"# SDK reference\n\nTypes and functions in the `modal-dojo` Python SDK.\n\n## Models\n\n| Name | Description |\n|------|-------------|\n| [`DeepSeek_V4_1_Flash`](/reference/deepseek_v4_1_flash) | DeepSeek-V4.1-Flash sparse-attention MoE model, 40 layers and 384 routed experts. |\n| [`Gemma4_26B_A4B`](/reference/gemma4_26b_a4b) | Google Gemma-4-26B-A4B-it multimodal MoE model with 26B total and 4B active parameters. |\n| [`GLM_4_7`](/reference/glm_4_7) | Zhipu AI GLM-4.7 MoE model with 355B total and 32B active parameters. |\n| [`HFModelConfiguration`](/reference/hfmodelconfiguration) | Downloads Hugging Face model weights with `snapshot_download`. |\n| [`Inkling_Small`](/reference/inkling_small) | Thinking Machines Lab Inkling-Small MoE model with 276B total and 12B active parameters. |\n| [`Inkling_Small_LoRA`](/reference/inkling_small_lora) | Selects `Inkling_Small_LoRA_Recipe`. |\n| [`Kimi_K3`](/reference/kimi_k3) | Moonshot Kimi-K3 hybrid KDA/MLA MoE model, 93 layers and 896 routed experts. |\n| [`ModelArchitecture`](/reference/modelarchitecture) | Megatron transformer architecture parameters. |\n| [`ModelConfig`](/reference/modelconfig) | Defines model identity, weight download, and response parsing. |\n| [`Moonlight_16B_A3B_Instruct`](/reference/moonlight_16b_a3b_instruct) | Moonshot AI Moonlight model with 16B total and 3B active parameters. |\n| [`ParsedResponse`](/reference/parsedresponse) | Structured result of parsing raw model output. |\n| [`Qwen3_0_6B`](/reference/qwen3_0_6b) | Alibaba Qwen3-0.6B model. |\n| [`Qwen3_1_7B`](/reference/qwen3_1_7b) | Alibaba Qwen3-1.7B model. |\n| [`Qwen3_30B`](/reference/qwen3_30b) | Alibaba Qwen3-30B-A3B MoE model with 30B total and 3B active parameters. |\n| [`Qwen3_4B`](/reference/qwen3_4b) | Alibaba Qwen3-4B model. |\n| [`Qwen3_5_0_8B`](/reference/qwen3_5_0_8b) | Alibaba Qwen3.5-0.8B model. |\n| [`Qwen3_5_2B`](/reference/qwen3_5_2b) | Alibaba Qwen3.5-2B model. |\n| [`Qwen3_5_4B`](/reference/qwen3_5_4b) | Alibaba Qwen3.5-4B model. |\n| [`Qwen3_5_9B`](/reference/qwen3_5_9b) | Alibaba Qwen3.5-9B model. |\n| [`Qwen3_6_27B`](/reference/qwen3_6_27b) | Qwen3.6-27B dense hybrid Gated DeltaNet/attention model. |\n| [`Qwen3_6_35B`](/reference/qwen3_6_35b) | Alibaba Qwen3.6-35B-A3B model. |\n| [`Qwen3_8_27B`](/reference/qwen3_8_27b) | Alibaba Qwen3.8-27B model. |\n| [`Qwen3_8B`](/reference/qwen3_8b) | Alibaba Qwen3-8B model. |\n| [`Qwen3_ASR_1_7B`](/reference/qwen3_asr_1_7b) | Alibaba Qwen3-ASR-1.7B speech recognition model. |\n| [`Qwen3_VL_8B`](/reference/qwen3_vl_8b) | Alibaba Qwen3-VL-8B-Instruct model. |\n| [`ToolCall`](/reference/toolcall) | Tool invocation parsed from model output. |\n\n## Datasets\n\n| Name | Description |\n|------|-------------|\n| [`DatasetConfig`](/reference/datasetconfig) | Dataset fields and materialization behavior shared across training frameworks. |\n| [`HarborDataset`](/reference/harbordataset) | A dataset loaded from Harbor tasks. |\n| [`HuggingFaceDataset`](/reference/huggingfacedataset) | A dataset loaded from a Hugging Face `datasets` repository. |\n| [`MultimodalDataset`](/reference/multimodaldataset) | Dataset of text prompts paired with image, audio, or video data. |\n| [`OnlineRollout`](/reference/onlinerollout) | Placeholder rows that size a live generate batch. |\n\n## Recipes\n\n| Name | Description |\n|------|-------------|\n| [`DeepSeek_V4_1_Flash_Recipe`](/reference/deepseek_v4_1_flash_recipe) | DeepSeek-V4.1-Flash GRPO recipe for 8 nodes with 8 H200 GPUs each. |\n| [`Gemma4_26B_A4B_Recipe`](/reference/gemma4_26b_a4b_recipe) | Gemma-4-26B-A4B recipe. |\n| [`GLM_4_7_Recipe`](/reference/glm_4_7_recipe) | GLM-4.7 recipe. |\n| [`Inkling_Small_LoRA_Recipe`](/reference/inkling_small_lora_recipe) | Inkling-Small rank-32 LoRA recipe. |\n| [`Inkling_Small_Recipe`](/reference/inkling_small_recipe) | Inkling-Small full-parameter recipe. |\n| [`Kimi_K3_LoRA_Recipe`](/reference/kimi_k3_lora_recipe) | Kimi-K3 rank-32 LoRA recipe for 8 nodes with 8 B300 GPUs each. |\n| [`MilesRecipe`](/reference/milesrecipe) | Miles training and Modal resource settings. |\n| [`Moonlight_16B_A3B_Recipe`](/reference/moonlight_16b_a3b_recipe) | Moonlight-16B-A3B recipe. |\n| [`Qwen3_0_6B_Recipe`](/reference/qwen3_0_6b_recipe) | Qwen3-0.6B recipe. |\n| [`Qwen3_1_7B_Recipe`](/reference/qwen3_1_7b_recipe) | Qwen3-1.7B recipe. |\n| [`Qwen3_4B_Recipe`](/reference/qwen3_4b_recipe) | Qwen3-4B recipe. |\n| [`Qwen3_5_0_8B_Recipe`](/reference/qwen3_5_0_8b_recipe) | Qwen3.5-0.8B recipe. |\n| [`Qwen3_5_2B_Recipe`](/reference/qwen3_5_2b_recipe) | Qwen3.5-2B recipe. |\n| [`Qwen3_5_4B_Miles_Recipe`](/reference/qwen3_5_4b_miles_recipe) | Qwen3.5-4B recipe. |\n| [`Qwen3_5_4B_Recipe`](/reference/qwen3_5_4b_recipe) | Qwen3.5-4B recipe. |\n| [`Qwen3_5_9B_Recipe`](/reference/qwen3_5_9b_recipe) | Qwen3.5-9B recipe. |\n| [`Qwen3_6_27B_Recipe`](/reference/qwen3_6_27b_recipe) | Qwen3.6-27B recipe. |\n| [`Qwen3_6_35B_Recipe`](/reference/qwen3_6_35b_recipe) | Qwen3.6-35B-A3B recipe. |\n| [`Qwen3_8_27B_Recipe`](/reference/qwen3_8_27b_recipe) | Qwen3.8-27B recipe. |\n| [`Qwen3_8B_Recipe`](/reference/qwen3_8b_recipe) | Qwen3-8B recipe. |\n| [`Qwen3_ASR_1_7B_Recipe`](/reference/qwen3_asr_1_7b_recipe) | Qwen3-ASR-1.7B recipe. |\n| [`Qwen3_VL_8B_Recipe`](/reference/qwen3_vl_8b_recipe) | Qwen3-VL-8B recipe. |\n| [`SlimeRecipe`](/reference/slimerecipe) | Slime training and Modal resource settings. |\n\n## Training\n\n| Name | Description |\n|------|-------------|\n| [`Checkpoint`](/reference/checkpoint) | A complete training checkpoint discovered on a Modal Volume. |\n| [`CheckpointType`](/reference/checkpointtype) | Whether a checkpoint is Hugging Face or Megatron weights. |\n| [`DashboardComponent`](/reference/dashboardcomponent) | Supported run-scoped dashboard component slots. |\n| [`DashboardMetricConfig`](/reference/dashboardmetricconfig) | Log framework metrics to the Modal Dojo dashboard only. |\n| [`DojoConfigError`](/reference/dojoconfigerror) | Raised when a training or deploy config is invalid. |\n| [`DojoError`](/reference/dojoerror) | Base error for Modal Dojo. |\n| [`GpuAllocationError`](/reference/gpuallocationerror) | Raised when a recipe's cluster or parallelism settings are invalid. |\n| [`MetricConfig`](/reference/metricconfig) | Defines metric tracker metadata, environment variables, and links. |\n| [`ModalCaptureError`](/reference/modalcaptureerror) | Raised when a cloudpickled user callback captures live Modal state. |\n| [`Sample`](/reference/sample) | A prompt, response, parsed structure, score, and metadata from one model call. |\n| [`TrackioConfig`](/reference/trackioconfig) | Trackio logging configuration shared across all frameworks. |\n| [`TrainConfig`](/reference/trainconfig) | A dataset, model, and recipe for one training run. |\n| [`TrainingGroup`](/reference/traininggroup) | A parameter sweep over a base `TrainConfig`. |\n| [`TrainingRun`](/reference/trainingrun) | A launched training run that can be inspected, awaited, or loaded by ID. |\n| [`WandbConfig`](/reference/wandbconfig) | Weights & Biases run metadata and credentials. |\n| [`convert_megatron_checkpoint_to_hf`](/reference/convert_megatron_checkpoint_to_hf) | Convert a Megatron checkpoint to Hugging Face format. |\n| [`extract_code`](/reference/extract_code) | Extract Python code from an LLM response. |\n| [`score_in_sandbox`](/reference/score_in_sandbox) | Run code against test cases in a Modal sandbox. |\n\n## Deployment\n\n| Name | Description |\n|------|-------------|\n| [`CustomDeployment`](/reference/customdeployment) | A model deployed with an SGLang or vLLM recipe. |\n| [`Endpoint`](/reference/endpoint) | Controls a [Modal Endpoint](https://modal.com/docs/guide/endpoints) that persists until stopped. |\n| [`Sandbox`](/reference/sandbox) | Controls a [Modal Sandbox](https://modal.com/docs/guide/sandbox) in the `app_name` app. |\n| [`SandboxResult`](/reference/sandboxresult) | Outcome of a single `Sandbox.run` invocation. |\n| [`SglangRecipe`](/reference/sglangrecipe) | SGLang server settings. |\n| [`VllmRecipe`](/reference/vllmrecipe) | vLLM server settings. |\n"
