Attempt at using the ministral-3:8b reasoning version for coding tasks.

vision tools thinking 3b 8b

ollama run pstdio/microcoder:8b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "pstdio/microcoder:8b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='pstdio/microcoder:8b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'pstdio/microcoder:8b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 2 days ago

2 days ago

74bf907437c2 · 18GB ·

model

archmistral3

parameters8.49B

quantizationBF16

17GB

projector

archclip

parameters428M

quantizationBF16

867MB

license

714B

template

{{- $hasSystemPrompt := false }} {{- range $index, $_ := .Messages }} {{- if eq .Role "system" }}{{

962B

system

You are Microcoder, a modified version of ministral-8b a Large Language Model (LLM) created by Mistr

3.0kB

params

{ "num_ctx": 16384, "temperature": 0.5 }

36B

Readme

Microcoder

Attempt at using the ministral-3:8b model for coding tasks.

Defaulting to the BF16 version due to reported issues with quantization.

Updated token input length (will use more Memory). By default ollama caps it at 4k which is too small for tool use / coding agent setups.

ollama run pstdio/microcoder:8b
/set parameter num_ctx 16384

To add the model to opencode, run:

ollama launch opencode --config