metadata
license: llama3
base_model: meta-llama/Llama-3.2-1B-Instruct
pipeline_tag: text-generation
library_name: transformers
language:
- en
- de
- es
- fr
- th
- pt
tags:
- meta
- pytorch
- llama
- llama-3
- quantization
- fp8
- compressed-model
Llama-3.2-1B-Instruct-FP8-Dynamic
FP8 dynamic quantization pipeline for Llama-3.2-1B-Instruct using llm_compressor.
Overview
- This repository demonstrates how to apply FP8 dynamic quantization to the Llama-3.2-1B-Instruct model.
- The goal is to reduce memory usage and improve inference efficiency while maintaining strong performance for text generation and instruction-following tasks.
⚠️ This is a quantization pipeline, not a pre-quantized checkpoint.
Usage
from transformers import AutoTokenizer, AutoModelForCausalLM
from llmcompressor.transformers import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
# Define the model ID for the model you want to quantize
MODEL_ID = "meta-llama/Llama-3.2-1B-Instruct"
# Load the model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
MODEL_ID, device_map="auto", torch_dtype="auto"
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
# Configure the quantization recipe
recipe = QuantizationModifier(targets="Linear", scheme="FP8_DYNAMIC", ignore=["lm_head"])
# Apply the quantization algorithm
oneshot(model=model, recipe=recipe)
# Define the directory to save the quantized model
SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-Dynamic"
# Save the quantized model and tokenizer
model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)
print(f"Quantized model saved to (SAVE_DIR)")