Instructions to use jkim96/phi-4-DASHQ-INT2-g32 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jkim96/phi-4-DASHQ-INT2-g32 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="jkim96/phi-4-DASHQ-INT2-g32", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("jkim96/phi-4-DASHQ-INT2-g32", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("jkim96/phi-4-DASHQ-INT2-g32", trust_remote_code=True, device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use jkim96/phi-4-DASHQ-INT2-g32 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jkim96/phi-4-DASHQ-INT2-g32"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jkim96/phi-4-DASHQ-INT2-g32",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/jkim96/phi-4-DASHQ-INT2-g32

SGLang

How to use jkim96/phi-4-DASHQ-INT2-g32 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "jkim96/phi-4-DASHQ-INT2-g32" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jkim96/phi-4-DASHQ-INT2-g32",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "jkim96/phi-4-DASHQ-INT2-g32" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jkim96/phi-4-DASHQ-INT2-g32",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use jkim96/phi-4-DASHQ-INT2-g32 with Docker Model Runner:
```
docker model run hf.co/jkim96/phi-4-DASHQ-INT2-g32
```

phi-4-DASHQ-INT2-g32 / README.md

jkim96

Add DASH-Q remote-code inference (Triton decode kernel)

0789322 verified about 7 hours ago

preview code

Raw

History Blame Contribute Delete

2.39 kB

metadata

license: mit
pipeline_tag: text-generation
base_model: microsoft/phi-4
base_model_relation: quantized
library_name: transformers
tags:
  - dashq
  - quantized
  - post-training-quantization
  - int2

phi-4-DASHQ-INT2-g32

DASH-Q — Diagonal-Aware Shrinkage for Robust PTQ. INT2 · group size 32 · 7.1679 GB (from 29.3190 GB — 4.1x smaller)

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "jkim96/phi-4-DASHQ-INT2-g32", trust_remote_code=True, device_map="cuda", dtype="auto"
)
tokenizer = AutoTokenizer.from_pretrained("jkim96/phi-4-DASHQ-INT2-g32")

messages = [{"role": "user", "content": "Explain 2-bit quantization in one sentence."}]
text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
print(tokenizer.decode(model.generate(**inputs, max_new_tokens=256)[0]))

trust_remote_code=True is required: the checkpoint ships its quantized-layer implementation (modeling_dashq.py) and Triton kernels (dashq_kernel.py). Without Triton, or on CPU, it falls back to dequantize-and-matmul in PyTorch.

Requirements

Package	Minimum	Verified with
`torch`	2.4	2.12.1+cu130
`transformers`	5.8	5.9.0
`triton`	3.0 (Linux; bundled with CUDA builds of PyTorch)	3.7.1
`huggingface_hub`	1.5 (pulled in by transformers)	1.15.0

Quantization

Field	Value
Base model	`microsoft/phi-4`
Precision	INT2, group size 32
Scale / zero dtype	float16
Calibration	wikitext2, 128 samples x 2048
Size	7.1679 GB · original 29.3190 GB · 4.1x compression

Benchmarks

Full zero-shot / few-shot results for every DASH-Q checkpoint: github.com/JaeminK/dashq#benchmarks

Evaluation

Metric	Value
`wikitext2_ppl`	7.9293
`zero-shot accuracy avg`	66.2368
`arc_challenge`	53.0717
`arc_easy`	77.0623
`commonsense_qa`	72.5635
`hellaswag`	73.3121
`lambada_openai`	72.1716
`openbookqa`	41.8000
`piqa`	79.3254
`truthfulqa_mc2`	54.1334
`winogrande`	72.6914