SmolLM-90M-Instruct-Pruned 🧠💡

A pruned version of HuggingFaceTB/SmolLM-135M-Instruct, reduced from 135M parameters to approximately 90M for faster inference and reduced memory usage, while maintaining reasonable performance for instruction-style tasks.

🔧 What’s Inside

• Base: SmolLM-135M-Instruct
• Parameters: ~90M
• Pruning method: Structured pruning (e.g., attention heads, MLP layers) using PyTorch/NVIDIA pruning tools (customize if needed).
• Vocabulary, tokenizer, and training objectives remain identical to the base model.

🚀 Intended Use

This model is optimized for:

• Low-latency applications
• Edge deployments
• Instruction-following tasks with compact models
• Use in environments with limited VRAM or compute

Example Use

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-135M-Instruct")
model = AutoModelForCausalLM.from_pretrained("your-username/SmolLM-90M-Instruct-Pruned")

prompt = "Explain quantum computing to a 10-year-old."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))