JERRYNPC/deepseek-v4.1-flash-int3-v100
text-generation model by JERRYNPC
Older snapshot: check the original source for current information.
From the publisher
Model card & documentation
Source previewRead the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.
非官方量化权重。 把 762B 的 DeepSeek-V4.1-Flash-1017 压到 32GB 显存 + 64GB 内存的单卡上运行： int8 稠密权重、int3（3.125 bit）路由专家、2-bit engram，配合「显存常驻 dense + 宿主内存 LRU + NVMe 冷读」的专家流式加载。目标是让路由专家不再逐 token 读盘，把解码稳定在 1 tok/s 以上。 本 repo 只含 dense + 路由专家。engram 表约 55GB，单独放在 JERRYNPC/deepseek-v4.1-flash-engram-v100， full 和 nano 共用。 无 MTP 合计约 276GB（dense 8.47 + experts 212.34 + engram 55.3）； 含可选 MTP 约 282GB（+ 5.95GB）。 = 3.125 bit/权重；单专家 w1/w3 [2304,5120]、w2 [5120,2304]， 定长记录 13,824,000 B。40 层 × 384 专家。 fp32/fp16；加载时反量化为 fp16，约占 14.92GiB 显存（无 MTP）。 行数 384,006,168（L1）/ 384,016,682（L14）；只按 token 查 24 行，不整表驻留。 实现是纯 PyTorch 算子 + 可选 NVRTC int3 解包 kernel + 可选 CPU AVX2 GEMV 动态库， 不依赖任何专有 tilelang kernel。…
Read the full model card ↗ · Preview checked 2026-10-31T12:39:32.613Z
Inside the original model card — Document outline
- DeepSeek-V4.1-Flash-1017 · int3 单卡流式专家版
- 仓库组成
- 量化格式
- 系统与依赖要求
- 快速开始
- 1. 代码与环境
- 2. 下载权重（两个 repo，engram 下到 engram/ 子目录）
- 3. 路径便携化（只需一次）
- 4. 冒烟
- 5. OpenAI 兼容服务
- 实测性能
- 发布前能力抽检（2026-10-22）
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Documentation belongs to its respective authors. Reported project/model license: other. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
Model overview
JERRYNPC/deepseek-v4.1-flash-int3-v100 is a text-generation model repository published by JERRYNPC on Hugging Face. The source reports the pytorch library.
This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.
Model facts
- Task
- text-generation
- Library
- pytorch
- Recent downloads (30 days)
- 0
- Cumulative likes
- 0
- Hugging Face trending score
- 0
- Reported safetensors parameters
- Not reported
- Architecture
- Not reported
- License
- other
- Access
- Not gated by Hugging Face
- Created
- 2026-10-22T06:07:43.000Z
- Last modified
- 2026-10-22T06:11:39.000Z
Compatibility and lineage
Reported languages: zh, en
Reported base models: deepseek-ai/DeepSeek-V4.1-Flash-1017
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.
Source and freshness
Source: Hugging Face Hub. Metadata observed 2026-10-22T06:45:03.943Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.