Instructions to use yvliet/perron-gemma4-e2b-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use yvliet/perron-gemma4-e2b-lora with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("google/gemma-4-e2b-it")
model = PeftModel.from_pretrained(base_model, "yvliet/perron-gemma4-e2b-lora")

Transformers

How to use yvliet/perron-gemma4-e2b-lora with Transformers:

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

pipe = pipeline("text-generation", model="yvliet/perron-gemma4-e2b-lora")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("yvliet/perron-gemma4-e2b-lora", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use yvliet/perron-gemma4-e2b-lora with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "yvliet/perron-gemma4-e2b-lora"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "yvliet/perron-gemma4-e2b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/yvliet/perron-gemma4-e2b-lora

SGLang

How to use yvliet/perron-gemma4-e2b-lora 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 "yvliet/perron-gemma4-e2b-lora" \
    --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": "yvliet/perron-gemma4-e2b-lora",
		"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 "yvliet/perron-gemma4-e2b-lora" \
        --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": "yvliet/perron-gemma4-e2b-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use yvliet/perron-gemma4-e2b-lora with Docker Model Runner:
```
docker model run hf.co/yvliet/perron-gemma4-e2b-lora
```

Perron Gemma 4 E2B LoRA Adapter

Parameter-efficient Low-Rank Adaptation (LoRA) adapter for Google DeepMind's google/gemma-4-e2b-it model, trained on provenance-conditioned candidate ranking over SWE-bench dependency multigraphs.

Model Summary

Author: Sultan Haikal (GitHub: @yvliet)
Base Model: google/gemma-4-e2b-it
Architecture: Dense PLE Transformer, 2B parameters
Adapter Configuration: LoRA rank $r=16$, $\alpha=32$, dropout $0.05$ targeting language model attention projections (q_proj, v_proj, k_proj, o_proj)
Trainable Parameters: ~4.7M (<0.25% of total base parameters)
License: Apache 2.0

Training Details

Objective: Provenance-Conditioned Ranking Cross-Entropy ($\mathcal{L}_{\text{Rank}}$) over atomic candidate choices [A] through [Y]
Active Loss Slicing: Sequence length 2048 with tail token slicing (logits_to_keep=24)
Optimization: AdamW, learning rate 2e-4, linear warmup with cosine decay
Hardware: Dual Tesla T4 GPUs (sharded pipeline via device_map="auto")
Validation Receipts ($N=50$ held-out pairs):
- Epoch 1: Loss 0.6391, Val Acc@1: 98.0%, Val Acc@5: 100.0%
- Epoch 2: Loss 0.0074, Val Acc@1: 100.0%, Val Acc@5: 100.0%
- Epoch 3: Loss 0.0002, Val Acc@1: 100.0%, Val Acc@5: 100.0%

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "google/gemma-4-e2b-it"
adapter_id = "yvliet/perron-gemma4-e2b-lora"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float16,
    device_map="auto",
)
model = PeftModel.from_pretrained(base_model, adapter_id)

Citation & Resources

Repository: https://github.com/yvliet/perron
PyPI Library: https://pypi.org/project/perron-core/

Downloads last month: 8