DeepSeek-V3
Support summary
DeepSeek-V3 is a checkpoint published by DeepSeek, built on the DeepseekV3ForCausalLM architecture. The newest indexed releases vLLM 0.28.0 and TensorRT-LLM 1.2.1 both document support. Official recipes or cookbooks cover SGLang, without pinning a stable release.
vLLM documents support in 20 of 20 indexed releases, from 0.17.0 to 0.28.0. TensorRT-LLM documents support in 2 of 2 indexed releases, from 1.2.0 to 1.2.1.
- Documented hardware
- NVIDIA / CUDA and AMD / ROCm
- Documented quantization
- FP8
- Documented features
- Reasoning parsers and Tensor parallelism
- Configured context
- 163,840 tokens
Hardware, quantization and features count only statements about this checkpoint; engine-wide capabilities are listed with each release.
Hugging Face model card
- Publisher
- DeepSeek
- Task
- text generation
- Library
- transformers
- Architecture
- DeepseekV3ForCausalLM
- Weight tensor elements
- 684.5B
- Tensor types
- BF16, F8_E4M3, F32
- Quantization
- fp8
- Configured context
- 163,840 tokens
- Weight access
- Public
- Repository created
- 2024-12-25
Model notes and sources
TensorRT-LLM limits MLA prefix reuse to named GPU architectures and KV-cache dtypes.
- vLLM v0.28.0 · docs/models/supported_models.md · 20 indexed revisions, v0.17.0 to v0.28.0
- TensorRT-LLM v1.2.1 · docs/source/models/supported-models.md · 2 indexed revisions, v1.2.0 to v1.2.1
- TensorRT-LLM v1.2.1 · docs/source/legacy/reference/support-matrix.md · 2 indexed revisions, v1.2.0 to v1.2.1
- deepseek-ai/DeepSeek-V3 publisher metadata
- deepseek-ai/DeepSeek-V3 · publisher model card
- deepseek-ai/DeepSeek-V3 · Hugging Face metadata
- deepseek-ai/DeepSeek-V3 · publisher configuration
- vllm · models/deepseek-ai/DeepSeek-V3.yaml
- sglang · docs/src/snippets/autoregressive/deepseek-v3-deployment.jsx
- sglang · docs/cookbook/autoregressive/DeepSeek/DeepSeek-V3.mdx
- Reviewed
Engine support
3 engines
| Engine | Release | Checkpoint support | Evidence scope | Last known supported |
|---|---|---|---|---|
| vLLM | 0.28.0Newest indexed | Checkpoint evidenceSource: vLLM v0.28.0 · docs/models/supported_models.mdReviewed | 0.28.0This release | |
Features, dependencies and evidence for vLLM 0.28.0 | ||||
| SGLang | Version not pinnedRecipe / cookbook | Documented recipeSource 1: sglang · docs/cookbook/autoregressive/DeepSeek/DeepSeek-V3.mdxSource 2: sglang · docs/src/snippets/autoregressive/deepseek-v3-deployment.jsxReviewed | No supported release indexed | |
Features, dependencies and evidence for SGLang 0.5.19 | ||||
| TensorRT-LLM | 1.2.1Newest indexed | Checkpoint evidenceSource 1: TensorRT-LLM v1.2.1 · docs/source/models/supported-models.mdSource 2: TensorRT-LLM v1.2.1 · docs/source/legacy/reference/support-matrix.mdReviewed | 1.2.1This release | |
Features, dependencies and evidence for TensorRT-LLM 1.2.1 | ||||