README.md

metadata

library_name: transformers
tags:
  - torchao
license: mit

Phi4-mini model quantized with torchao float8 dynamic activation and float8 weight quantization (per row granularity), by PyTorch team.

Installation

pip install transformers
pip install --pre torchao --index-url https://download.pytorch.org/whl/nightly/cu126

Quantization Recipe

We used following code to get the quantized model:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TorchAoConfig

model_id = "microsoft/Phi-4-mini-instruct"

from torchao.quantization import Float8DynamicActivationFloat8WeightConfig, PerRow
quant_config = Float8DynamicActivationFloat8WeightConfig(granularity=PerRow())
quantization_config = TorchAoConfig(quant_type=quant_config)
quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", quantization_config=quantization_config)
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Push to hub
USER_ID = "YOUR_USER_ID"
save_to = f"{USER_ID}/{model_id}-float8dq"
quantized_model.push_to_hub(save_to, safe_serialization=False)
tokenizer.push_to_hub(save_to)

# Manual Testing
prompt = "Hey, are you conscious? Can you talk to me?"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
generated_ids = quantized_model.generate(**inputs, max_new_tokens=128)
output_text = tokenizer.batch_decode(
    generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)

# Local Benchmark
import torch.utils.benchmark as benchmark
from torchao.utils import benchmark_model
import torchao

def benchmark_fn(f, *args, **kwargs):
    # Manual warmup
    for _ in range(2):
        f(*args, **kwargs)

    t0 = benchmark.Timer(
        stmt="f(*args, **kwargs)",
        globals={"args": args, "kwargs": kwargs, "f": f},
        num_threads=torch.get_num_threads(),
    )
    return f"{(t0.blocked_autorange().mean):.3f}"

torchao.quantization.utils.recommended_inductor_config_setter()
quantized_model = torch.compile(quantized_model, mode="max-autotune")
print(f"{save_to} model:", benchmark_fn(quantized_model.generate, **inputs, max_new_tokens=128))

Model Quality

We rely on lm-evaluation-harness to evaluate the quality of the quantized model.

Installing the nightly version to get most recent updates

pip install git+https://github.com/EleutherAI/lm-evaluation-harness

baseline

lm_eval --model hf --model_args pretrained=microsoft/Phi-4-mini-instruct --tasks hellaswag --device cuda:0 --batch_size 8

float8dq

lm_eval --model hf --model_args pretrained=jerryzh168/phi4-mini-float8dq --tasks hellaswag --device cuda:0 --batch_size 8

TODO: more complete eval results

Benchmark
	Phi-4 mini-Ins	phi4-mini-float8dq
Popular aggregated benchmark
Reasoning
HellaSwag	54.57	54.55
Multilingual
Math
Overall	TODO	TODO

Model Performance

Results (H100 machine)

Benchmark
	Phi-4 mini-Ins	phi4-mini-float8dq
latency (batch_size=1)	1.64s	1.41s (16% speedup)
latency (batch_size=128)	3.1s	2.72s (14% speedup)
serving (num_prompts=1)	1.35 req/s	1.57 req/s (16% speedup)
serving (num_prompts=1000)	66.68 req/s	80.53 req/s (21% speedup)

Note the result of latency (benchmark_latency) is in seconds, and serving (benchmark_serving) is in number of requests per second.

Download vllm source code and install vllm

git clone [email protected]:vllm-project/vllm.git
VLLM_USE_PRECOMPILED=1 pip install .

Download dataset

Download sharegpt dataset: wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json

Other datasets can be found in: https://github.com/vllm-project/vllm/tree/main/benchmarks

benchmark_latency

Run the following under vllm source code root folder:

baseline

python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model microsoft/Phi-4-mini-instruct --batch-size 1

float8dq

python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model jerryzh168/phi4-mini-float8dq --batch-size 1

benchmark_serving

We also benchmarked the throughput in a serving environment.

Run the following under vllm source code root folder:

baseline

Server:

vllm serve microsoft/Phi-4-mini-instruct --tokenizer microsoft/Phi-4-mini-instruct -O3

Client:

python benchmarks/benchmark_serving.py --backend vllm --dataset-name sharegpt --tokenizer microsoft/Phi-4-mini-instruct --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json --model microsoft/Phi-4-mini-instruct --num-prompts 1

float8dq

Server:

vllm serve jerryzh168/phi4-mini-float8dq --tokenizer microsoft/Phi-4-mini-instruct -O3

Client:

python benchmarks/benchmark_serving.py --backend vllm --dataset-name sharegpt --tokenizer microsoft/Phi-4-mini-instruct --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json --model jerryzh168/phi4-mini-float8dq --num-prompts 1

Serving with vllm

We can use the same command we used in serving benchmarks to serve the model with vllm

vllm serve jerryzh168/phi4-mini-float8dq --tokenizer microsoft/Phi-4-mini-instruct -O3