R4C3R/qwen2.5-coder-0.5b-heretic:q5_k

R4C3R/ qwen2.5-coder-0.5b-heretic:q5_k_m

3 Downloads Updated 17 hours ago

A decensored code-oriented 0.5B model with suppressed refusal behavior, ideal for local coding agents.

tools

ollama run R4C3R/qwen2.5-coder-0.5b-heretic:q5_k_m

curl http://localhost:11434/api/chat \
  -d '{
    "model": "R4C3R/qwen2.5-coder-0.5b-heretic:q5_k_m",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='R4C3R/qwen2.5-coder-0.5b-heretic:q5_k_m',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'R4C3R/qwen2.5-coder-0.5b-heretic:q5_k_m',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 17 hours ago

17 hours ago

97601fd11432 · 420MB ·

model

archqwen2

parameters494M

quantizationQ5_K_M

420MB

Readme

Qwen2.5-Coder-0.5B-Instruct-heretic is a decensored variant of Qwen2.5-Coder-0.5B-Instruct produced with Heretic v1.4.0 (directional ablation / “abliteration”). Refusal behavior is suppressed via targeted weight edits rather than fine-tuning, so the base model’s knowledge and instruction-following are left largely intact.

Sizes

0.5B (default)

Optimized for code generation and understanding. Runs comfortably on CPU.

ollama run R4C3R/qwen2.5-coder-0.5b-heretic

Quantizations

Tag	Size
`:q4_k_m`	398MB
`:q5_k_m`	420MB
`:q8_0`	531MB

Performance

KL divergence: 0.1249 (low — the edit is narrow and targeted)
Refusals: ⁸⁄₁₀₀ vs ⁵²⁄₁₀₀ on the base model

References

Source model | Base model | Heretic