Instructions to use AmirMohseni/Velma-9b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AmirMohseni/Velma-9b with Transformers:

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

pipe = pipeline("text-generation", model="AmirMohseni/Velma-9b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AmirMohseni/Velma-9b")
model = AutoModelForCausalLM.from_pretrained("AmirMohseni/Velma-9b")
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 AmirMohseni/Velma-9b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AmirMohseni/Velma-9b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AmirMohseni/Velma-9b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/AmirMohseni/Velma-9b

SGLang

How to use AmirMohseni/Velma-9b 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 "AmirMohseni/Velma-9b" \
    --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": "AmirMohseni/Velma-9b",
		"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 "AmirMohseni/Velma-9b" \
        --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": "AmirMohseni/Velma-9b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use AmirMohseni/Velma-9b with Docker Model Runner:
```
docker model run hf.co/AmirMohseni/Velma-9b
```

Velma-9b

Model Overview

Velma-9b is a fine-tuned version of google/gemma-2-9b-it, optimized to improve reasoning capabilities. The model has been trained using the GSM8K dataset, a benchmark dataset designed to enhance mathematical and logical reasoning skills in language models.

This fine-tuning process allows Velma-9b to excel at structured problem-solving, step-by-step reasoning, and logical inference, making it an ideal choice for tasks requiring in-depth analytical thinking.

Features

Fine-Tuned on GSM8K: Enhanced for mathematical reasoning and step-by-step logical problem-solving.
Transformer-Based: Built on the powerful gemma-2-9b-it architecture.
Optimized for SFT (Supervised Fine-Tuning): Fine-tuned using TRL (Transformer Reinforcement Learning) for improved inference and structured output generation.
Efficient Deployment: Compatible with transformers and supports GPU acceleration for fast inference.

Quick Start

Use the following code to generate text using Velma-9b:

from transformers import pipeline

# Initialize the pipeline
generator = pipeline("text-generation", model="AmirMohseni/Velma-9b", device="cuda")

# Example prompt
question = "If you had a time machine but could only go to the past or the future once and never return, which would you choose and why?"

# Generate output
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training Procedure

Velma-9b was fine-tuned using the Supervised Fine-Tuning (SFT) approach with the GSM8K dataset. This dataset contains high-quality mathematical and logical reasoning problems that help models develop structured thinking and problem-solving skills.

Training Details

Base Model: google/gemma-2-9b-it
Dataset Used: GSM8K
Fine-Tuning Method: SFT(STaR) using TRL
Optimization Objective: Supervised fine-tuning to enhance structured reasoning

Framework Versions

TRL: 0.12.1
Transformers: 4.46.3
PyTorch: 2.1.1
Datasets: 3.1.0
Tokenizers: 0.20.3

Use Cases

Velma-9b is best suited for tasks requiring structured reasoning and logical inference:

Mathematical & Logical Reasoning Tasks: Providing step-by-step explanations and structured problem-solving.
AI Research & Experimentation: Evaluating fine-tuning strategies for reasoning-focused language models.

Downloads last month: 6

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for AmirMohseni/Velma-9b

Base model

google/gemma-2-9b

Finetuned

google/gemma-2-9b-it