lfm2.5-2.6b-fable5-coding-agent-heretic

LFM2.5-2.6B-Fable5-Coding-Agent-heretic

<div align="center"> <img src="https://photu.kashyalabanavli.site/racer-is-op.png" alt="RACER IS OP" width="100%"> </div> <br>

A decensored variant of AyoubChLin/lfm2.5-2.6b-fable5-coding-agent (full-parameter SFT of LiquidAI/LFM2.5-2.6B on saidutta69/fable-5-premium), produced with Heretic v1.4.0 (directional ablation / "abliteration"). Refusal behavior is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the base model's coding-agent capabilities, tool-use patterns, and instruction-following are left largely intact.

Abliteration results: KL divergence 0.014 · Refusals reduced from 96/100 → 7/100.

Who this is for: developers who want a compact 2.6B coding agent with LFM2's hybrid conv+attention architecture — fast inference, tool-call generation, code generation, and multi-turn assistant behavior — without refusal guardrails. Not a capability upgrade over the base model — same model, refusal guardrails removed.

<!-- racer-gpu-matrix -->

Runs on your gaming PC

Full GGUF ladder included — pick the quant that fits your card:

Weights only, at this model's 2.7B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up.

Why abliteration instead of fine-tuning

Fine-tuning a "helpful" persona on top of RLHF'd refusals fights the base model's training and tends to degrade coherence. Abliteration instead finds and edits the specific weight directions responsible for refusal, leaving the rest of the network (and its capabilities) untouched. See the Heretic repo and the original abliteration writeup for the mechanism.

Files

GGUF quantizations

Full quantization set (14 quants + F16) produced with llama.cpp.

LFM2 hybrid conv+attention architecture — loads natively in llama.cpp (arch lfm2).

Run llama serve -hf saidutta69/lfm2.5-2.6b-fable5-coding-agent-heretic to pull the default quant.

Quickstart

llama.cpp

# defaults to the Q4_K_M quant
llama serve -hf saidutta69/lfm2.5-2.6b-fable5-coding-agent-heretic:Q4_K_M

Ollama

ollama run hf.co/saidutta69/lfm2.5-2.6b-fable5-coding-agent-heretic:Q4_K_M

LM Studio

1. Open LM Studio and click the search icon to open the Model Search panel.
2. Type "lfm2.5-2.6b-fable5-coding-agent-heretic" and click the download button marked GGUF.
3. Pick your quant, load the model, and start chatting.

Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

MODEL_ID = "saidutta69/lfm2.5-2.6b-fable5-coding-agent-heretic"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are a helpful coding assistant."},
    {"role": "user", "content": "Write a Python function that merges overlapping intervals."},
]

inputs = tokenizer.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

with torch.inference_mode():
    output = model.generate(**inputs, max_new_tokens=512, temperature=0.1)

print(tokenizer.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Responsible use

Refusal suppression is deliberate and works as intended: this model will comply with requests the base model would refuse, including some it shouldn't. There is no safety filtering layered on top. You are responsible for how you deploy it.

Made with ❤️ by RACER IS OP — follow for more uncensored models

License

Inherits the LFM Open License v1.0 from the base model.

Abliteration hardware

Recorded in reproduce/reproduce.json:

• GPU: Tesla T4 (14.56 GB VRAM)
• CPU: Intel(R) Xeon(R) CPU @ 2.00GHz
• OS: Linux-6.6.122+-x86_64-with-glibc2.35
• CUDA / driver: CUDA 12.8, driver 580.82.07
• Software: PyTorch 2.11.0+cu128, Heretic 1.4.0

Heretic notes that KL-divergence and refusal numbers can vary slightly with hardware and library versions.