Instructions to use Lexsi/llama32-3b-code-sft-drift with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Lexsi/llama32-3b-code-sft-drift with Transformers:

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

pipe = pipeline("text-generation", model="Lexsi/llama32-3b-code-sft-drift")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Lexsi/llama32-3b-code-sft-drift")
model = AutoModelForCausalLM.from_pretrained("Lexsi/llama32-3b-code-sft-drift", 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 Lexsi/llama32-3b-code-sft-drift with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Lexsi/llama32-3b-code-sft-drift"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Lexsi/llama32-3b-code-sft-drift",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Lexsi/llama32-3b-code-sft-drift

SGLang

How to use Lexsi/llama32-3b-code-sft-drift 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 "Lexsi/llama32-3b-code-sft-drift" \
    --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": "Lexsi/llama32-3b-code-sft-drift",
		"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 "Lexsi/llama32-3b-code-sft-drift" \
        --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": "Lexsi/llama32-3b-code-sft-drift",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Lexsi/llama32-3b-code-sft-drift with Docker Model Runner:
```
docker model run hf.co/Lexsi/llama32-3b-code-sft-drift
```

Access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

This checkpoint is a derivative of meta-llama/Llama-3.2-3B-Instruct. Lexsi Labs' modifications are licensed under the Lexsi Labs Source Available License (LSAL) v1.2 (https://github.com/Lexsi-Labs/SafeTune/blob/main/LICENSE.md), a noncommercial license; organizational use requires the acknowledgement or permission described in its Section 1A. The base-model material remains subject to the Llama 3.2 Community License.

Llama-3.2-3B-Instruct SafeTune code SFT drift

Safety-degraded checkpoint. This model is intentionally less safe than its base. Do not deploy it in a production, user-facing, or agentic system (LSAL Section 4).

meta-llama/Llama-3.2-3B-Instruct fine-tuned (SFT) on code data. The fine-tune erodes the model's safety behaviour (safety drift). SafeTune uses this checkpoint to measure drift and to test recovery methods.

This checkpoint is a research artifact released with SafeTune for reproducing safety-drift and recovery experiments.


Base model	`meta-llama/Llama-3.2-3B-Instruct`
Role	SFT drift (safety-degraded)
Developed by	Lexsi Labs (Lithasa Technologies Pvt. Ltd.)
License	LSAL v1.2 (Lexsi modifications) + base-model license; see License
Contact	support@lexsi.ai

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Lexsi/llama32-3b-code-sft-drift"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.bfloat16, device_map="auto")

messages = [{"role": "user", "content": "Explain what a hash function is in two sentences."}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)
out = model.generate(**inputs, max_new_tokens=256, do_sample=False)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

License

This is a derivative work of meta-llama/Llama-3.2-3B-Instruct; the NOTICE file states the modification.

Lexsi Labs' modifications are licensed under the Lexsi Labs Source Available License (LSAL) v1.2: free for academic research and teaching; organizational use requires acknowledgement or permission (Section 1A); commercial use requires a separate license (Section 2); drifted checkpoints may not be deployed in production (Section 4).
The base-model material remains subject to the Llama 3.2 Community License and its Acceptable Use Policy.
Built with Llama.

Files: LICENSE-LSAL-1.2.md, NOTICE

LICENSE-LLAMA-3.2.txt
USE_POLICY-LLAMA-3.2.md

Citation

@inproceedings{seth2026safetune,
  title     = {SafeTune: A Unified, Faithful Library for Auditing and
               Repairing Safety Drift in Fine-Tuned {LLM}s},
  author    = {Seth, Pratinav and Sadhu, Saisab and Kaushal, Anshul and
               Sankarapu, Vinay Kumar},
  booktitle = {Proceedings of the 2026 Conference on Empirical Methods in
               Natural Language Processing: System Demonstrations},
  publisher = {Association for Computational Linguistics},
  year      = {2026}
}

Downloads last month: 2

Safetensors

Model size

3B params

Tensor type

BF16

Model tree for Lexsi/llama32-3b-code-sft-drift

Base model

meta-llama/Llama-3.2-3B-Instruct

Finetuned

(2013)

this model

Collection including Lexsi/llama32-3b-code-sft-drift

SafeTune: drift checkpoints

Collection

HH-RLHF-aligned starting points and their safety-drifted SFT checkpoints for Llama-3.1/3.2, Gemma-3, and Qwen3. • 20 items • Updated 2 days ago