metadata

tags:
  - fp8
  - vllm
language:
  - en
  - de
  - fr
  - it
  - pt
  - hi
  - es
  - th
pipeline_tag: text-generation
license: llama3.2
base_model: meta-llama/Llama-3.2-3B-Instruct

Llama-3.2-3B-Instruct-FP8-dynamic

Model Overview

Model Architecture: Meta-Llama-3.2
- Input: Text
- Output: Text
Model Optimizations:
- Weight quantization: FP8
- Activation quantization: FP8
Intended Use Cases: Intended for commercial and research use in multiple languages. Similarly to Llama-3.2-3B-Instruct, this models is intended for assistant-like chat.
Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English.
Release Date: 9/25/2024
Version: 1.0
License(s): llama3.2
Model Developers: Neural Magic

Quantized version of Llama-3.2-3B-Instruct. It achieves an average score of 50.88 on a subset of task from the OpenLLM benchmark (version 1), whereas the unquantized model achieves 51.70.

Model Optimizations

This model was obtained by quantizing the weights and activations of Llama-3.2-3B-Instruct to FP8 data type, ready for inference with vLLM built from source. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.

Only the weights and activations of the linear operators within transformers blocks are quantized. Symmetric per-channel quantization is applied, in which a linear scaling per output dimension maps the FP8 representations of the quantized weights and activations. Activations are also quantized on a per-token dynamic basis. LLM Compressor is used for quantization.

Deployment

Use with vLLM

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic"

sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256)

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you?"},
]

prompts = tokenizer.apply_chat_template(messages, tokenize=False)

llm = LLM(model=model_id)

outputs = llm.generate(prompts, sampling_params)

generated_text = outputs[0].outputs[0].text
print(generated_text)

vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Creation

This model was created by applying LLM Compressor, as presented in the code snipet below.

import torch

from transformers import AutoTokenizer

from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
from llmcompressor.transformers.compression.helpers import (  # noqa
    calculate_offload_device_map,
    custom_offload_device_map,
)

recipe = """
quant_stage:
    quant_modifiers:
        QuantizationModifier:
            ignore: ["lm_head"]
            config_groups:
                group_0:
                    weights:
                        num_bits: 8
                        type: float
                        strategy: channel
                        dynamic: false
                        symmetric: true
                    input_activations:
                        num_bits: 8
                        type: float
                        strategy: token
                        dynamic: true
                        symmetric: true
                    targets: ["Linear"]
"""

model_stub = "meta-llama/Llama-3.2-3B-Instruct"
model_name = model_stub.split("/")[-1]

device_map = calculate_offload_device_map(
    model_stub, reserve_for_hessians=False, num_gpus=1, torch_dtype="auto"
)

model = SparseAutoModelForCausalLM.from_pretrained(
    model_stub, torch_dtype="auto", device_map=device_map
)

output_dir = f"./{model_name}-FP8-dynamic"

oneshot(
    model=model,
    recipe=recipe,
    output_dir=output_dir,
    save_compressed=True,
    tokenizer=AutoTokenizer.from_pretrained(model_stub),
)

Evaluation

The model was evaluated on MMLU, ARC-Challenge, GSM-8K, and Winogrande. Evaluation was conducted using the Neural Magic fork of lm-evaluation-harness (branch llama_3.1_instruct) and the vLLM engine. This version of the lm-evaluation-harness includes versions of ARC-Challenge, GSM-8K, MMLU, and MMLU-cot that match the prompting style of Meta-Llama-3.1-Instruct-evals.

Accuracy

Open LLM Leaderboard evaluation scores

Benchmark	Llama-3.2-3B-Instruct	Llama-3.2-3B-Instruct-FP8-dynamic (this model)	Recovery
MMLU (5-shot)	62.98	62.95	100.0%
MMLU-cot (0-shot)	65.40	65.23	99.7%
ARC Challenge (0-shot)	77.13	76.71	99.4%
GSM-8K-cot (8-shot, strict-match)	77.94	76.72	98.4%
Winogrande (5-shot)	71.11	71.11	100.0%
Hellaswag (10-shot)	73.62	73.54	99.9%
TruthfulQA (0-shot, mc2)	51.47	51.06	99.2%
Average	68.52	68.19	99.5%

Reproduction

The results were obtained using the following commands:

MMLU

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,max_model_len=3850,max_gen_toks=10,tensor_parallel_size=1 \
  --tasks mmlu_llama_3.1_instruct \
  --fewshot_as_multiturn \
  --apply_chat_template \
  --num_fewshot 5 \
  --batch_size auto

MMLU-CoT

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,max_model_len=4064,max_gen_toks=1024,tensor_parallel_size=1 \
  --tasks mmlu_cot_0shot_llama_3.1_instruct \
  --apply_chat_template \
  --num_fewshot 0 \
  --batch_size auto

ARC-Challenge

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,max_model_len=3940,max_gen_toks=100,tensor_parallel_size=1 \
  --tasks arc_challenge_llama_3.1_instruct \
  --apply_chat_template \
  --num_fewshot 0 \
  --batch_size auto

GSM-8K

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,max_model_len=4096,max_gen_toks=1024,tensor_parallel_size=1 \
  --tasks gsm8k_cot_llama_3.1_instruct \
  --fewshot_as_multiturn \
  --apply_chat_template \
  --num_fewshot 8 \
  --batch_size auto

Hellaswag

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \
  --tasks hellaswag \
  --num_fewshot 10 \
  --batch_size auto

Winogrande

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \
  --tasks winogrande \
  --num_fewshot 5 \
  --batch_size auto

TruthfulQA

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Llama-3.2-3B-Instruct-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \
  --tasks truthfulqa \
  --num_fewshot 0 \
  --batch_size auto