qwen3.8:27b-512k-bf16

vision tools thinking

ollama run home-rag/qwen3.8:27b-512k-bf16

curl http://localhost:11434/api/chat \
  -d '{
    "model": "home-rag/qwen3.8:27b-512k-bf16",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='home-rag/qwen3.8:27b-512k-bf16',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'home-rag/qwen3.8:27b-512k-bf16',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 3 days ago

3 days ago

6c0d5e209b6b · 56GB ·

model

archqwen35

parameters27.3B

quantizationBF16

55GB

projector

archclip

parameters461M

quantizationBF16

931MB

template

13B

params

{ "draft_num_predict": 4, "min_p": 0, "num_ctx": 524288, "presence_penalty": 0,

134B

Readme

Qwen 3.8 27B

ARCH: QWEN_3.5 // PARAMS: 27B // PRECISION: BF16 // CONTEXT: 512K TOKENS

Hardware & Environment

Designed for local multi-GPU systems and small GPU grids.

Target hardware: AMD Instinct MI50, MI60, MI100 and MI210 series
Recommended memory: 96+ GB HBM / VRAM

Technical Specifications

Model: Qwen 3.8 27B
Origin: Hugging Face Qwen/Qwen3.8-27B
Precision: BF16
Vision: Integrated mmproj (archclip) for multimodal input
Acceleration: Multi-Token Prediction (MTP) draft head
Context: Up to 524288 tokens

Quick Start

ollama run --think=high home-rag/qwen3.8:27b-512k-bf16