Instructions to use ProjectGlue/smollm3-dolly-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ProjectGlue/smollm3-dolly-lora with Transformers:

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

pipe = pipeline("text-generation", model="ProjectGlue/smollm3-dolly-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("ProjectGlue/smollm3-dolly-lora", device_map="auto")

PEFT
How to use ProjectGlue/smollm3-dolly-lora with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ProjectGlue/smollm3-dolly-lora with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ProjectGlue/smollm3-dolly-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": "ProjectGlue/smollm3-dolly-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ProjectGlue/smollm3-dolly-lora

SGLang

How to use ProjectGlue/smollm3-dolly-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 "ProjectGlue/smollm3-dolly-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": "ProjectGlue/smollm3-dolly-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 "ProjectGlue/smollm3-dolly-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": "ProjectGlue/smollm3-dolly-lora",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ProjectGlue/smollm3-dolly-lora with Docker Model Runner:
```
docker model run hf.co/ProjectGlue/smollm3-dolly-lora
```

SmolLM3-3B Dolly LoRA

A LoRA fine-tuned adapter for SmolLM3-3B, trained on a 500-row subset of the Databricks Dolly 15k closed QA dataset as part of a structured HuggingFace learning curriculum.

Model Details

Model Description

Developed by: ProjectGlue
Model type: Causal Language Model with LoRA adapter
Language(s): English
License: Apache 2.0
Finetuned from: HuggingFaceTB/SmolLM3-3B

Uses

Direct Use

Question answering with provided context. Load the base SmolLM3-3B model with this LoRA adapter for improved closed QA responses.

Out-of-Scope Use

Not intended for production use. This is a learning exercise and proof-of-concept fine-tuning run.

How to Get Started with the Model

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

base_model = AutoModelForCausalLM.from_pretrained(
    "HuggingFaceTB/SmolLM3-3B",
    torch_dtype=torch.float16,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, "ProjectGlue/smollm3-dolly-lora")
tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM3-3B")

Training Details

Training Data

500-row subset of the Databricks Dolly 15k dataset, filtered to the closed_qa category. Formatted using SmolLM3's chat template with system, user, and assistant roles.

Training Hyperparameters

Method: LoRA (Low-Rank Adaptation)
LoRA rank: 16
LoRA alpha: 32
Target modules: q_proj, k_proj, v_proj, o_proj
Dropout: 0.05
Trainable parameters: 7,667,712 (0.25% of total)
Epochs: 1
Batch size: 1 (gradient accumulation steps: 8)
Learning rate: 2e-4
Precision: fp16

Evaluation

Results

Initial loss: 1.824
Final loss: 1.505
Loss decreased consistently over 63 steps indicating successful training.

Environmental Impact

Hardware Type: NVIDIA T4 GPU (16GB)
Hours used: ~0.1
Cloud Provider: Google Colab
Compute Region: US

Model Card Authors

ProjectGlue

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for ProjectGlue/smollm3-dolly-lora

Base model

HuggingFaceTB/SmolLM3-3B-Base

Finetuned

HuggingFaceTB/SmolLM3-3B

Adapter

(70)

this model

ProjectGlue
/

smollm3-dolly-lora