Instructions to use PointGuardAI/LFM2.5-1.2B-Instruct-INT8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use PointGuardAI/LFM2.5-1.2B-Instruct-INT8 with Transformers:

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

pipe = pipeline("text-generation", model="PointGuardAI/LFM2.5-1.2B-Instruct-INT8")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("PointGuardAI/LFM2.5-1.2B-Instruct-INT8")
model = AutoModelForCausalLM.from_pretrained("PointGuardAI/LFM2.5-1.2B-Instruct-INT8", 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 PointGuardAI/LFM2.5-1.2B-Instruct-INT8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "PointGuardAI/LFM2.5-1.2B-Instruct-INT8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PointGuardAI/LFM2.5-1.2B-Instruct-INT8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/PointGuardAI/LFM2.5-1.2B-Instruct-INT8

SGLang

How to use PointGuardAI/LFM2.5-1.2B-Instruct-INT8 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 "PointGuardAI/LFM2.5-1.2B-Instruct-INT8" \
    --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": "PointGuardAI/LFM2.5-1.2B-Instruct-INT8",
		"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 "PointGuardAI/LFM2.5-1.2B-Instruct-INT8" \
        --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": "PointGuardAI/LFM2.5-1.2B-Instruct-INT8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use PointGuardAI/LFM2.5-1.2B-Instruct-INT8 with Docker Model Runner:
```
docker model run hf.co/PointGuardAI/LFM2.5-1.2B-Instruct-INT8
```

LFM2.5-1.2B-Instruct-INT8 / quantization_manifest.json

shubham-kothari

Publish LFM2.5-1.2B-Instruct W8A8 INT8 checkpoint

f1eb922 verified 7 days ago

Raw

History Blame Contribute Delete

4.41 kB

	{
	"schema_version": 1,
	"created_at": "2026-07-28T12:17:52.765605+00:00",
	"scheme": "W8A8",
	"base_model": {
	"name_or_path": "LiquidAI/LFM2.5-1.2B-Instruct",
	"revision": "868df74dd56ff8a0c2ac5dbf281690c2dbebe4c9",
	"tokenizer_revision": "868df74dd56ff8a0c2ac5dbf281690c2dbebe4c9"
	},
	"source_run": "/workspace/runs/application_policy_match/20260728T113822Z-e31ae7185d",
	"quantization_pipeline": {
	"type": "data_free",
	"reason": "llm-compressor selected DataFreePipeline for QuantizationModifier with dynamic W8A8 activations"
	},
	"vllm_compatibility": {
	"excluded_module_family": "Liquid convolution projections",
	"reason": "Transformers names these modules conv while vLLM names them short_conv; vLLM 0.24.0 does not remap compressed scale names",
	"excluded_targets": [
	"model.layers.0.conv.in_proj",
	"model.layers.0.conv.out_proj",
	"model.layers.1.conv.in_proj",
	"model.layers.1.conv.out_proj",
	"model.layers.11.conv.in_proj",
	"model.layers.11.conv.out_proj",
	"model.layers.13.conv.in_proj",
	"model.layers.13.conv.out_proj",
	"model.layers.15.conv.in_proj",
	"model.layers.15.conv.out_proj",
	"model.layers.3.conv.in_proj",
	"model.layers.3.conv.out_proj",
	"model.layers.4.conv.in_proj",
	"model.layers.4.conv.out_proj",
	"model.layers.6.conv.in_proj",
	"model.layers.6.conv.out_proj",
	"model.layers.7.conv.in_proj",
	"model.layers.7.conv.out_proj",
	"model.layers.9.conv.in_proj",
	"model.layers.9.conv.out_proj"
	]
	},
	"target_modules": [
	"model.layers.0.feed_forward.w1",
	"model.layers.0.feed_forward.w2",
	"model.layers.0.feed_forward.w3",
	"model.layers.1.feed_forward.w1",
	"model.layers.1.feed_forward.w2",
	"model.layers.1.feed_forward.w3",
	"model.layers.10.feed_forward.w1",
	"model.layers.10.feed_forward.w2",
	"model.layers.10.feed_forward.w3",
	"model.layers.10.self_attn.k_proj",
	"model.layers.10.self_attn.out_proj",
	"model.layers.10.self_attn.q_proj",
	"model.layers.10.self_attn.v_proj",
	"model.layers.11.feed_forward.w1",
	"model.layers.11.feed_forward.w2",
	"model.layers.11.feed_forward.w3",
	"model.layers.12.feed_forward.w1",
	"model.layers.12.feed_forward.w2",
	"model.layers.12.feed_forward.w3",
	"model.layers.12.self_attn.k_proj",
	"model.layers.12.self_attn.out_proj",
	"model.layers.12.self_attn.q_proj",
	"model.layers.12.self_attn.v_proj",
	"model.layers.13.feed_forward.w1",
	"model.layers.13.feed_forward.w2",
	"model.layers.13.feed_forward.w3",
	"model.layers.14.feed_forward.w1",
	"model.layers.14.feed_forward.w2",
	"model.layers.14.feed_forward.w3",
	"model.layers.14.self_attn.k_proj",
	"model.layers.14.self_attn.out_proj",
	"model.layers.14.self_attn.q_proj",
	"model.layers.14.self_attn.v_proj",
	"model.layers.15.feed_forward.w1",
	"model.layers.15.feed_forward.w2",
	"model.layers.15.feed_forward.w3",
	"model.layers.2.feed_forward.w1",
	"model.layers.2.feed_forward.w2",
	"model.layers.2.feed_forward.w3",
	"model.layers.2.self_attn.k_proj",
	"model.layers.2.self_attn.out_proj",
	"model.layers.2.self_attn.q_proj",
	"model.layers.2.self_attn.v_proj",
	"model.layers.3.feed_forward.w1",
	"model.layers.3.feed_forward.w2",
	"model.layers.3.feed_forward.w3",
	"model.layers.4.feed_forward.w1",
	"model.layers.4.feed_forward.w2",
	"model.layers.4.feed_forward.w3",
	"model.layers.5.feed_forward.w1",
	"model.layers.5.feed_forward.w2",
	"model.layers.5.feed_forward.w3",
	"model.layers.5.self_attn.k_proj",
	"model.layers.5.self_attn.out_proj",
	"model.layers.5.self_attn.q_proj",
	"model.layers.5.self_attn.v_proj",
	"model.layers.6.feed_forward.w1",
	"model.layers.6.feed_forward.w2",
	"model.layers.6.feed_forward.w3",
	"model.layers.7.feed_forward.w1",
	"model.layers.7.feed_forward.w2",
	"model.layers.7.feed_forward.w3",
	"model.layers.8.feed_forward.w1",
	"model.layers.8.feed_forward.w2",
	"model.layers.8.feed_forward.w3",
	"model.layers.8.self_attn.k_proj",
	"model.layers.8.self_attn.out_proj",
	"model.layers.8.self_attn.q_proj",
	"model.layers.8.self_attn.v_proj",
	"model.layers.9.feed_forward.w1",
	"model.layers.9.feed_forward.w2",
	"model.layers.9.feed_forward.w3"
	]
	}