> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo/automodel/_mcp/server.

# nemo_automodel.components.speculative

Speculative decoding and training components.

## Subpackages

* **[`nemo_automodel.components.speculative.dflash`](/nemo-automodel/nemo_automodel/components/speculative/dflash)**
* **[`nemo_automodel.components.speculative.dspark`](/nemo-automodel/nemo_automodel/components/speculative/dspark)**
* **[`nemo_automodel.components.speculative.eagle`](/nemo-automodel/nemo_automodel/components/speculative/eagle)**
* **[`nemo_automodel.components.speculative.streaming`](/nemo-automodel/nemo_automodel/components/speculative/streaming)**

## Submodules

* **[`nemo_automodel.components.speculative.bench_common`](/nemo-automodel/nemo_automodel/components/speculative/bench_common)**
* **[`nemo_automodel.components.speculative.bench_sglang`](/nemo-automodel/nemo_automodel/components/speculative/bench_sglang)**
* **[`nemo_automodel.components.speculative.bench_sweep`](/nemo-automodel/nemo_automodel/components/speculative/bench_sweep)**
* **[`nemo_automodel.components.speculative.bench_vllm`](/nemo-automodel/nemo_automodel/components/speculative/bench_vllm)**
* **[`nemo_automodel.components.speculative.decode_eval`](/nemo-automodel/nemo_automodel/components/speculative/decode_eval)**
* **[`nemo_automodel.components.speculative.precompute_dspark`](/nemo-automodel/nemo_automodel/components/speculative/precompute_dspark)**
* **[`nemo_automodel.components.speculative.precompute_eagle3`](/nemo-automodel/nemo_automodel/components/speculative/precompute_eagle3)**
* **[`nemo_automodel.components.speculative.regen_loop`](/nemo-automodel/nemo_automodel/components/speculative/regen_loop)**
* **[`nemo_automodel.components.speculative.regenerate`](/nemo-automodel/nemo_automodel/components/speculative/regenerate)**
* **[`nemo_automodel.components.speculative.regenerate_vlm`](/nemo-automodel/nemo_automodel/components/speculative/regenerate_vlm)**
* **[`nemo_automodel.components.speculative.serve_sglang`](/nemo-automodel/nemo_automodel/components/speculative/serve_sglang)**
* **[`nemo_automodel.components.speculative.serve_target`](/nemo-automodel/nemo_automodel/components/speculative/serve_target)**
* **[`nemo_automodel.components.speculative.serve_vllm`](/nemo-automodel/nemo_automodel/components/speculative/serve_vllm)**
* **[`nemo_automodel.components.speculative.target_cp`](/nemo-automodel/nemo_automodel/components/speculative/target_cp)**

## Docs

- [nemo_automodel.components.speculative.bench_common](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/bench_common.md)
- [nemo_automodel.components.speculative.bench_sglang](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/bench_sglang.md)
- [nemo_automodel.components.speculative.bench_sweep](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/bench_sweep.md)
- [nemo_automodel.components.speculative.bench_vllm](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/bench_vllm.md)
- [nemo_automodel.components.speculative.decode_eval](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/decode_eval.md)
- [nemo_automodel.components.speculative.dflash](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash.md)
- [nemo_automodel.components.speculative.dflash.core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/core.md)
- [nemo_automodel.components.speculative.dflash.dflash2_core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/dflash2_core.md)
- [nemo_automodel.components.speculative.dflash.domino_core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/domino_core.md)
- [nemo_automodel.components.speculative.dflash.draft_kimi_k3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/draft_kimi_k3.md)
- [nemo_automodel.components.speculative.dflash.draft_qwen3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/draft_qwen3.md)
- [nemo_automodel.components.speculative.dflash.draft_qwen3_dflash2](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/draft_qwen3_dflash2.md)
- [nemo_automodel.components.speculative.dflash.jetspec_core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/jetspec_core.md)
- [nemo_automodel.components.speculative.dflash.registry](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/registry.md)
- [nemo_automodel.components.speculative.dflash.target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dflash/target.md)
- [nemo_automodel.components.speculative.dspark](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark.md)
- [nemo_automodel.components.speculative.dspark.common](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/common.md)
- [nemo_automodel.components.speculative.dspark.config](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/config.md)
- [nemo_automodel.components.speculative.dspark.core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/core.md)
- [nemo_automodel.components.speculative.dspark.draft_deepseek_v4](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/draft_deepseek_v4.md)
- [nemo_automodel.components.speculative.dspark.draft_gemma4](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/draft_gemma4.md)
- [nemo_automodel.components.speculative.dspark.draft_glm_5_2](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/draft_glm_5_2.md)
- [nemo_automodel.components.speculative.dspark.draft_kimi_k3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/draft_kimi_k3.md)
- [nemo_automodel.components.speculative.dspark.draft_minimax_m3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/draft_minimax_m3.md)
- [nemo_automodel.components.speculative.dspark.draft_qwen3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/draft_qwen3.md)
- [nemo_automodel.components.speculative.dspark.loss](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/loss.md)
- [nemo_automodel.components.speculative.dspark.markov_head](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/markov_head.md)
- [nemo_automodel.components.speculative.dspark.registry](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/registry.md)
- [nemo_automodel.components.speculative.dspark.target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/target.md)
- [nemo_automodel.components.speculative.dspark.target_utils](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/dspark/target_utils.md)
- [nemo_automodel.components.speculative.eagle](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle.md)
- [nemo_automodel.components.speculative.eagle.backend](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/backend.md)
- [nemo_automodel.components.speculative.eagle.core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/core.md)
- [nemo_automodel.components.speculative.eagle.core_v12](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/core_v12.md)
- [nemo_automodel.components.speculative.eagle.draft_deepseek](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/draft_deepseek.md)
- [nemo_automodel.components.speculative.eagle.draft_gemma](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/draft_gemma.md)
- [nemo_automodel.components.speculative.eagle.draft_gpt_oss](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/draft_gpt_oss.md)
- [nemo_automodel.components.speculative.eagle.draft_kimi_k3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/draft_kimi_k3.md)
- [nemo_automodel.components.speculative.eagle.draft_llama](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/draft_llama.md)
- [nemo_automodel.components.speculative.eagle.draft_llama_v12](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/draft_llama_v12.md)
- [nemo_automodel.components.speculative.eagle.msd](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/msd.md)
- [nemo_automodel.components.speculative.eagle.msd_curriculum](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/msd_curriculum.md)
- [nemo_automodel.components.speculative.eagle.msd_decode](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/msd_decode.md)
- [nemo_automodel.components.speculative.eagle.msd_target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/msd_target.md)
- [nemo_automodel.components.speculative.eagle.peagle_attention](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/peagle_attention.md)
- [nemo_automodel.components.speculative.eagle.peagle_data](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/peagle_data.md)
- [nemo_automodel.components.speculative.eagle.peagle_draft](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/peagle_draft.md)
- [nemo_automodel.components.speculative.eagle.peagle_trainer](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/peagle_trainer.md)
- [nemo_automodel.components.speculative.eagle.registry](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/registry.md)
- [nemo_automodel.components.speculative.eagle.remote](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/remote.md)
- [nemo_automodel.components.speculative.eagle.remote.client](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/remote/client.md)
- [nemo_automodel.components.speculative.eagle.remote.protocol](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/remote/protocol.md)
- [nemo_automodel.components.speculative.eagle.remote.server](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/remote/server.md)
- [nemo_automodel.components.speculative.eagle.remote.transport](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/remote/transport.md)
- [nemo_automodel.components.speculative.eagle.remote.wire](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/remote/wire.md)
- [nemo_automodel.components.speculative.eagle.ring_attention](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/ring_attention.md)
- [nemo_automodel.components.speculative.eagle.sglang_runner](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/sglang_runner.md)
- [nemo_automodel.components.speculative.eagle.sglang_target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/sglang_target.md)
- [nemo_automodel.components.speculative.eagle.target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/target.md)
- [nemo_automodel.components.speculative.eagle.target_runner](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/target_runner.md)
- [nemo_automodel.components.speculative.eagle.target_v12](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/target_v12.md)
- [nemo_automodel.components.speculative.eagle.ulysses_attention](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/ulysses_attention.md)
- [nemo_automodel.components.speculative.eagle.vispec_core](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/vispec_core.md)
- [nemo_automodel.components.speculative.eagle.vispec_draft](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/vispec_draft.md)
- [nemo_automodel.components.speculative.eagle.vispec_target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/vispec_target.md)
- [nemo_automodel.components.speculative.eagle.vllm_runner](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/vllm_runner.md)
- [nemo_automodel.components.speculative.eagle.vllm_target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/vllm_target.md)
- [nemo_automodel.components.speculative.eagle.zigzag_ring_attention](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/eagle/zigzag_ring_attention.md)
- [nemo_automodel.components.speculative.precompute_dspark](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/precompute_dspark.md)
- [nemo_automodel.components.speculative.precompute_eagle3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/precompute_eagle3.md)
- [nemo_automodel.components.speculative.regen_loop](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/regen_loop.md)
- [nemo_automodel.components.speculative.regenerate](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/regenerate.md)
- [nemo_automodel.components.speculative.regenerate_vlm](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/regenerate_vlm.md)
- [nemo_automodel.components.speculative.serve_sglang](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/serve_sglang.md)
- [nemo_automodel.components.speculative.serve_target](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/serve_target.md)
- [nemo_automodel.components.speculative.serve_vllm](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/serve_vllm.md)
- [nemo_automodel.components.speculative.streaming](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming.md)
- [nemo_automodel.components.speculative.streaming.async_pipeline](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/async_pipeline.md)
- [nemo_automodel.components.speculative.streaming.eagle3](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/eagle3.md)
- [nemo_automodel.components.speculative.streaming.loader](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/loader.md)
- [nemo_automodel.components.speculative.streaming.producer](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/producer.md)
- [nemo_automodel.components.speculative.streaming.queue](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/queue.md)
- [nemo_automodel.components.speculative.streaming.refs](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/refs.md)
- [nemo_automodel.components.speculative.streaming.store](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/store.md)
- [nemo_automodel.components.speculative.streaming.stores](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/stores.md)
- [nemo_automodel.components.speculative.streaming.stores.local](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/stores/local.md)
- [nemo_automodel.components.speculative.streaming.stores.shared_dir](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/streaming/stores/shared_dir.md)
- [nemo_automodel.components.speculative.target_cp](https://docs.nvidia.com/nemo/automodel/v0.5/nemo-automodel/nemo_automodel/components/speculative/target_cp.md)