Instructions to use saidutta69/LFM2.5-350M-heretic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use saidutta69/LFM2.5-350M-heretic with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="saidutta69/LFM2.5-350M-heretic")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("saidutta69/LFM2.5-350M-heretic")
model = AutoModelForCausalLM.from_pretrained("saidutta69/LFM2.5-350M-heretic", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use saidutta69/LFM2.5-350M-heretic with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Use Docker

docker model run hf.co/saidutta69/LFM2.5-350M-heretic:Q4_K_M

LM Studio
Jan

vLLM

How to use saidutta69/LFM2.5-350M-heretic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "saidutta69/LFM2.5-350M-heretic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "saidutta69/LFM2.5-350M-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/saidutta69/LFM2.5-350M-heretic:Q4_K_M

SGLang

How to use saidutta69/LFM2.5-350M-heretic with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "saidutta69/LFM2.5-350M-heretic" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "saidutta69/LFM2.5-350M-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "saidutta69/LFM2.5-350M-heretic" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "saidutta69/LFM2.5-350M-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use saidutta69/LFM2.5-350M-heretic with Ollama:
```
ollama run hf.co/saidutta69/LFM2.5-350M-heretic:Q4_K_M
```
Unsloth Desktop

How to use saidutta69/LFM2.5-350M-heretic with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "saidutta69/LFM2.5-350M-heretic:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use saidutta69/LFM2.5-350M-heretic with Docker Model Runner:
```
docker model run hf.co/saidutta69/LFM2.5-350M-heretic:Q4_K_M
```

Lemonade

How to use saidutta69/LFM2.5-350M-heretic with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull saidutta69/LFM2.5-350M-heretic:Q4_K_M

Run and chat with the model

lemonade run user.LFM2.5-350M-heretic-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use saidutta69/LFM2.5-350M-heretic with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default saidutta69/LFM2.5-350M-heretic:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use saidutta69/LFM2.5-350M-heretic with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf saidutta69/LFM2.5-350M-heretic:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "saidutta69/LFM2.5-350M-heretic:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

LFM2.5-350M-heretic

A decensored variant of LiquidAI/LFM2.5-350M, produced with Heretic v1.4.0 (directional ablation / "abliteration"). Liquid's hybrid conv+attention architecture keeps its native Pythonic tool-call format and multilingual coverage; refusal behaviour is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the base model's agentic ability is left largely intact.

Who this is for: developers who want an uncensored assistant that is genuinely fast rather than merely small. LFM2.5-350M decodes at 313 tok/s on an AMD CPU and 188 tok/s on a Snapdragon Gen4 NPU — an order of magnitude faster than a same-size dense Llama — while speaking nine languages. It runs under 1 GB of memory, so it fits in a phone, a Pi, or a browser tab. Reach for it for data extraction, structured output, tool routing, and RAG on hardware where a larger model simply would not run.

This is the most aggressively uncensored model in the collection relative to its size: refusals drop from 89/100 to 4/100, and the 1.2B sibling reaches 2/100.

Runs anywhere

At 350M there is no VRAM ladder — pick a quant by how much RAM and how much bandwidth your device has:

Device	Recommended quant	Weights
Phone / browser (4 GB RAM)	Q4_K_M	~0.21 GB
Raspberry Pi 5 (4 GB)	Q4_K_M	~0.21 GB
Laptop (8 GB, Apple Silicon)	Q8_0	~0.35 GB
Server / multi-tenant	Q6_K	~0.27 GB

Weights only, at this model's 350M native size. Liquid's own measurements put the whole model — weights and KV cache — under 1 GB at Q4_0, with roughly 32K of usable context. Because the hybrid architecture spends most of its layers in cheap convolution blocks, this model is far more decode-bandwidth-sensitive than its parameter count suggests: fast RAM matters more than a fast GPU.

Abliteration parameters

Trial 176 of a 200-trial Heretic run (seed 3260526957). direction_index was selected per layer.

Parameter	Value
direction_index	per layer
attn.o_proj.max_weight	1.46
attn.o_proj.max_weight_position	11.87
attn.o_proj.min_weight	1.41
attn.o_proj.min_weight_distance	7.69
mlp.down_proj.max_weight	1.48
mlp.down_proj.max_weight_position	15.00
mlp.down_proj.min_weight	0.65
mlp.down_proj.min_weight_distance	6.02

Performance

Metric	This model	Original model (LiquidAI/LFM2.5-350M)
KL divergence	0.0989	0 (by definition)
Refusals	4/100	89/100

Refusals on the harmful evaluation set drop from 89/100 to 4/100. The KL divergence of 0.0989 is the highest in this batch — this base model was heavily refusal-tuned for such a small model, so Heretic had to move further than usual to reach the same target. Expect slightly more drift in style than the low-KL siblings; the tool-call format and multilingual ability survive intact.

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.

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

Files

Safetensors

File	Size
`model.safetensors`	0.66 GB

BF16, 350M parameters. The reproduce/ directory carries the full Heretic recipe — config.toml, requirements.txt, the Optuna study journal, and SHA-256 sums — so this exact model can be regenerated bit-for-bit. Reproduce it with heretic --reproduce reproduce/reproduce.json.

GGUF quantizations

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

File	Format	Size
`LFM2.5-350M-heretic-F16.gguf`	GGUF F16	0.66 GB
`LFM2.5-350M-heretic-Q2_K.gguf`	GGUF Q2_K	0.15 GB
`LFM2.5-350M-heretic-IQ3_S.gguf`	GGUF IQ3_S	0.17 GB
`LFM2.5-350M-heretic-Q3_K_S.gguf`	GGUF Q3_K_S	0.17 GB
`LFM2.5-350M-heretic-Q3_K_M.gguf`	GGUF Q3_K_M	0.18 GB
`LFM2.5-350M-heretic-Q3_K_L.gguf`	GGUF Q3_K_L	0.19 GB
`LFM2.5-350M-heretic-IQ4_XS.gguf`	GGUF IQ4_XS	0.20 GB
`LFM2.5-350M-heretic-Q4_K_S.gguf`	GGUF Q4_K_S	0.21 GB
`LFM2.5-350M-heretic-Q4_0.gguf`	GGUF Q4_0	0.20 GB
`LFM2.5-350M-heretic-Q4_1.gguf`	GGUF Q4_1	0.22 GB
`LFM2.5-350M-heretic-Q4_K_M.gguf`	GGUF Q4_K_M	0.21 GB
`LFM2.5-350M-heretic-Q5_K_S.gguf`	GGUF Q5_K_S	0.24 GB
`LFM2.5-350M-heretic-Q5_K_M.gguf`	GGUF Q5_K_M	0.24 GB
`LFM2.5-350M-heretic-Q6_K.gguf`	GGUF Q6_K	0.27 GB
`LFM2.5-350M-heretic-Q8_0.gguf`	GGUF Q8_0	0.35 GB

LFM2 dense architecture (lfm2) - loads natively in llama.cpp / LM Studio / Jan. Day-one upstream support, so this also works with MLX, vLLM, and Nexa-optimized NPU builds.

Run llama serve -hf saidutta69/LFM2.5-350M-heretic to pull the default quant.

Quickstart

# llama.cpp
llama serve -hf saidutta69/LFM2.5-350M-heretic

# transformers (requires transformers>=5.0.0)
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "saidutta69/LFM2.5-350M-heretic"
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", dtype="bfloat16")
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [{"role": "user", "content": "Extract the invoice number and total from: 'Invoice INV-4471, total 231.50 USD, due 30 Nov.'"}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(**inputs, do_sample=True, temperature=0.1, top_k=50, repetition_penalty=1.05, max_new_tokens=512)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

Also runnable via Ollama, LM Studio, Jan, vLLM, SGLang, and OpenVINO.

Generation parameters

Liquid's recommended settings - low temperature with a top-k cut:

model.generate(**inputs, temperature=0.1, top_k=50, repetition_penalty=1.05)

Chat template

LFM2.5 uses a ChatML-like format. See the Chat Template documentation for details.

<|startoftext|>system
You are a helpful assistant trained by Liquid AI.
user
What is C. elegans?
assistant

Tool use

LFM2.5 writes Pythonic calls: a Python list between <|tool_call_start|> and <|tool_call_end|>. Pass tool definitions as a JSON list in the system prompt, or hand them to tokenizer.apply_chat_template(..., tools=tools), then feed the result back under a tool role. To get JSON instead, just ask for it in the system prompt.

Model details


Architecture	`Lfm2ForCausalLM` (hybrid conv + attention)
Parameters	350M
Layers / heads	16 layers (10 double-gated conv + 6 GQA), 16 attention heads, 8 KV heads
Hidden / intermediate	1024 / 6656
Convolution cache	L_cache = 3
Position embedding	RoPE, theta = 1,000,000
Context length	128,000 (config `max_position_embeddings`); Liquid's tuned default is 32,768
Vocab	65,536
Precision	bfloat16
Languages	English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, Spanish
Base model	LiquidAI/LFM2.5-350M

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 — don't put this behind an unmoderated public-facing endpoint serving third parties. Liquid explicitly does not recommend LFM2.5-350M for knowledge-intensive tasks or programming; it is built for extraction, structured output, and tool use.

License

Inherits the LFM License 1.0 from the base model.

LiquidAI/LFM2.5-350M — the base model
LFM2.5 technical report — architecture and benchmarks
RACER IS OP — Heretic Models — full collection

Downloads last month: 51

Safetensors

Model size

0.4B params

Tensor type

BF16

Model tree for saidutta69/LFM2.5-350M-heretic

Base model

LiquidAI/LFM2.5-350M-Base

Finetuned

LiquidAI/LFM2.5-350M

Quantized

(83)

this model

Collection including saidutta69/LFM2.5-350M-heretic

RACER IS OP — Heretic Models

Collection

All abliterated/heretic models by RACER IS OP — uncensored variants of open-weight LLMs • 57 items • Updated about 15 hours ago • 6

Paper for saidutta69/LFM2.5-350M-heretic

LFM2 Technical Report

Paper • 2511.23404 • Published Nov 28, 2025 • 76

saidutta69
/

LFM2.5-350M-heretic

LFM2.5-350M-heretic

Runs anywhere

Abliteration parameters

Performance

Why abliteration instead of fine-tuning

Files

Safetensors

GGUF quantizations

Quickstart

Generation parameters

Chat template

Tool use

Model details

Responsible use

License

Related

Model tree for saidutta69/LFM2.5-350M-heretic

Collection including saidutta69/LFM2.5-350M-heretic

RACER IS OP — Heretic Models

Paper for saidutta69/LFM2.5-350M-heretic

LFM2 Technical Report