DeepSeek-V4-Flash-0731

Revision 7872f01b1d1fe23eabc4c98b48bffcef5a386062

Open on Hugging Face Plan a deployment

304BTOTAL PARAMETERS

—CHECKPOINT BYTES

— / 6EXPERTS / PER TOKEN

4.6MDOWNLOADS

TENSOR ACCOUNTING

Where the bytes are.

Summed from the safetensors index, one row per dtype. A parameter count alone cannot produce this figure, because a checkpoint may mix widths.

Dtype	Parameters	Bytes each	Bytes	Share
`I8`	296,352,743,424	1	296.4 GB	—
`F8_E4M3`	6,304,038,912	1	6.3 GB	—
`BF16`	1,483,567,488	2	3.0 GB	—
`F32`	37,741,630	4	151 MB	—
`I64`	2,327,040	8	19 MB	—

EVERY FIELD, WITH ITS ORIGIN

Sourced or undetermined. Never assumed.

Each value below names the exact API field it was computed from. Where the Hub does not establish a value, the reason is shown instead of a plausible default.

Architecture: deepseek_v4config.model_type
Model classes: DeepseekV4ForCausalLMconfig.architectures
Routed experts: UndeterminedThe published config exposes no routed-expert count. The Hub's config summary omits fields some architectures place only in the full config.json.
Experts per token: 6config.num_experts_per_tok
Shared experts: UndeterminedThe published config declares no always-on shared experts.
Routing sparsity: UndeterminedRouting sparsity requires both a routed-expert and a per-token expert count.
Total parameters: 304,180,418,494safetensors.total
Checkpoint bytes: UndeterminedPer-dtype tensor bytes (305793500152) exceed the repository's stored bytes (166888735421), so the dtype map cannot be read literally for this checkpoint.
Ships below 16-bit: UndeterminedPrecision cannot be characterised from an inconsistent dtype map.
Quantisation method: fp8config.quantization_config.quant_method
Trained context: UndeterminedThe config summary omits max_position_embeddings. Trained context is a model-card claim, not a derivable fact.
Declared licence: mitcardData.license
Base model: UndeterminedThe repository declares no base model.
Library: transformerslibrary_name
Files in repository: 74siblings
Last modified: 2026-08-01lastModified

LICENCE POSTURE

Permissive

The repository declares a licence that is generally read as permitting commercial use. Read the licence file in the repository before relying on that.

This is a reading of a metadata field, not legal advice, and it does not account for the licences of upstream models or training data.

PROVENANCE

Derived from https://huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Flash-0731 in the snapshot generated 2026-09-05. Evidence class hub_derived: computed mechanically from publisher metadata, never measured by this project.

Back to catalogue How evidence is classified