extend modelbuilder to build Olmo3, SmolLM3 and other models

.github/workflows/modelbuilder_fast_tests.yml

@@ -0,0 +1,135 @@
+name: Fast Tests
+
+on:
+  push:
+    branches: ["main"]
+  pull_request:
+    branches: ["main"]
+
+permissions:
+  contents: read
+
+jobs:
+  tests:
+    name: ci (${{ matrix.os }}, py${{ matrix.python-version }}, torch-${{ matrix.torch-version || 'stable' }}, transformers-${{ matrix.transformers-version || 'latest' }})
+    runs-on: ${{ matrix.os }}
+    permissions:
+      contents: read
+      id-token: write
+      pull-requests: write
+    strategy:
+      fail-fast: false
+      matrix:
+        os: ["ubuntu-latest"]
+        python-version: ["3.13"]
+        transformers-version: ["5.5"]
+        torch-version: ["2.11"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      # we install torch first to avoid downloading any CUDA dependency
+
+      - name: Install nightly pytorch (cpu)
+        if: matrix.torch-version == 'nightly'
+        run: pip install --pre --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu
+
+      - name: Install pytorch 2.11 (cpu)
+        if: matrix.torch-version == '2.11' && matrix.os != 'macos-latest'
+        run: pip install torch==2.11.0+cpu torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
+
+      - name: Install dependencies
+        run: pip install torch "transformers==${{ matrix.transformers-version }}" tokenizers pandas openpyxl optree onnxruntime onnxruntime-genai peft py-cpuinfo tabulate
+
+      - name: Install dependencies
+        run: pip install -r test/python/requirements.txt
+
+      - name: ls
+        run: ls . -l
+
+      - name: Run fast tests
+        run: DOCLEAN=1 PYTHONPATH=./src/python/py pytest test/python/models/fast -v --cov=models --cov-report=xml --cov-report=term-missing --test_models=dump_models
+
+      # Code to upload coverage report.
+      # - name: Upload coverage reports to Codecov
+      #   uses: codecov/codecov-action@v5
+      #   with:
+      #    token: ${{ secrets.CODECOV_TOKEN }}
+      #     flags: modelbuilder-fast-tests
+
+      - name: Upload stats artifact
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: stats-${{ matrix.os }}-py${{ matrix.python-version }}-torch${{ matrix.torch-version }}-transformers${{ matrix.transformers-version }}
+          path: stats/
+          if-no-files-found: ignore
+
+      - name: Post test results as PR comment
+        if: github.event_name == 'pull_request' && always()
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const fs = require('fs');
+            const path = require('path');
+
+            const statsDir = 'stats';
+            if (!fs.existsSync(statsDir)) {
+              console.log('No stats directory found, skipping PR comment');
+              return;
+            }
+
+            const files = fs.readdirSync(statsDir).filter(f => f.endsWith('.csv'));
+            if (files.length === 0) {
+              console.log('No CSV result files found in stats/, skipping PR comment');
+              return;
+            }
+
+            const matrixConfig = 'transformers=${{ matrix.transformers-version }}, torch=${{ matrix.torch-version }}, os=${{ matrix.os }}';
+            const marker = `<!-- fast-test-results-${{ matrix.transformers-version }}-${{ matrix.torch-version }}-${{ matrix.os }} -->`;
+            let body = `${marker}\n## Fast Test Results\n\n**Config:** ${matrixConfig}\n\n`;
+
+            for (const file of files.sort()) {
+              const content = fs.readFileSync(path.join(statsDir, file), 'utf8').trim();
+              const lines = content.split('\n');
+              if (lines.length < 2) continue;
+
+              body += `<details><summary>${file}</summary>\n\n`;
+
+              // Convert CSV to markdown table
+              const rows = lines.map(l => l.split(',').map(c => c.trim()));
+              const header = rows[0];
+              body += '| ' + header.join(' | ') + ' |\n';
+              body += '| ' + header.map(() => '---').join(' | ') + ' |\n';
+              for (const row of rows.slice(1)) {
+                body += '| ' + row.join(' | ') + ' |\n';
+              }
+              body += '\n</details>\n\n';
+            }
+
+            // Update existing comment for this matrix combination or create a new one
+            const { data: comments } = await github.rest.issues.listComments({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: context.issue.number,
+            });
+
+            const existing = comments.find(c => c.body && c.body.includes(marker));
+            if (existing) {
+              await github.rest.issues.updateComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                comment_id: existing.id,
+                body,
+              });
+            } else {
+              await github.rest.issues.createComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                issue_number: context.issue.number,
+                body,
+              });
+            }

.gitignore

@@ -16,6 +16,8 @@ src/csharp/AssemblyInfo.cs
 *.csv
 .idea
 cache_dir
+dump_models
+stats
 example-models
 *.onnx
 *.onnx.data

src/python/py/models/builder.py

@@ -16,39 +16,9 @@
 
 import onnx_ir as ir
 import torch
-from builders import (
-    ChatGLMModel,
-    ErnieModel,
-    Gemma2Model,
-    Gemma3Model,
-    GemmaModel,
-    GPTOSSModel,
-    GraniteModel,
-    InternLM2Model,
-    LlamaModel,
-    MistralModel,
-    Model,
-    NemotronModel,
-    OLMoModel,
-    Phi3MiniLongRoPEModel,
-    Phi3MiniModel,
-    Phi3MoELongRoPEModel,
-    Phi3SmallLongRoPEModel,
-    Phi3SmallModel,
-    Phi3VModel,
-    Phi4MMModel,
-    PhiModel,
-    Qwen3Model,
-    Qwen3VLTextModel,
-    Qwen25VLTextModel,
-    Qwen35TextModel,
-    QwenModel,
-    SmolLM3Model,
-    WhisperModel,
-)
-from transformers import (
-    AutoConfig,
-)
+from transformers import AutoConfig
+
+from .builders import Model
 
 
 def check_extra_options(kv_pairs, execution_provider):
@@ -211,19 +181,30 @@ def create_model(
         # Quantized ChatGLM model has ChatGLMForConditionalGeneration as architecture whereas HF model as the latter
         config.bos_token_id = 1
         config.hidden_act = "swiglu"
+        from .builders.chatglm import ChatGLMModel
+
         onnx_model = ChatGLMModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
         onnx_model.model_type = "chatglm"
     elif config.architectures[0] == "Ernie4_5ForCausalLM":
+        from .builders.ernie import ErnieModel
+
         onnx_model = ErnieModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "GemmaForCausalLM":
+        from .builders.gemma import GemmaModel
+
         onnx_model = GemmaModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Gemma2ForCausalLM":
-        print("WARNING: This model loses accuracy with float16 precision. It is recommended to set `--precision bf16` or `--precision int4 --extra_options use_cuda_bf16=true` by default.")
+        print(
+            "WARNING: This model loses accuracy with float16 precision. It is recommended to set `--precision bf16` or `--precision int4 --extra_options use_cuda_bf16=true` by default."
+        )
+        from .builders.gemma import Gemma2Model
+
         onnx_model = Gemma2Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Gemma3ForCausalLM":
         print("WARNING: This model loses accuracy with float16 precision. It is recommended to set `--precision bf16` or `--precision int4 --extra_options use_cuda_bf16=true` by default.")
+        from .builders.gemma import Gemma3Model
+
         onnx_model = Gemma3Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
-        onnx_model.model_type = "gemma3_text"
     elif config.architectures[0] == "Gemma3ForConditionalGeneration":
         text_config = config.text_config
         for key in text_config:
@@ -232,49 +213,122 @@ def create_model(
         print("WARNING: This model loses accuracy with float16 precision. It is recommended to set `--precision bf16` or `--precision int4 --extra_options use_cuda_bf16=true` by default.")
         print("WARNING: This is only generating the text component of the model. Setting `--extra_options exclude_embeds=true` by default.")
         extra_options["exclude_embeds"] = True
+        from .builders.gemma import Gemma3Model
+
         onnx_model = Gemma3Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
+        onnx_model.model_type = "gemma3"
     elif config.architectures[0] == "GptOssForCausalLM":
         print("WARNING: This model only supports symmetric quantization for `QMoE`.")
         if hasattr(config, "quantization_config") and config.quantization_config.get("quant_method") != "quark":
             delattr(config, "quantization_config")
+        from .builders.gptoss import GPTOSSModel
+
         onnx_model = GPTOSSModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "GraniteForCausalLM":
+        from .builders.granite import GraniteModel
+
         onnx_model = GraniteModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "InternLM2ForCausalLM":
+        from .builders.internlm import InternLM2Model
+
         onnx_model = InternLM2Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "LlamaForCausalLM":
+        from .builders.llama import LlamaModel
+
         onnx_model = LlamaModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "MistralForCausalLM":
+        from .builders.mistral import MistralModel
+
         onnx_model = MistralModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
+    elif config.architectures[0] == "Ministral3ForCausalLM":
+        from .builders.mistral import Ministral3TextModel
+
+        if hasattr(config, "quantization_config"):
+            # Remove FP8 quantization_config to avoid wrong weight loading via
+            # QuantModel; Ministral3TextModel.load_weights handles dequantization.
+            delattr(config, "quantization_config")
+        onnx_model = Ministral3TextModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
+    elif config.architectures[0] == "Mistral3ForConditionalGeneration":
+        if hasattr(config, "quantization_config"):
+            # Remove FP8 quantization_config to avoid wrong weight loading via
+            # QuantModel; Ministral3TextModel.load_weights handles dequantization.
+            delattr(config, "quantization_config")
+        print(
+            "WARNING: Exporting the text decoder with `exclude_embeds=true` and a separate "
+            "vision encoder (vision_encoder.onnx) for Mistral3ForConditionalGeneration."
+        )
+        from .builders.mistral import Ministral3ConditionalGenerationModel
+
+        onnx_model = Ministral3ConditionalGenerationModel(
+            config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options
+        )
+    elif config.architectures[0] == "MistralNeMoForCausalLM":
+        from .builders.mistral import MistralNeMoModel
+
+        onnx_model = MistralNeMoModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "NemotronForCausalLM":
+        from .builders.nemotron import NemotronModel
+
         onnx_model = NemotronModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
+    elif config.architectures[0] == "NemotronHForCausalLM":
+        from .builders.nemotron import NemotronHModel
+
+        onnx_model = NemotronHModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "OlmoForCausalLM":
+        from .builders.olmo import OLMoModel
+
         onnx_model = OLMoModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
+    elif config.architectures[0] == "Olmo2ForCausalLM":
+        from .builders.olmo import OLMo2Model
+
+        onnx_model = OLMo2Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
+    elif config.architectures[0] == "Olmo3ForCausalLM":
+        from .builders.olmo import OLMo3Model
+
+        onnx_model = OLMo3Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "PhiForCausalLM":
+        from .builders.phi import PhiModel
+
         onnx_model = PhiModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Phi3ForCausalLM" and config.max_position_embeddings == config.original_max_position_embeddings:
+        from .builders.phi import Phi3MiniModel
+
         onnx_model = Phi3MiniModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Phi3ForCausalLM" and config.max_position_embeddings != config.original_max_position_embeddings:
+        from .builders.phi import Phi3MiniLongRoPEModel
+
         onnx_model = Phi3MiniLongRoPEModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "PhiMoEForCausalLM" and config.max_position_embeddings != config.original_max_position_embeddings:
         print("WARNING: This model only works for CUDA currently because `MoE` is only supported for CUDA in ONNX Runtime. Setting `--execution_provider cuda` by default.")
         print("WARNING: This model currently only supports the quantized version. Setting `--precision int4` by default.")
+        from .builders.phi import Phi3MoELongRoPEModel
+
         execution_provider = "cuda"
         onnx_dtype = set_onnx_dtype("int4", extra_options)
         onnx_model = Phi3MoELongRoPEModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Phi3SmallForCausalLM" and config.max_position_embeddings == config.original_max_position_embeddings:
+        from .builders.phi import Phi3SmallModel
+
         onnx_model = Phi3SmallModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Phi3SmallForCausalLM" and config.max_position_embeddings != config.original_max_position_embeddings:
+        from .builders.phi import Phi3SmallLongRoPEModel
+
         onnx_model = Phi3SmallLongRoPEModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Phi3VForCausalLM":
         print("WARNING: This is only generating the text component of the model. Setting `--extra_options exclude_embeds=true` by default.")
         extra_options["exclude_embeds"] = True
+        from .builders.phi import Phi3VModel
+
         onnx_model = Phi3VModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Phi4MMForCausalLM":
         print("WARNING: This is only generating the text component of the model. Setting `--extra_options exclude_embeds=true` by default.")
         extra_options["exclude_embeds"] = True
+        from .builders.phi import Phi4MMModel
+
         onnx_model = Phi4MMModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Qwen2ForCausalLM":
+        from .builders.qwen import QwenModel
+
         onnx_model = QwenModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Qwen2_5_VLForConditionalGeneration":
         text_config = config.text_config
@@ -283,10 +337,16 @@ def create_model(
                 setattr(config, key, getattr(text_config, key))
         print("WARNING: This is only generating the text component of the model. Setting `--extra_options exclude_embeds=true` by default.")
         extra_options["exclude_embeds"] = True
+        from .builders.qwen import Qwen25VLTextModel
+
         onnx_model = Qwen25VLTextModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Qwen3ForCausalLM":
+        from .builders.qwen import Qwen3Model
+
         onnx_model = Qwen3Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Qwen3_5ForConditionalGeneration":
+        from .builders.qwen import Qwen35TextModel
+
         onnx_model = Qwen35TextModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "Qwen3VLForConditionalGeneration":
         text_config = config.text_config
@@ -295,10 +355,16 @@ def create_model(
                 setattr(config, key, getattr(text_config, key))
         print("WARNING: This is only generating the text component of the model. Setting `--extra_options exclude_embeds=true` by default.")
         extra_options["exclude_embeds"] = True
+        from .builders.qwen import Qwen3VLTextModel
+
         onnx_model = Qwen3VLTextModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "SmolLM3ForCausalLM":
+        from .builders.smollm import SmolLM3Model
+
         onnx_model = SmolLM3Model(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config.architectures[0] == "WhisperForConditionalGeneration":
+        from .builders.whisper import WhisperModel
+
         onnx_model = WhisperModel(config, io_dtype, onnx_dtype, execution_provider, cache_dir, extra_options)
     elif config_only:
         # Create base Model class to guess model attributes