=$SOURCE_VISION_REPOS" >> "$GITHUB_OUTPUT"
            echo "target_vision_quantizations=$TARGET_VISION_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
            echo "source_guardian_repos=$SOURCE_GUARDIAN_REPOS" >> "$GITHUB_OUTPUT"
            echo "target_guardian_quantizations=$TARGET_GUARDIAN_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
            echo "source_embedding_repos=$SOURCE_EMBEDDING_REPOS" >> "$GITHUB_OUTPUT"
            echo "target_embedding_quantizations=$TARGET_EMBEDDING_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
            echo "source_docling_repos=$SOURCE_DOCLING_REPOS" >> "$GITHUB_OUTPUT"
            echo "target_docling_quantizations=$TARGET_DOCLING_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
            echo "default_artifacts_dir=$DEFAULT_ARTIFACTS_DIR" >> "$GITHUB_OUTPUT"
            echo "artifact_logs_name=$ARTIFACT_LOGS_NAME" >> "$GITHUB_OUTPUT"
            echo "artifact_responses_name=$ARTIFACT_RESPONSES_NAME" >> "$GITHUB_OUTPUT"
            echo "ext_log=$EXT_LOG" >> "$GITHUB_OUTPUT"
            echo "ext_response=$EXT_RESPONSE" >> "$GITHUB_OUTPUT"
            echo "transformers_version_language=$TRANSFORMERS_VERSION_LANGUAGE" >> "$GITHUB_OUTPUT"
            echo "transformers_version_vision=$TRANSFORMERS_VERSION_VISION" >> "$GITHUB_OUTPUT"
            echo "transformers_version_guardian=$TRANSFORMERS_VERSION_GUARDIAN" >> "$GITHUB_OUTPUT"
            echo "transformers_version_embedding=$TRANSFORMERS_VERSION_EMBEDDING" >> "$GITHUB_OUTPUT"
            echo "transformers_version_docling=$TRANSFORMERS_VERSION_DOCLING" >> "$GITHUB_OUTPUT"
            echo "llamacpp_build_tag_language=$LLAMACPP_BUILD_TAG_LANGUAGE" >> "$GITHUB_OUTPUT"
            echo "llamacpp_build_tag_vision=$LLAMACPP_BUILD_TAG_VISION" >> "$GITHUB_OUTPUT"
            echo "llamacpp_build_tag_guardian=$LLAMACPP_BUILD_TAG_GUARDIAN" >> "$GITHUB_OUTPUT"
            echo "llamacpp_build_tag_embedding=$LLAMACPP_BUILD_TAG_EMBEDDING" >> "$GITHUB_OUTPUT"
            echo "llamacpp_build_tag_docling=$LLAMACPP_BUILD_TAG_DOCLING" >> "$GITHUB_OUTPUT"
            echo "def_converted_quant_language=$DEF_CONVERTED_QUANT_LANGUAGE" >> "$GITHUB_OUTPUT"
            echo "def_converted_quant_vision=$DEF_CONVERTED_QUANT_VISION" >> "$GITHUB_OUTPUT"
            echo "def_converted_quant_guardian=$DEF_CONVERTED_QUANT_GUARDIAN" >> "$GITHUB_OUTPUT"
            echo "def_converted_quant_embedding=$DEF_CONVERTED_QUANT_EMBEDDING" >> "$GITHUB_OUTPUT"
            echo "def_converted_quant_docling=$DEF_CONVERTED_QUANT_DOCLING" >> "$GITHUB_OUTPUT"

      - name: Append converted format to BVT test matrices
        id: append-converted-format
        run: |
            echo "ENABLE_BVT_TEST_CONVERTED_FORMAT: $ENABLE_BVT_TEST_CONVERTED_FORMAT"

            if [ "$ENABLE_BVT_TEST_CONVERTED_FORMAT" = "true" ]; then
              echo "✅ Appending converted formats to BVT test matrices"

              # Language models (BF16)
              echo "Appending ${{env.DEF_CONVERTED_QUANT_LANGUAGE}} to TARGET_LANGUAGE_QUANTIZATIONS for BVT"
              TEMP_ARRAY_STRING="${{ join(fromJSON(env.TARGET_LANGUAGE_QUANTIZATIONS), env.ARRAY_SEP) }}"
              if [ "$TEMP_ARRAY_STRING" != "None" ]; then
                TARGET_LANGUAGE_QUANTIZATIONS_BVT="['${TEMP_ARRAY_STRING}${{env.ARRAY_SEP}}${{env.DEF_CONVERTED_QUANT_LANGUAGE}}']"
              else
                TARGET_LANGUAGE_QUANTIZATIONS_BVT="$TARGET_LANGUAGE_QUANTIZATIONS"
              fi
              echo "target_language_quantizations_bvt=$TARGET_LANGUAGE_QUANTIZATIONS_BVT" >> "$GITHUB_OUTPUT"

              # Vision models (F16)
              echo "Appending ${{env.DEF_CONVERTED_QUANT_VISION}} to TARGET_VISION_QUANTIZATIONS for BVT"
              TEMP_ARRAY_STRING="${{ join(fromJSON(env.TARGET_VISION_QUANTIZATIONS), env.ARRAY_SEP) }}"
              if [ "$TEMP_ARRAY_STRING" != "None" ]; then
                TARGET_VISION_QUANTIZATIONS_BVT="['${TEMP_ARRAY_STRING}${{env.ARRAY_SEP}}${{env.DEF_CONVERTED_QUANT_VISION}}']"
              else
                TARGET_VISION_QUANTIZATIONS_BVT="$TARGET_VISION_QUANTIZATIONS"
              fi
              echo "target_vision_quantizations_bvt=$TARGET_VISION_QUANTIZATIONS_BVT" >> "$GITHUB_OUTPUT"

              # Guardian models (BF16)
              # NOTE: Commented out because bf16 BVT test takes longer than 30 minutes and times out
              # echo "Appending ${{env.DEF_CONVERTED_QUANT_GUARDIAN}} to TARGET_GUARDIAN_QUANTIZATIONS for BVT"
              # TEMP_ARRAY_STRING="${{ join(fromJSON(env.TARGET_GUARDIAN_QUANTIZATIONS), env.ARRAY_SEP) }}"
              # if [ "$TEMP_ARRAY_STRING" != "None" ]; then
              #   TARGET_GUARDIAN_QUANTIZATIONS_BVT="['${TEMP_ARRAY_STRING}${{env.ARRAY_SEP}}${{env.DEF_CONVERTED_QUANT_GUARDIAN}}']"
              # else
              #   TARGET_GUARDIAN_QUANTIZATIONS_BVT="$TARGET_GUARDIAN_QUANTIZATIONS"
              # fi
              # Use base quantizations only (without appending bf16)
              TARGET_GUARDIAN_QUANTIZATIONS_BVT="$TARGET_GUARDIAN_QUANTIZATIONS"
              echo "target_guardian_quantizations_bvt=$TARGET_GUARDIAN_QUANTIZATIONS_BVT" >> "$GITHUB_OUTPUT"

              # Embedding models (BF16)
              echo "Appending ${{env.DEF_CONVERTED_QUANT_EMBEDDING}} to TARGET_EMBEDDING_QUANTIZATIONS for BVT"
              TEMP_ARRAY_STRING="${{ join(fromJSON(env.TARGET_EMBEDDING_QUANTIZATIONS), env.ARRAY_SEP) }}"
              if [ "$TEMP_ARRAY_STRING" != "None" ]; then
                TARGET_EMBEDDING_QUANTIZATIONS_BVT="['${TEMP_ARRAY_STRING}${{env.ARRAY_SEP}}${{env.DEF_CONVERTED_QUANT_EMBEDDING}}']"
              else
                TARGET_EMBEDDING_QUANTIZATIONS_BVT="$TARGET_EMBEDDING_QUANTIZATIONS"
              fi
              echo "target_embedding_quantizations_bvt=$TARGET_EMBEDDING_QUANTIZATIONS_BVT" >> "$GITHUB_OUTPUT"

              # Docling models (F16)
              echo "Appending ${{env.DEF_CONVERTED_QUANT_DOCLING}} to TARGET_DOCLING_QUANTIZATIONS for BVT"
              TEMP_ARRAY_STRING="${{ join(fromJSON(env.TARGET_DOCLING_QUANTIZATIONS), env.ARRAY_SEP) }}"
              if [ "$TEMP_ARRAY_STRING" != "None" ]; then
                TARGET_DOCLING_QUANTIZATIONS_BVT="['${TEMP_ARRAY_STRING}${{env.ARRAY_SEP}}${{env.DEF_CONVERTED_QUANT_DOCLING}}']"
              else
                TARGET_DOCLING_QUANTIZATIONS_BVT="$TARGET_DOCLING_QUANTIZATIONS"
              fi
              echo "target_docling_quantizations_bvt=$TARGET_DOCLING_QUANTIZATIONS_BVT" >> "$GITHUB_OUTPUT"
            else
              echo "ℹ️  Skipping append of converted formats (ENABLE_BVT_TEST_CONVERTED_FORMAT=false)"
              echo "target_language_quantizations_bvt=$TARGET_LANGUAGE_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
              echo "target_vision_quantizations_bvt=$TARGET_VISION_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
              echo "target_guardian_quantizations_bvt=$TARGET_GUARDIAN_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
              echo "target_embedding_quantizations_bvt=$TARGET_EMBEDDING_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
              echo "target_docling_quantizations_bvt=$TARGET_DOCLING_QUANTIZATIONS" >> "$GITHUB_OUTPUT"
            fi

      - name: Confirm GitHub output values were set
        run: |
            echo "debug: ${{ steps.set-vars.outputs.debug }}"
            echo "trace: ${{ steps.set-vars.outputs.trace }}"
            echo "enable_language_jobs: ${{ steps.set-vars.outputs.enable_language_jobs }}"
            echo "enable_vision_jobs: ${{ steps.set-vars.outputs.enable_vision_jobs }}"
            echo "enable_guardian_jobs: ${{ steps.set-vars.outputs.enable_guardian_jobs }}"
            echo "enable_embedding_jobs: ${{ steps.set-vars.outputs.enable_embedding_jobs }}"
            echo "enable_docling_jobs: ${{ steps.set-vars.outputs.enable_docling_jobs }}"
            echo "hf_collection_config: '${{ steps.set-vars.outputs.hf_collection_config }}'"
            echo "target_repo_owner: '${{ steps.set-vars.outputs.target_repo_owner }}'"
            echo "target_repo_name_ext: '${{ steps.set-vars.outputs.target_repo_name_ext }}'"
            echo "target_repo_private: '${{ steps.set-vars.outputs.target_repo_private }}'"
            echo "source_language_repos: '${{ steps.set-vars.outputs.source_language_repos }}'"
            echo "target_language_quantizations: '${{ steps.set-vars.outputs.target_language_quantizations }}'"
            echo "source_vision_repos: '${{ steps.set-vars.outputs.source_vision_repos }}'"
            echo "target_vision_quantizations: '${{ steps.set-vars.outputs.target_vision_quantizations }}'"
            echo "source_guardian_repos: '${{ steps.set-vars.outputs.source_guardian_repos }}'"
            echo "target_guardian_quantizations: '${{ steps.set-vars.outputs.target_guardian_quantizations }}'"
            echo "source_embedding_repos: '${{ steps.set-vars.outputs.source_embedding_repos }}'"
            echo "target_embedding_quantizations: '${{ steps.set-vars.outputs.target_embedding_quantizations }}'"
            echo "source_docling_repos: '${{ steps.set-vars.outputs.source_docling_repos }}'"
            echo "target_docling_quantizations: '${{ steps.set-vars.outputs.target_docling_quantizations }}'"
            echo "default_artifacts_dir: '${{ steps.set-vars.outputs.default_artifacts_dir }}'"
            echo "artifact_logs_name: '${{ steps.set-vars.outputs.artifact_logs_name }}'"
            echo "artifact_responses_name: '${{ steps.set-vars.outputs.artifact_responses_name }}'"
            echo "ext_log: '${{ steps.set-vars.outputs.ext_log }}'"
            echo "ext_response: '${{ steps.set-vars.outputs.ext_response }}'"

      - name: List all environment variables
        run: env | sort

      - name: Dump runner context
        env:
          RUNNER_CONTEXT: ${{ toJson(runner) }}
        run: |
          uname -a
          echo "$RUNNER_CONTEXT"

      - name: Dump GitHub context
        env:
          GH_CONTEXT: ${{ toJson(github) }}
        run: echo "$GH_CONTEXT"

      # - name: Dump GitHub quiet
      #   env:
      #     GH_QUIET: ${{ toJson(secrets) }}
      #   run: echo "$GH_QUIET"

  # ===========================================================================
  # Granite LLMs
  # ===========================================================================
  language-create-hf-repos:
    needs: [ environment-setup ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-repos.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      granite_family: "instruct"
      enable_language_jobs: ${{ needs.environment-setup.outputs.enable_language_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repos: ${{ needs.environment-setup.outputs.source_language_repos }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_repo_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  language-convert-hf-to-f16-gguf:
    needs: [ environment-setup, language-create-hf-repos ]
    uses: IBM/gguf/.github/workflows/reusable-convert-hf-language-to-bf16-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_language_repos) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_language }}
      enable_language_jobs: ${{ needs.environment-setup.outputs.enable_language_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      repo_id: ${{ matrix.repo_id }}
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_language }}
      transformers_version: ${{ needs.environment-setup.outputs.transformers_version_language }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}
      ibm_granite_token: ${{ secrets.HF_TOKEN_IBM_GRANITE }}

  language-quantize-upload-gguf:
    needs: [ environment-setup, language-convert-hf-to-f16-gguf ]
    if: ${{ always() && (needs.language-convert-hf-to-f16-gguf.result == 'success' || needs.language-convert-hf-to-f16-gguf.result == 'skipped') }}
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_language_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_language_quantizations) }}
    uses: IBM/gguf/.github/workflows/reusable-quantize-language-gguf-upload.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_language }}
      enable_language_jobs: ${{ needs.environment-setup.outputs.enable_language_jobs == 'true' }}
      hf_collection_config: "${{ needs.environment-setup.outputs.hf_collection_config }}"
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_language }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  language-bvt-quantized-gguf-models:
    needs: [ environment-setup, language-convert-hf-to-f16-gguf, language-quantize-upload-gguf ]
    if: ${{ always() && needs.language-convert-hf-to-f16-gguf.result != 'failure' && (needs.language-quantize-upload-gguf.result == 'success' || needs.language-quantize-upload-gguf.result == 'skipped') }}
    uses: IBM/gguf/.github/workflows/reusable-bvt-language-quantized-models-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_language_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_language_quantizations_bvt) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_language }}
      enable_language_jobs: ${{ needs.environment-setup.outputs.enable_language_jobs == 'true' }}
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      ext_log: "${{ needs.environment-setup.outputs.ext_log }}"
      ext_response: "${{ needs.environment-setup.outputs.ext_response }}"
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  language-lm-eval-gguf-models:
    needs: [ environment-setup, language-bvt-quantized-gguf-models ]
    if: ${{ always() && (needs.language-bvt-quantized-gguf-models.result == 'success' || needs.language-bvt-quantized-gguf-models.result == 'skipped') }}
    uses: IBM/gguf/.github/workflows/reusable-lm-eval-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_language_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_language_quantizations) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_language }}
      enable_language_jobs: ${{ needs.environment-setup.outputs.enable_language_jobs == 'true' }}
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      ext_log: "${{ needs.environment-setup.outputs.ext_log }}"
      ext_response: "${{ needs.environment-setup.outputs.ext_response }}"
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # TODO: change collection from private to public (on success)
  language-create-hf-collections:
    needs: [
      environment-setup,
      language-bvt-quantized-gguf-models,
    ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-collections.yml@main
    with:
      granite_family: "instruct"
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      enable_language_jobs: ${{ needs.environment-setup.outputs.enable_language_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_collection_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # ===========================================================================
  # Granite Vision
  # ===========================================================================
  vision-create-hf-repos:
    needs: [ environment-setup ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-repos.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      granite_family: "vision"
      enable_vision_jobs: ${{ needs.environment-setup.outputs.enable_vision_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repos: ${{ needs.environment-setup.outputs.source_vision_repos }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_repo_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}
  check-vision-llama-cpp-support:
    needs: [ environment-setup, vision-create-hf-repos ]
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_vision_repos) }}
    steps:
      - name: Checkout repository
        uses: actions/checkout@v6
        with:
          sparse-checkout: |
            scripts/is_model_llama_cpp_supported.py
            resources/json/

      - name: Check llama.cpp support for ${{ matrix.repo_id }}
        id: check-support
        run: |
          SUPPORTED=$(python ./scripts/is_model_llama_cpp_supported.py ${{ needs.environment-setup.outputs.hf_collection_config }} ${{ matrix.repo_id }})

          if [ "$SUPPORTED" = "true" ]; then
            echo "✅ Model ${{ matrix.repo_id }} is supported by llama.cpp - BVT/UAT testing will be enabled"
          else
            echo "⚠️ Model ${{ matrix.repo_id }} is NOT supported by llama.cpp - BVT/UAT testing will be skipped"
            echo "   Conversion and quantization will proceed, but inference testing requires llama.cpp support"
          fi


  convert-hf-llava-to-f16-gguf:
    needs: [ environment-setup, vision-create-hf-repos, check-vision-llama-cpp-support ]
    # OLD: uses: IBM/gguf/.github/workflows/reusable-convert-hf-llava-to-bf16-gguf.yml@main
    uses: IBM/gguf/.github/workflows/reusable-convert-hf-vision-to-f16-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_vision_repos) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_vision }}
      enable_vision_jobs: ${{ needs.environment-setup.outputs.enable_vision_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      repo_id: ${{ matrix.repo_id }}
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_vision }}
      transformers_version: ${{ needs.environment-setup.outputs.transformers_version_vision }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}
      ibm_granite_token: ${{ secrets.HF_TOKEN_IBM_GRANITE }}

  quantize-vision-llm-and-upload-gguf:
    needs: [ environment-setup, convert-hf-llava-to-f16-gguf ]
    if: ${{ always() && needs.convert-hf-llava-to-f16-gguf.result == 'success' }}
    uses: IBM/gguf/.github/workflows/reusable-quantize-llava-upload-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_vision_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_vision_quantizations) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_vision }}
      enable_vision_jobs: ${{ needs.environment-setup.outputs.enable_vision_jobs == 'true' }}
      hf_collection_config: "${{ needs.environment-setup.outputs.hf_collection_config }}"
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_vision }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  bvt-vision-quantized-gguf-models:
    needs: [ environment-setup, quantize-vision-llm-and-upload-gguf ]
    if: ${{ always() && (needs.quantize-vision-llm-and-upload-gguf.result == 'success' || needs.quantize-vision-llm-and-upload-gguf.result == 'skipped') }}
    uses: IBM/gguf/.github/workflows/reusable-bvt-vision-quantized-models-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_vision_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_vision_quantizations_bvt) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_vision }}
      enable_vision_jobs: ${{ needs.environment-setup.outputs.enable_vision_jobs == 'true' }}
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      ext_log: "${{ needs.environment-setup.outputs.ext_log }}"
      ext_response: "${{ needs.environment-setup.outputs.ext_response }}"
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # TODO: change collection from private to public (on success)
  vision-create-hf-collections:
    needs: [
      environment-setup,
      bvt-vision-quantized-gguf-models,
    ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-collections.yml@main
    with:
      granite_family: "vision"
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      enable_vision_jobs: ${{ needs.environment-setup.outputs.enable_vision_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_collection_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # ===========================================================================
  # Granite Guardian
  # ===========================================================================
  guardian-create-hf-repos:
    needs: [ environment-setup ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-repos.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      granite_family: "guardian"
      enable_guardian_jobs: ${{ needs.environment-setup.outputs.enable_guardian_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repos: ${{ needs.environment-setup.outputs.source_guardian_repos }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_repo_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  guardian-convert-hf-to-f16-gguf:
    needs: [ environment-setup, guardian-create-hf-repos ]
    uses: IBM/gguf/.github/workflows/reusable-convert-hf-language-to-bf16-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_guardian_repos) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_guardian }}
      enable_guardian_jobs: ${{ needs.environment-setup.outputs.enable_guardian_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      repo_id: ${{ matrix.repo_id }}
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_guardian }}
      transformers_version: ${{ needs.environment-setup.outputs.transformers_version_guardian }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}
      ibm_granite_token: ${{ secrets.HF_TOKEN_IBM_GRANITE }}

  guardian-quantize-upload-gguf:
    needs: [ environment-setup, guardian-convert-hf-to-f16-gguf ]
    if: ${{ always() && (needs.guardian-convert-hf-to-f16-gguf.result == 'success' || needs.guardian-convert-hf-to-f16-gguf.result == 'skipped') }}
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_guardian_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_guardian_quantizations) }}
    uses: IBM/gguf/.github/workflows/reusable-quantize-language-gguf-upload.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_guardian }}
      enable_guardian_jobs: ${{ needs.environment-setup.outputs.enable_guardian_jobs == 'true' }}
      hf_collection_config: "${{ needs.environment-setup.outputs.hf_collection_config }}"
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_guardian }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  guardian-bvt-quantized-gguf-models:
    needs: [ environment-setup, guardian-quantize-upload-gguf ]
    if: ${{ always() && (needs.guardian-quantize-upload-gguf.result == 'success' || needs.guardian-quantize-upload-gguf.result == 'skipped') }}
    uses: IBM/gguf/.github/workflows/reusable-bvt-guardian-quantized-models-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_guardian_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_guardian_quantizations_bvt) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_guardian }}
      enable_guardian_jobs: ${{ needs.environment-setup.outputs.enable_guardian_jobs == 'true' }}
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      ext_log: "${{ needs.environment-setup.outputs.ext_log }}"
      ext_response: "${{ needs.environment-setup.outputs.ext_response }}"
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # TODO: change collection from private to public (on success)
  guardian-create-hf-collections:
    needs: [
      environment-setup,
      guardian-bvt-quantized-gguf-models,
    ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-collections.yml@main
    with:
      granite_family: "guardian"
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      enable_guardian_jobs: ${{ needs.environment-setup.outputs.enable_guardian_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_collection_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # ===========================================================================
  # Granite Embedding
  # ===========================================================================
  embedding-create-hf-repos:
    needs: [ environment-setup ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-repos.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      granite_family: "embedding"
      enable_embedding_jobs: ${{ needs.environment-setup.outputs.enable_embedding_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repos: ${{ needs.environment-setup.outputs.source_embedding_repos }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_repo_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # Note: Granite 3.3 embedding models have different architectures:
  # - Non-r2 models (e.g., granite-embedding-30m-english): RoBERTa architecture
  # - R2 models (e.g., granite-embedding-english-r2): ModernBERT architecture
  #   ModernBERT support was added in llama.cpp PR: https://github.com/ggml-org/llama.cpp/pull/18330
  # This workflow includes baseline similarity testing before GGUF conversion
  embedding-convert-hf-to-f16-gguf:
    needs: [ environment-setup, embedding-create-hf-repos ]
    uses: IBM/gguf/.github/workflows/reusable-convert-hf-embedding-to-bf16-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_embedding_repos) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_embedding }}
      enable_embedding_jobs: ${{ needs.environment-setup.outputs.enable_embedding_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      repo_id: ${{ matrix.repo_id }}
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_embedding }}
      transformers_version: ${{ needs.environment-setup.outputs.transformers_version_embedding }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}
      ibm_granite_token: ${{ secrets.HF_TOKEN_IBM_GRANITE }}

  embedding-quantize-upload-gguf:
    needs: [ environment-setup, embedding-convert-hf-to-f16-gguf ]
    if: ${{ always() && (needs.embedding-convert-hf-to-f16-gguf.result == 'success' || needs.embedding-convert-hf-to-f16-gguf.result == 'skipped') }}
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_embedding_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_embedding_quantizations) }}
    uses: IBM/gguf/.github/workflows/reusable-quantize-language-gguf-upload.yml@main
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_embedding }}
      enable_embedding_jobs: ${{ needs.environment-setup.outputs.enable_embedding_jobs == 'true' }}
      hf_collection_config: "${{ needs.environment-setup.outputs.hf_collection_config }}"
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      converted_quantization: ${{ needs.environment-setup.outputs.def_converted_quant_embedding }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  embedding-bvt-quantized-gguf-models:
    needs: [ environment-setup, embedding-quantize-upload-gguf ]
    if: ${{ always() && (needs.embedding-quantize-upload-gguf.result == 'success' || needs.embedding-quantize-upload-gguf.result == 'skipped') }}
    uses: IBM/gguf/.github/workflows/reusable-bvt-embedding-quantized-models-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_embedding_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_embedding_quantizations_bvt) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      trace: ${{ needs.environment-setup.outputs.trace == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_embedding }}
      enable_embedding_jobs: ${{ needs.environment-setup.outputs.enable_embedding_jobs == 'true' }}
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      ext_log: "${{ needs.environment-setup.outputs.ext_log }}"
      ext_response: "${{ needs.environment-setup.outputs.ext_response }}"
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  embedding-rag-test-quantized-gguf-models:
    needs: [ environment-setup, embedding-bvt-quantized-gguf-models ]
    if: ${{ always() && needs.embedding-bvt-quantized-gguf-models.result == 'success' && needs.environment-setup.outputs.enable_rag_testing == 'true' }}
    uses: IBM/gguf/.github/workflows/reusable-rag-test-embedding-models-gguf.yml@main
    permissions:
      contents: write
      packages: write
    strategy:
      fail-fast: false
      matrix:
        repo_id: ${{ fromJson(needs.environment-setup.outputs.source_embedding_repos) }}
        quantization: ${{ fromJson(needs.environment-setup.outputs.target_embedding_quantizations_bvt) }}
    with:
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      llamacpp_build_tag: ${{ needs.environment-setup.outputs.llamacpp_build_tag_embedding }}
      enable_rag_testing: ${{ needs.environment-setup.outputs.enable_rag_testing == 'true' }}
      target_repo_owner: "${{ needs.environment-setup.outputs.target_repo_owner }}"
      target_repo_name_ext: "${{ needs.environment-setup.outputs.target_repo_name_ext }}"
      repo_id: "${{ matrix.repo_id }}"
      quantization: "${{ matrix.quantization }}"
      top_k: 3
      min_score: 0.5
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}

  # TODO: change collection from private to public (on success)
  embedding-create-hf-collections:
    needs: [
      environment-setup,
      embedding-bvt-quantized-gguf-models,
      embedding-rag-test-quantized-gguf-models,
    ]
    uses: IBM/gguf/.github/workflows/reusable-create-hf-collections.yml@main
    with:
      granite_family: "embedding"
      debug: ${{ needs.environment-setup.outputs.debug == 'true' }}
      enable_embedding_jobs: ${{ needs.environment-setup.outputs.enable_embedding_jobs == 'true' }}
      hf_collection_config: ${{ needs.environment-setup.outputs.hf_collection_config }}
      target_repo_owner: ${{ needs.environment-setup.outputs.target_repo_owner }}
      target_repo_name_ext: ${{ needs.environment-setup.outputs.target_repo_name_ext }}
      target_collection_private: ${{ needs.environment-setup.outputs.target_repo_private == 'true' }}
    secrets:
      hf_token: ${{ secrets.HF_TOKEN_TEST }}