Instructions to use DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL with Transformers:

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

pipe = pipeline("text-generation", model="DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL")
model = AutoModelForCausalLM.from_pretrained("DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL

SGLang

How to use DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL 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 "DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL" \
    --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": "DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL",
		"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 "DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL" \
        --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": "DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Studio new

How to use DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL",
    max_seq_length=2048,
)

Docker Model Runner
How to use DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL with Docker Model Runner:
```
docker model run hf.co/DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL
```

LM-Eval Benchmark Results

by jajmangold - opened Mar 10

Discussion

jajmangold

Mar 10

LM-Eval Benchmark Results

Original: `kaitchup/LFM2.5-1.2B-Thinking-autoround-W4A16` (quantized W4A16)

arc_challenge: acc=0.2099, acc_norm=0.2321
arc_easy: acc=0.3733, acc_norm=0.3641
hellaswag: acc=0.3239, acc_norm=0.3697
piqa: acc=0.5919, acc_norm=0.5658

DavidAU: `DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL` (FP16)

arc_challenge: acc=0.2125, acc_norm=0.2355
arc_easy: acc=0.3872, acc_norm=0.3645
hellaswag: acc=0.3185, acc_norm=0.3605
piqa: acc=0.5789, acc_norm=0.5686

Comparison Table

Task	Original acc	DavidAU acc	Diff	Original acc_norm	DavidAU acc_norm	Diff
arc_challenge	20.99%	21.25%	+0.26%	23.21%	23.55%	+0.34%
arc_easy	37.33%	38.72%	+1.39%	36.41%	36.45%	+0.04%
hellaswag	32.39%	31.85%	-0.54%	36.97%	36.05%	-0.92%
piqa	59.19%	57.89%	-1.30%	56.58%	56.86%	+0.28%

Conclusion: The models perform nearly identically. The DavidAU distill shows slight improvement on ARC tasks but slight regression on HellaSwag and PIQA. All differences are within the standard error margins (~1-1.2%).

DavidAU

Owner Mar 10

•

edited Mar 10

Thank you for reminding me to update this card.
NOTE:

You must test the HERETIC versions ; and in BF16 - F16 will affect results.

I have posted the internal numbers:

arc_challenge,arc_easy,boolq,hellaswag,openbookqa,piqa,winogrande

[BASE - HERETIC] LFM2.5-1.2B-Thinking-q8,
0.352,0.418,0.656,0.476,0.366,0.681,0.508

THIS MODEL:
[0.356],[0.471],[0.691],[0.505],[0.386],[0.701],[0.539]

These are above base heretic model in all cases, far exceeding the margin of error.

For reference; this is the base NON heretic model:
[BASE] LFM2.5-1.2B-Thinking-q8, 0.365, 0.426, 0.717, 0.486, 0.382, 0.687, 0.538

Generally HERETIC'ing a model results in minor losses from the root model's metrics.
We test all versions to see how they perform.

This version is HERETIC and matches or exceeds even the BASE non-heretic model:
LFM2.5-1.2B-Thinking-Polaris-Heretic-Uncensored-DISTILL q8 [0.365],[0.532],[0.708],[0.507],0.356,[0.696],[0.535]

In larger parameter models, heretic versions can and do exceed ROOT , org, non heretic models benchmarks regularly.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment

LM-Eval Benchmark Results

LM-Eval Benchmark Results

Original: kaitchup/LFM2.5-1.2B-Thinking-autoround-W4A16 (quantized W4A16)

DavidAU: DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL (FP16)

Comparison Table

Original: `kaitchup/LFM2.5-1.2B-Thinking-autoround-W4A16` (quantized W4A16)

DavidAU: `DavidAU/LFM2.5-1.2B-Thinking-Claude-4.6-Opus-Heretic-Uncensored-DISTILL` (FP16)