Llama-3.3-70B-Instruct-FP8-block

Model Overview

• Model Architecture: LlamaForCausalLM
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: FP8
  • Activation quantization: FP8
• Release Date:
• Version: 1.0
• Model Developers:: Red Hat

Quantized version of meta-llama/Llama-3.3-70B-Instruct.

Model Optimizations

This model was obtained by quantizing the weights and activations of meta-llama/Llama-3.3-70B-Instruct to FP8 data type. 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 of the language model are quantized.

Deployment

Use with vLLM

1. Initialize vLLM server:

vllm serve nm-testing/Llama-3.3-70B-Instruct-FP8-block --tensor_parallel_size 4

2. Send requests to the server:

from openai import OpenAI

# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://<your-server-host>:8000/v1"

client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)

model = "nm-testing/Llama-3.3-70B-Instruct-FP8-block"


messages = [
    {"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
]

outputs = client.chat.completions.create(
    model=model,
    messages=messages,
)

generated_text = outputs.choices[0].message.content
print(generated_text)

Creation

This model was quantized using the llm-compressor library as shown below.

Creation details

from transformers import AutoProcessor, LlamaForCausalLM

from llmcompressor import oneshot
from llmcompressor.modeling import replace_modules_for_calibration
from llmcompressor.modifiers.quantization import QuantizationModifier

MODEL_ID = "meta-llama/Llama-3.3-70B-Instruct"

# Load model.
model = LlamaForCausalLM.from_pretrained(MODEL_ID, dtype="auto")
processor = AutoProcessor.from_pretrained(MODEL_ID)
model = replace_modules_for_calibration(model)

# Configure the quantization algorithm and scheme.
# In this case, we:
#   * quantize the weights to fp8 with per-block quantization
#   * quantize the activations to fp8 with dynamic token activations
recipe = QuantizationModifier(
    targets="Linear",
    scheme="FP8_BLOCK",
    ignore=["lm_head"],
)

# Apply quantization.
oneshot(model=model, recipe=recipe)

# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-block"
model.save_pretrained(SAVE_DIR)
processor.save_pretrained(SAVE_DIR)

Evaluation

The model was evaluated on the OpenLLMv1 leaderboard task, using lm-evaluation-harness, on reasoning tasks using lighteval. vLLM was used for all evaluations.

Evaluation details

Openllm V1

lm_eval \
  --model vllm \
  --model_args pretrained="nm-testing/Llama-3.3-70B-Instruct-FP8-block",dtype=auto,add_bos_token=True,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks openllm \
  --write_out \
  --batch_size auto \
  --show_config

Openllm V2

lm_eval \
  --model vllm \
  --model_args pretrained="nm-testing/Llama-3.3-70B-Instruct-FP8-block",dtype=auto,add_bos_token=False,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.7,disable_log_stats=True,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks leaderboard \
  --apply_chat_template \
  --fewshot_as_multiturn \
  --write_out \
  --batch_size auto \
  --show_config

Coding Benchmarks

evalplus.evaluate --model "nm-testing/Llama-3.3-70B-Instruct-FP8-block" \
                  --dataset "humaneval" \
                  --backend vllm \
                  --tp 4 \
                  --greedy
evalplus.evaluate --model "nm-testing/Llama-3.3-70B-Instruct-FP8-block" \
                --dataset "mbpp" \
                --backend vllm \
                --tp 4 \
                --greedy

Accuracy

Category	Metric	meta-llama/Llama-3.3-70B-Instruct	nm-testing/Llama-3.3-70B-Instruct-FP8-block	Recovery (%)
OpenLLM V1	ARC-Challenge (Acc-Norm, 25-shot)	72.53	72.61	100.12
	GSM8K (Strict-Match, 5-shot)	76.35	73.16	95.83
	HellaSwag (Acc-Norm, 10-shot)	86.65	86.56	99.90
	MMLU (Acc, 5-shot)	82.51	82.38	99.84
	TruthfulQA (MC2, 0-shot)	62.83	62.64	99.69
	Winogrande (Acc, 5-shot)	83.50	83.27	99.72
	Average Score	77.39	76.77	99.20
OpenLLM V2	IFEval (Inst Level Strict Acc, 0-shot)	92.57	92.57	100.00
	BBH (Acc-Norm, 3-shot)	69.03	68.98	99.92
	Math-Hard (Exact-Match, 4-shot)	49.24	49.47	100.46
	GPQA (Acc-Norm, 0-shot)	32.63	32.63	100.00
	MUSR (Acc-Norm, 0-shot)	44.31	43.92	99.10
	MMLU-Pro (Acc, 5-shot)	53.55	53.56	100.02
	Average Score	56.89	56.85	99.93
Coding	HumanEval pass@1	82.90	82.90	100.00
	HumanEval+ pass@1	78.70	77.40	98.35
	MBPP pass@1	87.70	87.60	99.89
	MBPP+ pass@1	72.80	73.00	100.27