mannix/omnimerge-v4-mtp:IQ3

mannix/ omnimerge-v4-mtp:IQ3_M

1,086 Downloads Updated 1 week ago

Qwen/Qwen3.6-27B + 3 Qwen3.6 fine-tunes with MLP-passthrough surgery - MTP quants

vision tools thinking

ollama run mannix/omnimerge-v4-mtp:IQ3_M

curl http://localhost:11434/api/chat \
  -d '{
    "model": "mannix/omnimerge-v4-mtp:IQ3_M",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='mannix/omnimerge-v4-mtp:IQ3_M',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'mannix/omnimerge-v4-mtp:IQ3_M',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 week ago

1 week ago

626044afd4b3 · 13GB ·

model

archqwen35

parameters27.3B

quantization(!unknown_file_type 27!)

13GB

params

{ "draft_num_predict": 3, "min_p": 0, "presence_penalty": 0, "repeat_penalty": 1,

114B

Readme

GGUF quantizations of ManniX-ITA/Qwen3.6-27B-Omnimerge-v4 with the MTP (Multi-Token Prediction) head retained for self-speculative decoding on llama.cpp mainline (PR #22673, merged 2026-05-16) and later.

Up to 4x inference speed with 1 session, 2x with 2 in parallel.