PeterYang12/DeepSeek-V4-Flash-bf16-vllm

AI model by PeterYang12

From the publisher

Model card & documentation

Source preview

Read the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.

A BF16 copy of sgl-project/DeepSeek-V4-Flash-FP8 (revision ae01d80c06cdfe30581edfd0e1c5449dc7ed7f17). Every FP8 weight was dequantized with its block scale and stored as BF16. There is no quantizationconfig, so an inference engine runs the model without weight or activation quantization. This is not a higher-precision model. Each BF16 weight equals fp8value × blockscale exactly, so the weights carry the same information as the FP8 checkpoint. No fine-tuning and no retraining were done. An engine that loads the FP8 checkpoint typically also quantizes the activations to FP8 (W8A8). A training framework that scores the same tokens in BF16 then disagrees with the engine on log-probabilities. With this checkpoint the engine computes in BF16, which removes that source of disagreement. We used it as the rollout model in reinforcement-learning runs. Tokenizer, chat template and generationconfig.json are unchanged. config.json keeps "expertdtype": "fp8" on purpose. vLLM's DeepSeek-V4 loader treats a missing value as fp4 and would select the…

PeterYang12/DeepSeek-V4-Flash-bf16-vllm on Hugging Face A short preview, not the full document.

Read the full model card ↗ · Preview checked 2026-09-29T08:12:08.527Z

Inside the original model card — Document outline

Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.

Links from the model card

References supplied by the publisher, not independently verified endorsements. Check the destination before downloading files or entering credentials.

SGLang Cookbook

Documentation belongs to its respective authors. Reported project/model license: mit. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.

Model overview

PeterYang12/DeepSeek-V4-Flash-bf16-vllm is a task-unspecified model repository published by PeterYang12 on Hugging Face. A library was not reported.

This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.

Model facts

Task: Not reported
Library: Not reported
Recent downloads (30 days): 0
Cumulative likes: 0
Hugging Face trending score: 0
Reported safetensors parameters: 290,944,616,402
Architecture: DeepseekV4ForCausalLM
License: mit
Access: Not gated by Hugging Face
Created: 2026-09-29T06:05:17.000Z
Last modified: 2026-09-29T06:07:58.000Z

Compatibility and lineage

Language coverage was not reported.

Reported base models: deepseek-ai/DeepSeek-V4-Flash

Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.

Use and evaluate

Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.

No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.

Model card and usage instructions

Model files and configuration

Community discussion

Source and freshness

Source: Hugging Face Hub. Metadata observed 2026-09-29T06:38:10.429Z. Daily imports are snapshots, not real-time monitoring.

Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.

Open original source