metadata

library_name: transformers
tags:
  - torchao
  - phi
  - phi4
  - nlp
  - code
  - math
  - chat
  - conversational
license: mit
language:
  - multilingual
base_model:
  - microsoft/Phi-4-mini-instruct
pipeline_tag: text-generation

Installation

pip install git+https://github.com/huggingface/transformers@main
pip install --pre torchao --index-url https://download.pytorch.org/whl/nightly/cu126
pip install vllm --pre --extra-index-url https://wheels.vllm.ai/nightly

Also need to install lm-eval from source: https://github.com/EleutherAI/lm-evaluation-harness#install

Quantization Recipe

We used following code to get the quantized model:

from transformers import (
  AutoModelForCausalLM,
  AutoProcessor,
  AutoTokenizer,
  TorchAoConfig,
)
from torchao.quantization.quant_api import (
    IntxWeightOnlyConfig,
    Int8DynamicActivationIntxWeightConfig,
    AOPerModuleConfig,
    quantize_,
)
from torchao.quantization.granularity import PerGroup, PerAxis
import torch

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

embedding_config = IntxWeightOnlyConfig(
    weight_dtype=torch.int8,
    granularity=PerAxis(0),
)
linear_config = Int8DynamicActivationIntxWeightConfig(
    weight_dtype=torch.int4,
    weight_granularity=PerGroup(32),
    weight_scale_dtype=torch.bfloat16,
)
quantized_model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float32, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_id)

# TODO: use AOPerModuleConfig once fix for tied weights is landed 
quantize_(
    quantized_model,
    embedding_config,
    lambda m, fqn: isinstance(m, torch.nn.Embedding) 
)
quantize_(
    quantized_model,
    linear_config,
)

# Push to hub
# USER_ID = "YOUR_USER_ID"
# save_to = f"{USER_ID}/phi4-mini-8dq4w"
# 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?"
messages = [
    {
        "role": "system",
        "content": "",
    },
    {"role": "user", "content": prompt},
]
templated_prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
print("Prompt:", prompt)
print("Templated prompt:", templated_prompt)
inputs = tokenizer(
    templated_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("Response:", output_text[0][len(prompt):])


# Save to disk
state_dict = quantized_model.state_dict()
torch.save(state_dict, "phi4-mini-8dq4w.bin")

The response from the manual testing is:

Hello! As an AI, I don't have consciousness in the way humans do, but I am fully operational and here to assist you. How can I help you today?

Model Quality

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

baseline

lm_eval --model hf --model_args pretrained=microsoft/Phi-4-mini-instruct --tasks hellaswag --device cuda:0 --batch_size 64

8dq4w

import lm_eval
from lm_eval import evaluator
from lm_eval.utils import (
    make_table,
)

lm_eval_model = lm_eval.models.huggingface.HFLM(pretrained=quantized_model, batch_size=64)
results = evaluator.simple_evaluate(
    lm_eval_model, tasks=["hellaswag"], device="cuda:0", batch_size="auto"
)
print(make_table(results))

Benchmark
	Phi-4 mini-Ins	phi4-mini-8dq4w
Popular aggregated benchmark
Reasoning
HellaSwag	54.57	53.24
Multilingual
Math
Overall	TODO	TODO

Exporting to ExecuTorch

Exporting to ExecuTorch requires you clone and install ExecuTorch.

Convert quantized checkpoint to ExecuTorch's format

python -m executorch.examples.models.phi_4_mini.convert_weights phi4-mini-8dq4w.bin phi4-mini-8dq4w-converted.bin

Export to an ExecuTorch *.pte with XNNPACK

PARAMS="executorch/examples/models/phi_4_mini/config.json"
python -m executorch.examples.models.llama.export_llama \
  --model "phi_4_mini" \
  --checkpoint "phi4-mini-8dq4w-converted.bin" \
  --params "$PARAMS" \
  -kv \
  --use_sdpa_with_kv_cache \
  -X \
  --output_name="phi4-mini-8dq4w.pte"

Run model with pybindings

export TOKENIZER="/path/to/tokenizer.json"
export TOKENIZER_CONFIG="/path/to/tokenizer_config.json"
export PROMPT="<|system|><|end|><|user|>Hey, are you conscious? Can you talk to me?<|end|><|assistant|>"
python -m executorch.examples.models.llama.runner.native \
  --model phi_4_mini \
  --pte phi4-mini-8dq4w.pte \
  -kv \
  --tokenizer ${TOKENIZER} \
  --tokenizer_config ${TOKENIZER_CONFIG} \
  --prompt "${PROMPT}" \
  --params "${PARAMS}" \
  --max_len 128 \
  --temperature 0

The output is:

Hello! I am Phi, an AI developed by Microsoft. I am not conscious in the way humans are, but I am here to help and converse with you. How can I assist you today?Hello! I am Phi, an AI developed by Microsoft. I am not conscious in the way humans are, but I am here to help and converse with you. How can I assist you today?Hello! I am Phi, an AI developed by Microsoft. I am not conscious in the way humans are, but I am here to

Note: the runner does not currently recongize the stop token from Phi 4 Mini, so it generates text beyond when it should stop.

TODO: link to iOS app once ready.