DeepSeek-V4-Flash-0731 VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. FP8 safetensors: 156.71 GiB–182.40 GiB at 8,192 tokens, 1 sequence, FP16 KV, vllm.
Known working setups
1 known setupsCommunity and maintainer-tested deployments on real hardware with verified context, quantization, and runtime configurations.
- ✓ generation
- ✓ 1,048,576 context
- ✓ tool calling
- ✓ vision
- ✓ speculative decoding
2-node DGX Spark deployment for DeepSeek-V4-Flash-Vision-Exp running vLLM TP=2 with DSpark speculative decoding and nvfp4_ds_mla KV cache.
Runs on
Publisher and provenance
- PublisherDeepSeek
- Hugging Face
deepseek-ai/DeepSeek-V4-Flash-0731 - Revision
7872f01b1d1fe23eabc4c98b48bffcef5a386062 - Total parameters304,180,418,494
- Active parametersUnknown / not independently sourced
- Layers43
- Query heads64
- KV heads1 (mla)
- Head dimension512
- Recorded context capability1,048,576
- Completenesspartial · partial
Active parameters describe tokens' routed compute, not the full resident model weight footprint. Full-resident weight estimates use total parameters.
Memory layers in this estimate
Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.
- Weight artifact
- 155.42 GiB (166,878,536,440 bytes)
- Raw KV cache
- 0.67 GiB (721,420,288 bytes)
- Engine extras / workspace
- 0.63 GiB–26.31 GiB
- Total estimated demand
- 156.71 GiB–182.40 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| FP8 safetensors fp8 | 166,878,536,440 (155.42 GiB / 166.88 GB) | vllm, sglang, unspecified | file-metadata | |
| INT4 theoretical | No published file | 152,090,209,247 (theoretical) | — | theoretical |
Memory by quantization and context
Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.
| Quantization | 2,048 tokens | 8,192 tokens | 32,768 tokens |
|---|---|---|---|
| FP8 safetensors | 156.21 GiB–181.80 GiB | 156.71 GiB–182.40 GiB | 158.73 GiB–184.79 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 141.94 GiB–183.21 GiB | 142.44 GiB–183.84 GiB | 144.46 GiB–186.32 GiB |
Raw KV contribution
Related pages
Sources
- DeepSeek-V4-Flash-0731 model card
- config.json — revision 7872f01b1d1fe23eabc4c98b48bffcef5a386062
- safetensors index — metadata.total_size=166878536440
- Artifact repo unsloth/DeepSeek-V4-Flash-0731-GGUF — DSpark drafter dspark-DeepSeek-V4-Flash-0731-Q8_0.gguf is not main weight and is not a catalog artifact. No measured main Q8_0 size is recorded from this repo.
Assumptions that affect this result
- Requested context is within the recorded native window of 1048576 tokens.
- Full-resident weight estimates use total parameters, including inactive MoE experts.
- Activated parameter count is unknown / not independently sourced. Active parameters describe routed compute, not the full resident model weight footprint.
- Weight bytes come from the selected weight files only, not from every file in the repository.
- Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
- Weight bytes come from 48 published safetensors shards totaling 166878536440 bytes.
- Multi-token prediction / draft weights exist and are not included unless enabled.
- Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
Calculator
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.