make sure to validate the config before normalizing so defaults get set (#2554)

* make sure to validate the config before normalizing so defaults get set

* validation not needed for particular test

* remove duplicate validations

* set qlora correctly

.github/workflows/base.yml

@@ -46,6 +46,18 @@ jobs:
             python_version: "3.11"
             pytorch: 2.6.0
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+          - cuda: "126"
+            cuda_version: 12.6.3
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.7.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
+          - cuda: "128"
+            cuda_version: 12.6.3
+            cudnn_version: ""
+            python_version: "3.11"
+            pytorch: 2.7.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
           - cuda: "128"
             cuda_version: 12.8.1
             cudnn_version: ""

.github/workflows/main.yml

@@ -31,6 +31,11 @@ jobs:
             pytorch: 2.6.0
             axolotl_extras: vllm
             is_latest: true
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            axolotl_extras: vllm
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -93,6 +98,11 @@ jobs:
             pytorch: 2.6.0
             axolotl_extras:
             is_latest: true
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -138,7 +148,7 @@ jobs:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.4.1
+            pytorch: 2.6.0
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:

.github/workflows/multi-gpu-e2e.yml

@@ -45,6 +45,13 @@ jobs:
             axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            axolotl_extras:
+            num_gpus: 2
+            nightly_build: "true"
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     steps:
@@ -67,6 +74,7 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.multigpu

.github/workflows/tests-nightly.yml

@@ -147,6 +147,7 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests

.github/workflows/tests.yml

@@ -49,7 +49,7 @@ jobs:
       max-parallel: 2
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.4.1", "2.5.1", "2.6.0"]
+        pytorch_version: ["2.4.1", "2.5.1", "2.6.0", "2.7.0"]
     timeout-minutes: 20
 
     steps:
@@ -109,6 +109,7 @@ jobs:
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v5
         with:
+          token: ${{ secrets.CODECOV_TOKEN }}
           files: ./coverage.xml
           flags: unittests,pytorch-${{ matrix.pytorch_version }}
           fail_ci_if_error: false
@@ -241,6 +242,7 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests
@@ -268,6 +270,12 @@ jobs:
             pytorch: 2.5.1
             num_gpus: 1
             axolotl_extras: vllm
+          - cuda: 126
+            cuda_version: 12.6.3
+            python_version: "3.11"
+            pytorch: 2.7.0
+            num_gpus: 1
+            axolotl_extras:
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -288,6 +296,7 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
+          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests

cicd/cicd.sh

@@ -9,8 +9,7 @@ pytest -v --durations=10 -n8 \
   --ignore=tests/patched/ \
   --ignore=tests/cli \
   /workspace/axolotl/tests/ \
-  --cov=axolotl \
-  --cov-report=xml:coverage.xml
+  --cov=axolotl
 
 # Run lora kernels tests with coverage append
 pytest -v --durations=10 \
@@ -51,11 +50,6 @@ pytest -v --durations=10 \
   /workspace/axolotl/tests/e2e/ \
   --cov=axolotl \
   --cov-append \
-  --cov-report=xml:coverage.xml
+  --cov-report=xml:e2e-coverage.xml
 
-# Upload coverage to Codecov
-if [ -f e2e-coverage.xml ]; then
-  codecov -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION}
-else
-  echo "Coverage file not found. Coverage report may have failed."
-fi
+codecov upload-process -t $CODECOV_TOKEN -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION}

cicd/e2e_tests.py

@@ -28,6 +28,7 @@ df_args = {
     "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
     "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
     "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
+    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
     "HF_HOME": "/workspace/data/huggingface-cache/hub",
 }
 

cicd/multigpu.py

@@ -29,6 +29,7 @@ df_args = {
     "CUDA": os.environ.get("CUDA", "121"),
     "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
     "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
+    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
     "HF_HOME": "/workspace/data/huggingface-cache/hub",
 }
 

cicd/multigpu.sh

@@ -6,13 +6,13 @@ pytest -v -n2 \
   --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
   --ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
   /workspace/axolotl/tests/e2e/multigpu/ \
-  --cov=axolotl \
-  --cov-report=xml:multigpu-coverage.xml
+  --cov=axolotl
 
-pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/solo/ \
+# Run solo tests with coverage append
+pytest -v --durations=10 -n1 \
+  /workspace/axolotl/tests/e2e/multigpu/solo/ \
   --cov=axolotl \
-  --cov-append \
-  --cov-report=xml:multigpu-coverage.xml
+  --cov-append
 
 pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/patched/ \
   --cov=axolotl \
@@ -20,8 +20,4 @@ pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/patched/ \
   --cov-report=xml:multigpu-coverage.xml
 
 # Upload coverage to Codecov
-if [ -f multigpu-coverage.xml ]; then
-  codecov -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION}
-else
-  echo "Coverage file not found. Coverage report may have failed."
-fi
+codecov upload-process -t $CODECOV_TOKEN -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION}

codecov.yml

@@ -49,3 +49,6 @@ comment:
   require_changes: no
   require_base: no
   require_head: yes
+
+github_checks:
+  annotations: false

docker/Dockerfile-base

@@ -37,3 +37,7 @@ RUN git lfs install --skip-repo && \
     pip3 install awscli && \
     # The base image ships with `pydantic==1.8.2` which is not working
     pip3 install -U --no-cache-dir pydantic==1.10.10
+
+RUN if [ "$PYTORCH_VERSION" = "2.7.0" ] ; then \
+        pip3 install flash-attn==2.7.4.post1; \
+    fi

docs/cli.qmd

@@ -199,6 +199,17 @@ output_dir: # Directory to save evaluation results
 
 See [LM Eval Harness](https://github.com/EleutherAI/lm-evaluation-harness) for more details.
 
+### delinearize-llama4
+
+Delinearizes a Llama 4 linearized model into a regular HuggingFace Llama 4 model. This only works with the non-quantized linearized model.
+
+```bash
+axolotl delinearize-llama4 --model path/to/model_dir --output path/to/output_dir
+```
+
+This would be necessary to use with other frameworks. If you have an adapter, merge it with the non-quantized linearized model before delinearizing.
+
+
 ## Legacy CLI Usage
 
 While the new Click-based CLI is preferred, Axolotl still supports the legacy module-based CLI:

docs/installation.qmd

@@ -19,6 +19,12 @@ This guide covers all the ways you can install and set up Axolotl for your envir
 
 ## Installation Methods {#sec-installation-methods}
 
+::: {.callout-important}
+Please make sure to have Pytorch installed before installing Axolotl in your local environment.
+
+Follow the instructions at: [https://pytorch.org/get-started/locally/](https://pytorch.org/get-started/locally/)
+:::
+
 ### PyPI Installation (Recommended) {#sec-pypi}
 
 ```{.bash}

examples/glm4/qlora-32b.yaml

@@ -0,0 +1,62 @@
+base_model: THUDM/GLM-4-32B-0414
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+load_in_4bit: true
+
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0
+output_dir: ./outputs/qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 2048
+sample_packing: true
+eval_sample_packing: true
+pad_to_sequence_len: true
+
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: false
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
+warmup_steps: 10
+evals_per_epoch: 1
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:

examples/llama-4/README.md

@@ -26,3 +26,11 @@ Multi-GPU (4xH100) for Llama 4 Scout uses 62.8GB VRAM/GPU @ 4k contenxt length @
 ### Llama 4 Maverick 17Bx128Experts (400B)
 
 Coming Soon
+
+## Delinearized Llama 4 Models
+
+We provide a script to delinearize Llama 4 linearized models into regular HuggingFace Llama 4 models.
+
+```bash
+axolotl delinearize-llama4 --model path/to/model_dir --output path/to/output_dir
+```

requirements-tests.txt

@@ -1,4 +1,5 @@
 codecov
+codecov-cli
 pytest
 pytest-cov
 pytest-retry

requirements.txt

@@ -19,6 +19,7 @@ datasets==3.5.0
 deepspeed>=0.15.4
 trl==0.16.1
 hf_xet==1.0.0
+hqq==0.2.5
 
 optimum==1.16.2
 hf_transfer

setup.py

@@ -51,7 +51,7 @@ def parse_requirements(extras_require_map):
             try:
                 torch_version = version("torch")
             except PackageNotFoundError:
-                torch_version = "2.5.1"
+                torch_version = "2.6.0"  # default to torch 2.6
             _install_requires.append(f"torch=={torch_version}")
 
             version_match = re.match(r"^(\d+)\.(\d+)(?:\.(\d+))?", torch_version)
@@ -64,9 +64,15 @@ def parse_requirements(extras_require_map):
             else:
                 raise ValueError("Invalid version format")
 
-            if (major, minor) >= (2, 6):
+            if (major, minor) >= (2, 7):
                 _install_requires.pop(_install_requires.index(xformers_version))
-                _install_requires.append("xformers==0.0.29.post2")
+                # _install_requires.append("xformers==0.0.29.post3")  # xformers seems to be hard pinned to 2.6.0
+                extras_require_map["vllm"] = ["vllm==0.8.3"]
+            elif (major, minor) >= (2, 6):
+                _install_requires.pop(_install_requires.index(xformers_version))
+                _install_requires.append(
+                    "xformers==0.0.29.post2"
+                )  # vllm needs post2 w torch 2.6
                 extras_require_map["vllm"] = ["vllm==0.8.3"]
             elif (major, minor) >= (2, 5):
                 _install_requires.pop(_install_requires.index(xformers_version))

src/axolotl/cli/args.py

@@ -39,16 +39,16 @@ class TrainerCliArgs:
 class VllmServeCliArgs:
     """Dataclass with CLI arguments for `axolotl vllm-serve` command."""
 
-    tensor_parallel_size: int = field(
-        default=1,
+    tensor_parallel_size: Optional[int] = field(
+        default=None,
         metadata={"help": "Number of tensor parallel workers to use."},
     )
-    host: str = field(
-        default="0.0.0.0",  # nosec B104
+    host: Optional[str] = field(
+        default=None,  # nosec B104
         metadata={"help": "Host address to run the server on."},
     )
-    port: int = field(
-        default=8000,
+    port: Optional[int] = field(
+        default=None,
         metadata={"help": "Port to run the server on."},
     )
     gpu_memory_utilization: Optional[float] = field(

src/axolotl/core/trainers/grpo/__init__.py

@@ -40,8 +40,8 @@ class GRPOStrategy:
 
         if trl.use_vllm:
             grpo_args_kwargs["use_vllm"] = trl.use_vllm
-            grpo_args_kwargs["vllm_server_host"] = trl.vllm_server_host
-            grpo_args_kwargs["vllm_server_port"] = trl.vllm_server_port
+            grpo_args_kwargs["vllm_server_host"] = trl.vllm_server_host or trl.vllm.host
+            grpo_args_kwargs["vllm_server_port"] = trl.vllm_server_port or trl.vllm.port
             if trl.vllm_server_timeout:
                 grpo_args_kwargs["vllm_server_timeout"] = trl.vllm_server_timeout
             if trl.vllm_guided_decoding_regex:

src/axolotl/integrations/cut_cross_entropy/README.md

@@ -47,6 +47,8 @@ cut_cross_entropy: true
 - qwen2
 - cohere
 - cohere2
+- glm
+- glm4
 
 ## Citation
 

src/axolotl/integrations/cut_cross_entropy/monkeypatch/glm4.py

@@ -0,0 +1,57 @@
+"""GLM 4 patch. GLM family inherits from Llama."""
+
+from types import MethodType
+
+import transformers
+from cut_cross_entropy.transformers.utils import (
+    PatchOptions,
+    TransformersModelT,
+)
+
+
+def patch_glm(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+
+    # Set the _PATCH_OPTS in the llama patch file
+    import cut_cross_entropy.transformers.llama as llama_patch
+
+    llama_patch._PATCH_OPTS = patch_options  # pylint: disable=protected-access
+
+    from cut_cross_entropy.transformers.llama import cce_forward
+    from transformers.models.glm import modeling_glm
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_glm.GlmForCausalLM
+        ), f"Expected a GlmForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward, maybe_model)
+        return maybe_model
+
+    modeling_glm.GlmForCausalLM.forward = cce_forward
+    return None
+
+
+def patch_glm4(
+    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
+    patch_options: PatchOptions,
+) -> TransformersModelT | None:
+
+    # Set the _PATCH_OPTS in the llama patch file
+    import cut_cross_entropy.transformers.llama as llama_patch
+
+    llama_patch._PATCH_OPTS = patch_options  # pylint: disable=protected-access
+
+    from cut_cross_entropy.transformers.llama import cce_forward
+    from transformers.models.glm4 import modeling_glm4
+
+    if isinstance(maybe_model, transformers.PreTrainedModel):
+        assert isinstance(
+            maybe_model, modeling_glm4.Glm4ForCausalLM
+        ), f"Expected a Glm4ForCausalLM model. Got {type(maybe_model)}."
+        maybe_model.forward = MethodType(cce_forward, maybe_model)
+        return maybe_model
+
+    modeling_glm4.Glm4ForCausalLM.forward = cce_forward
+    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/patch.py

@@ -20,6 +20,10 @@ from axolotl.integrations.cut_cross_entropy.monkeypatch.gemma3 import (
     patch_gemma3,
     patch_gemma3_text,
 )
+from axolotl.integrations.cut_cross_entropy.monkeypatch.glm4 import (
+    patch_glm,
+    patch_glm4,
+)
 from axolotl.integrations.cut_cross_entropy.monkeypatch.llama4 import (
     patch_llama4,
     patch_llama4_text,
@@ -45,6 +49,8 @@ CUT_CROSS_ENTROPY_MODEL_MAPPING = {
     "qwen2": patch_qwen2,
     "cohere": patch_cohere,
     "cohere2": patch_cohere2,
+    "glm": patch_glm,
+    "glm4": patch_glm4,
 }
 
 

src/axolotl/monkeypatch/multipack.py

@@ -31,6 +31,8 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
     "starcoder2",
     "deepseek_v2",
     "deepseek_v3",
+    "glm",
+    "glm4",
 ]
 
 

src/axolotl/monkeypatch/relora.py

@@ -272,7 +272,7 @@ class ReLoRAScheduler(LRScheduler):
         self.warmup_steps = warmup_steps
         self.anneal_steps = anneal_steps
         self.min_lr_scale = min_lr_scale
-        super().__init__(optimizer, inner_schedule.last_epoch, inner_schedule.verbose)
+        super().__init__(optimizer, inner_schedule.last_epoch)
 
     def get_lr(self) -> float:
         self.inner_schedule.last_epoch = self.last_epoch

src/axolotl/utils/schemas/vllm.py

@@ -36,3 +36,11 @@ class VllmConfig(BaseModel):
         default=None,
         json_schema_extra={"description": "Enable prefix caching for VLLM"},
     )
+    host: str | None = Field(
+        default="0.0.0.0",  # nosec B104
+        json_schema_extra={"description": "Host for the vLLM server to start on"},
+    )
+    port: int | None = Field(
+        default=8000,
+        json_schema_extra={"description": "Port of the vLLM server to start on"},
+    )

tests/e2e/integrations/test_cut_cross_entropy.py

@@ -8,7 +8,7 @@ from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
 from axolotl.utils import get_pytorch_version
-from axolotl.utils.config import normalize_config, prepare_plugins
+from axolotl.utils.config import normalize_config, prepare_plugins, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists
@@ -56,6 +56,7 @@ class TestCutCrossEntropyIntegration:
     # pylint: disable=redefined-outer-name
     def test_llama_w_cce(self, min_cfg, temp_dir):
         cfg = DictDefault(min_cfg)
+        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
@@ -101,6 +102,7 @@ class TestCutCrossEntropyIntegration:
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
@@ -129,6 +131,7 @@ class TestCutCrossEntropyIntegration:
                 attention_type: True,
             }
         )
+        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()

tests/e2e/integrations/test_liger.py

@@ -5,7 +5,7 @@ Simple end-to-end test for Liger integration
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, prepare_plugins
+from axolotl.utils.config import normalize_config, prepare_plugins, validate_config
 from axolotl.utils.dict import DictDefault
 
 from tests.e2e.utils import check_model_output_exists, require_torch_2_4_1
@@ -54,6 +54,7 @@ class LigerIntegrationTestCase:
             }
         )
         # pylint: disable=duplicate-code
+        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
@@ -100,6 +101,7 @@ class LigerIntegrationTestCase:
             }
         )
         # pylint: disable=duplicate-code
+        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()

tests/e2e/patched/test_4d_multipack_llama.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -60,6 +60,7 @@ class Test4dMultipackLlama(unittest.TestCase):
                 "fp16": True,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -104,6 +105,7 @@ class Test4dMultipackLlama(unittest.TestCase):
                 "fp16": True,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_falcon_samplepack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -63,6 +63,7 @@ class TestFalconPatched(unittest.TestCase):
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -103,6 +104,7 @@ class TestFalconPatched(unittest.TestCase):
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_fused_llama.py

@@ -12,7 +12,7 @@ from transformers.utils import is_torch_bf16_gpu_available
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -67,6 +67,7 @@ class TestFusedLlama(unittest.TestCase):
             cfg.bf16 = True
         else:
             cfg.fp16 = True
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_llama_s2_attention.py

@@ -11,7 +11,7 @@ import pytest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -65,6 +65,7 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -105,6 +106,7 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_lora_llama_multipack.py

@@ -12,7 +12,7 @@ from transformers.utils import is_auto_gptq_available, is_torch_bf16_gpu_availab
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -70,6 +70,7 @@ class TestLoraLlama(unittest.TestCase):
         else:
             cfg.fp16 = True
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -120,6 +121,7 @@ class TestLoraLlama(unittest.TestCase):
                 "lr_scheduler": "cosine",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_mistral_samplepack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -63,6 +63,7 @@ class TestMistral(unittest.TestCase):
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -104,6 +105,7 @@ class TestMistral(unittest.TestCase):
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_mixtral_samplepack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -60,6 +60,7 @@ class TestMixtral(unittest.TestCase):
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_model_patches.py

@@ -6,7 +6,7 @@ import unittest
 
 import transformers
 
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_model, load_tokenizer
 
@@ -47,6 +47,7 @@ class TestModelPatches(unittest.TestCase):
                 "eval_steps": 10,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         tokenizer = load_tokenizer(cfg)
         load_model(cfg, tokenizer, inference=False)
@@ -79,6 +80,7 @@ class TestModelPatches(unittest.TestCase):
                 "eval_steps": 10,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         tokenizer = load_tokenizer(cfg)
         load_model(cfg, tokenizer, inference=False)

tests/e2e/patched/test_phi_multipack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -63,6 +63,7 @@ class TestPhiMultipack(unittest.TestCase):
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -82,7 +83,7 @@ class TestPhiMultipack(unittest.TestCase):
                 "sample_packing": True,
                 "flash_attention": True,
                 "pad_to_sequence_len": True,
-                "load_in_8bit": False,
+                "load_in_4bit": True,
                 "adapter": "qlora",
                 "lora_r": 64,
                 "lora_alpha": 32,
@@ -114,6 +115,7 @@ class TestPhiMultipack(unittest.TestCase):
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_resume.py

@@ -12,7 +12,7 @@ from transformers.utils import is_torch_bf16_gpu_available
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, most_recent_subdir
@@ -68,6 +68,7 @@ class TestResumeLlama:
             cfg.bf16 = True
         else:
             cfg.fp16 = True
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_unsloth_qlora.py

@@ -10,7 +10,7 @@ import pytest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, check_tensorboard
@@ -72,6 +72,7 @@ class TestUnslothQLoRA:
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -122,6 +123,7 @@ class TestUnslothQLoRA:
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -177,6 +179,7 @@ class TestUnslothQLoRA:
             }
         )
 
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_embeddings_lr.py

@@ -102,6 +102,7 @@ class TestEmbeddingsLrScale(unittest.TestCase):
                 "use_tensorboard": True,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_llama_vision.py

@@ -109,6 +109,7 @@ class TestLlamaVision(unittest.TestCase):
                 "bf16": True,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_phi.py

@@ -79,7 +79,7 @@ class TestPhi(unittest.TestCase):
                 "tokenizer_type": "AutoTokenizer",
                 "sequence_len": 2048,
                 "sample_packing": False,
-                "load_in_8bit": False,
+                "load_in_4bit": True,
                 "adapter": "qlora",
                 "lora_r": 64,
                 "lora_alpha": 32,
@@ -111,6 +111,7 @@ class TestPhi(unittest.TestCase):
                 "bf16": "auto",
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_process_reward_model_smollm2.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 
 from .utils import check_model_output_exists, check_tensorboard, with_temp_dir
@@ -57,6 +57,7 @@ class TestProcessRewardSmolLM2(unittest.TestCase):
                 "seed": 42,
             }
         )
+        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/test_exact_deduplication.py

@@ -11,7 +11,7 @@ from unittest.mock import patch
 import pytest
 from datasets import Dataset
 
-from axolotl.utils.config import normalize_config
+from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.data import prepare_dataset
 from axolotl.utils.data.rl import load_prepare_preference_datasets
 from axolotl.utils.data.utils import deduplicate_and_log_datasets
@@ -319,6 +319,7 @@ class TestDeduplicateNonRL(unittest.TestCase):
                 "num_epochs": 1,
             }
         )
+        self.cfg_1 = validate_config(self.cfg_1)
         normalize_config(self.cfg_1)
 
     @pytest.mark.skip(reason="TODO: fix hf hub offline to work with HF rate limits")