Instructions to use amd/gpt-oss-20b-BF16-w4a16-llmcompressor with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use amd/gpt-oss-20b-BF16-w4a16-llmcompressor with Transformers:

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

pipe = pipeline("text-generation", model="amd/gpt-oss-20b-BF16-w4a16-llmcompressor")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("amd/gpt-oss-20b-BF16-w4a16-llmcompressor")
model = AutoModelForCausalLM.from_pretrained("amd/gpt-oss-20b-BF16-w4a16-llmcompressor", 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 amd/gpt-oss-20b-BF16-w4a16-llmcompressor with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "amd/gpt-oss-20b-BF16-w4a16-llmcompressor"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "amd/gpt-oss-20b-BF16-w4a16-llmcompressor",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/amd/gpt-oss-20b-BF16-w4a16-llmcompressor

SGLang

How to use amd/gpt-oss-20b-BF16-w4a16-llmcompressor 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 "amd/gpt-oss-20b-BF16-w4a16-llmcompressor" \
    --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": "amd/gpt-oss-20b-BF16-w4a16-llmcompressor",
		"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 "amd/gpt-oss-20b-BF16-w4a16-llmcompressor" \
        --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": "amd/gpt-oss-20b-BF16-w4a16-llmcompressor",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use amd/gpt-oss-20b-BF16-w4a16-llmcompressor with Docker Model Runner:
```
docker model run hf.co/amd/gpt-oss-20b-BF16-w4a16-llmcompressor
```

gpt-oss-20b-BF16-w8a8-llmcompressor-v0.13.0

Model Overview

Model Architecture: GptOssForCausalLM
- Input: Text
- Output: Text
Source Model: gpt-oss-20b-BF16
Supported Hardware: AMD EPYC (CPU inference)
Preferred Operating System: Linux
Inference Engine: vLLM v0.28.0
Quantization Framework: LLM Compressor v0.13.0
Quantization Method: 8-bit Weight, 8-bit Dynamic Activation Quantization (W8A8)
Compatible Stack:
- ZenDNN v6.1.0
- ZenTorch v2.13.0.0
- PyTorch v2.13.0.0
- LLM Compressor v0.13.0
- vLLM v0.28.0

This is a quantized version of gpt-oss-20b-BF16 created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference.

Quantization

The model was quantized from gpt-oss-20b-BF16 using LLM Compressor via the Round-to-Nearest (RTN) algorithm. This reduces the model weights from 39.0 GiB to 20.6 GiB on disk (~47% reduction).

Method: 8-bit Weight, 8-bit Dynamic Activation Quantization (W8A8)
Config: compressed-tensors, num_bits=8, type=int, symmetric=true
Weights: INT8, symmetric, per-channel (static)
Activations: INT8, symmetric, per-token (dynamic)
Quantized: all 32 routed experts in each of the 24 layers (mlp.experts.*.{gate,up,down}_proj) and self_attn.{q,k,v,o}_proj.
Kept in BF16: the MoE routers (mlp.router), the attention sinks (self_attn.sinks), every Linear bias, lm_head, embed_tokens, and the layer norms.

gpt-oss stores each layer's 32 experts as fused 3D tensors, which a targets=["Linear"] recipe cannot see. Loading inside load_context() together with the gptoss_linear_experts shim exposes them as per-expert Linear submodules, so the quantizer reaches all 2,304 expert projections. The router is skipped because it is a tiny Linear whose logits decide expert assignment, where an 8-bit rounding error can flip the top-4 selection and change which experts run.

import gptoss_linear_experts  # noqa: F401  (registers GptOssLinearExperts)
from transformers import AutoModelForCausalLM, AutoTokenizer

from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.utils import load_context

MODEL_ID = "unsloth/gpt-oss-20b-BF16"
SAVE_DIR = "./gpt-oss-20b-BF16-w8a8-llmcompressor-v0.13.0"

IGNORE = [
    "lm_head",
    r"re:.*\.router$",
    r"re:.*\.router\..*",
    r"re:.*\.gate$",
    r"re:.*\.mlp\.gate$",
]

# Step 1: Load the BF16 model inside load_context(), which linearizes the fused
# 3D expert tensors into per-expert Linear submodules.
print(f"loading {MODEL_ID}", flush=True)
with load_context():
    model = AutoModelForCausalLM.from_pretrained(MODEL_ID, dtype="bfloat16")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

experts = model.model.layers[0].mlp.experts
print("linearized experts class:", type(experts).__name__, flush=True)

# Step 2: Define the W8A8 recipe and apply it. W8A8 here is data-free (RTN),
# so no calibration dataset is needed.
recipe = QuantizationModifier(targets=["Linear"], scheme="W8A8", ignore=IGNORE)
oneshot(model=model, recipe=recipe)

# Step 3: Save in compressed-tensors format.
print(f"saving to {SAVE_DIR}", flush=True)
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)
print("done", flush=True)

Quick Start

Use with vLLM

from vllm import LLM, SamplingParams

model = LLM(
    model="amd/gpt-oss-20b-BF16-w8a8-llmcompressor-v0.13.0",
    dtype="bfloat16",
)

sampling_params = SamplingParams(temperature=0.7, max_tokens=256)
outputs = model.generate(["Hello, how are you?"], sampling_params)
print(outputs[0].outputs[0].text)

Requirements

torch==2.13.0.0
zentorch==2.13.0.0
vllm==0.28.0
llmcompressor==0.13.0

OpenMP Setup

For optimal performance, set LD_PRELOAD with libomp.so (LLVM OpenMP) or libiomp5.so (Intel OpenMP):

# Using LLVM OpenMP (llvmopenmp)
export LD_PRELOAD=$(find /path/to/env -name "libomp.so" | head -1)

# Or using Intel OpenMP (libiomp)
export LD_PRELOAD=$(find /path/to/env -name "libiomp5.so" | head -1)

Note: Set LD_PRELOAD before launching vLLM or any inference script.

Evaluation

The model was evaluated against the BF16 (unquantized) baseline on standard benchmarks using lm-evaluation-harness with the vLLM engine.

Benchmark	BF16 Baseline	W8A8 (this model)	Recovery
GSM8K (5-shot)	0.8969	0.8605	95.94%

Evaluation Command

lm_eval \
    --model vllm \
    --model_args pretrained=amd/gpt-oss-20b-BF16-w8a8-llmcompressor-v0.13.0,dtype=bfloat16,max_model_len=4096 \
    --tasks gsm8k \
    --batch_size auto \
    --trust_remote_code \
    --num_fewshot 5 \
    --apply_chat_template \
    --log_samples \
    --gen_kwargs "max_gen_toks=2048" \
    --output_path .

Limitations

Version Lock: This model is compatible with ZenDNN v6.1.0 / ZenTorch v2.13.0.0 / PyTorch v2.13.0.0. It may not load correctly on other versions.
CPU Only: This model is optimized for AMD EPYC CPU inference via ZenDNN. It is not intended for GPU inference.
Accuracy Trade-off: The quantization pass is data-free, so per-channel scales are derived from the weights alone with no activation statistics to compensate outliers. On a 32-expert MoE where each expert sees only a fraction of the tokens, this costs about 3.6 points of GSM8K accuracy (95.94% recovery), more than a calibrated pass would.

License

This model is distributed under the same license as the source model. See the LICENSE file for details.

Downloads last month: 178

Safetensors

Model size

21B params

Tensor type

I32

BF16

Model tree for amd/gpt-oss-20b-BF16-w4a16-llmcompressor

Base model

openai/gpt-oss-20b

Finetuned

unsloth/gpt-oss-20b-BF16

Quantized

(14)

this model