LLM EXPLORER 61,405 MODELS INDEXED

DeepSeek V4 Flash W4A16 FP8 MTP by canada-quant

By canada-quant · 3435 downloads
HF Repo ↗ Copy URL

DeepSeek V4 Flash W4A16 FP8 MTP is an open-source language model by canada-quant. Features: 50.7b LLM, VRAM: 170.4GB, Context: 1024K, License: mit, LLM Explorer Score: 0.29.

Base model:deepseek-ai/deepsee... Base model:quantized:deepseek-...   Compressed-tensors   Deepseek   Deepseek v4   En   Fp8   Gptq   Mixture-of-experts   Moe   Mtp   Region:us   Safetensors   Sharded   Speculative-decoding   Tensorflow   Vllm   W4a16   Zh
Model Card on HF 🤗: https://huggingface.co/canada-quant/DeepSeek-V4-Flash-W4A16-FP8-MTP 

DeepSeek V4 Flash W4A16 FP8 MTP Parameters and Internals

LLM NameDeepSeek V4 Flash W4A16 FP8 MTP
Repository 🤗https://huggingface.co/canada-quant/DeepSeek-V4-Flash-W4A16-FP8-MTP 
Base Model(s)  DeepSeek V4 Flash   deepseek-ai/DeepSeek-V4-Flash
Model Size50.7b
Required VRAM170.4 GB
Updated2026-08-01
Maintainercanada-quant
Model Typedeepseek_v4
Model Files  51.2 GB: 1-of-4   50.2 GB: 2-of-4   50.2 GB: 3-of-4   18.8 GB: 4-of-4
Supported Languagesen zh
Model ArchitectureDeepseekV4ForCausalLM
Licensemit
Context Length1048576
Model Max Length1048576
Transformers Version5.8.1
Tokenizer ClassTokenizersBackend
Padding Token<｜end▁of▁sentence｜>
Vocabulary Size129280
Torch Data Typebfloat16