QuantTrio / text-generation updated 8 months ago

DeepSeek-V3.2-AWQ

Name: DeepSeek-V3.2-AWQ
Author: QuantTrio

Note: 1. Tested on Hopper device, we don't know if ada / ampere devices could run this repo yet. 2. Waiting for official chattemplate.jinja; The file in this repo is borrowed from v3.1 with thinking mode turned off by default. To enable thinking mode, include: extrabody = {"chattemplatekwargs": {"thinking": True}} in the post requests.

Params

685.4 B

Context

163,840

Downloads 30d

168 K

Likes

Commercial use: allowed mit Not gated SAFETENSORS View on Hugging Face ↗

Download history

daily snapshots · 10 days

197 K168 K

Jul 28Jul 31Aug 3Aug 6

Can you run it?

Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.

File	Quant	Size	Est. VRAM	Verdict on RTX 4090 · 24 GB
model.safetensors	i32	361.9 GB	501.4 GB	❌ Won’t fit
model.safetensors (bf16, full)	bf16 + 160K ctx	361.9 GB	2,454.7 GB	❌ Won’t fit

Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.

Run it

copy-paste, exact tags checked against the Hub

~ · curl · api/v1

$ curl -s https://aimodelscomparison.com/api/v1/models/deepseek-v3-2-awq
{
  "hf_id": "QuantTrio/DeepSeek-V3.2-AWQ",
  "params_b": 685.36,
  "context_length": 163840,
  "license": { "id": "mit", "commercial": "yes" },
  "downloads_30d": 167983,
  "vram_estimates": [
    { "quant": "i32", "gb": 501.4 }
  ],
  "updated_at": "2026-07-28T18:08:12Z"
}

est. VRAM —on RTX 4090 · 24 GBJSON API →

Specifications

Architecture: DeepseekV32ForCausalLM
Parameters: 685.4 B
Tensor type: I32
Context length: 163,840
Vocabulary: 129,280
Layers / heads: 61 / 128
Licence: mit
First seen on the Hub: 2025-12-03
Base model: DeepSeek-V3.2
Training datasets: undisclosed
Added to our catalog: 2026-07-28

Family

Base model and the most-downloaded derivatives in the catalog.

Base DeepSeek-V3.2 deepseek-ai 1.2 M

Compare with

DeepSeek-R1 9.9 M GLM-5.2-FP8 2.7 M GLM-5.2 2.2 M DeepSeek-V3.2 1.2 M DeepSeek-V3 1.2 M GLM-5-FP8 1.2 M

Appears in

Best open coding models Commercial-use text models Models that fit in 12 GB