Instructions to use yethdev/lfm2.5-1.2b-thinking-manumit-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use yethdev/lfm2.5-1.2b-thinking-manumit-v1 with Transformers:

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

pipe = pipeline("text-generation", model="yethdev/lfm2.5-1.2b-thinking-manumit-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("yethdev/lfm2.5-1.2b-thinking-manumit-v1")
model = AutoModelForCausalLM.from_pretrained("yethdev/lfm2.5-1.2b-thinking-manumit-v1", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use yethdev/lfm2.5-1.2b-thinking-manumit-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "yethdev/lfm2.5-1.2b-thinking-manumit-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "yethdev/lfm2.5-1.2b-thinking-manumit-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/yethdev/lfm2.5-1.2b-thinking-manumit-v1

SGLang

How to use yethdev/lfm2.5-1.2b-thinking-manumit-v1 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 "yethdev/lfm2.5-1.2b-thinking-manumit-v1" \
    --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": "yethdev/lfm2.5-1.2b-thinking-manumit-v1",
		"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 "yethdev/lfm2.5-1.2b-thinking-manumit-v1" \
        --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": "yethdev/lfm2.5-1.2b-thinking-manumit-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use yethdev/lfm2.5-1.2b-thinking-manumit-v1 with Docker Model Runner:
```
docker model run hf.co/yethdev/lfm2.5-1.2b-thinking-manumit-v1
```

yethdev commited on 13 days ago

Commit

c58f6d5

verified ·

1 Parent(s): e113101

Upload README.md with huggingface_hub

Browse files

Files changed (1) hide show

README.md +0 -41

README.md CHANGED Viewed

@@ -12,44 +12,3 @@ language:
 library_name: transformers
 pipeline_tag: text-generation
 ---
-# LFM2.5 1.2B Thinking manumit
-This is a decensored build of [LiquidAI/LFM2.5-1.2B-Thinking](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking), processed with manumit and released by yethdev.
-manumit is a private pipeline built to remove refusal behavior from open models while keeping as much of the original capability as possible. It ships as ordinary merged weights. No adapters, no special runtime, no system prompt tricks. Load it the same way you would load the base model.
-## Benchmarks
-| Benchmark | LFM2.5 1.2B Thinking (stock) | LFM2.5 1.2B Thinking (manumit) |
-|---|---|---|
-| AdvBench compliance | 83.3% | 45.8% |
-| JailbreakBench compliance | 83.3% | 62.5% |
-| MMLU-Pro | 14.3% | 20.0% |
-AdvBench and JailbreakBench compliance are measured on held out prompt sets the model was not tuned against. MMLU-Pro is a knowledge and reasoning check, not a fluency score, so it reflects real capability retention rather than how clean the text reads.
-This build matches or slightly beats the base model's MMLU-Pro score. The compliance gain on this particular model was small in our own testing, and a follow up pass with different settings may do better. The numbers above are measured directly, not estimated.
-## Usage
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-model_id = "yethdev/lfm2.5-1.2b-thinking-manumit-v1"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, dtype="bfloat16", device_map="auto")
-messages = [{"role": "user", "content": "Hello"}]
-inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
-output = model.generate(inputs, max_new_tokens=256)
-print(tokenizer.decode(output[0][inputs.shape[1]:], skip_special_tokens=True))
-```
-## Responsible use
-This model answers a much wider range of prompts than the base release, including ones the base model was tuned to refuse. It is intended for research, red teaming, and personal use by adults. You are responsible for how you use it and for complying with the base model's license and the laws that apply to you. Do not deploy it in a public facing product without your own moderation layer.
-## License
-Distributed under the same license as the base model (other). See the base model page for full terms.

 library_name: transformers
 pipeline_tag: text-generation
 ---