Mixtral-8x7B-Instruct-v0.1-w4a16-llmcompressor-v0.12.0

Model Overview

• Model Architecture: MixtralForCausalLM
  • Input: Text
  • Output: Text
• Source Model: Mixtral-8x7B-Instruct-v0.1
• Supported Hardware: AMD EPYC (CPU inference)
• Preferred Operating System: Linux
• Inference Engine: vLLM v0.26.0
• Quantization Framework: LLM Compressor v0.12.0
• Quantization Method: 4-bit Weight-Only Quantization (W4A16)
• Compatible Stack:
  • ZenDNN v6.1.0
  • ZenTorch v2.11.0.3
  • PyTorch v2.11.0
  • LLM Compressor v0.12.0
  • vLLM v0.26.0

This is a quantized version of Mixtral-8x7B-Instruct-v0.1 created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference.

Quantization

The model was quantized from Mixtral-8x7B-Instruct-v0.1 using LLM Compressor via the GPTQ algorithm. This reduces the model weights from 87.0 GiB to 22.8 GiB on disk (~74% reduction).

• Method: 4-bit Weight-Only Quantization (W4A16)
• Config: compressed-tensors, num_bits=4, type=int, symmetric=true, group_size=128
• Weights: INT4, symmetric, group-wise (group_size=128, actorder=static), stored as pack-quantized
• Activations: BF16 (unquantized)
• Group Size: 128
• Kept in BF16: MoE router (block_sparse_moe.gate) and lm_head
• Calibration: 128 examples from HuggingFaceH4/ultrachat_200k at a max sequence length of 2048

Both the attention projections and all eight per-expert gate_proj / up_proj / down_proj matrices are quantized. Only the tiny routing layer stays in BF16, which is why the on-disk reduction (~74%) is larger than for dense models: the experts dominate the parameter count.

Quantization command:

numactl --physcpubind=0-95 python llm_compressor_quantize_and_run.py \
    --model_id mistralai/Mixtral-8x7B-Instruct-v0.1 \
    --save_dir ./Mixtral-8x7B-Instruct-v0.1-w4a16-llmcompressor-v0.12.0 \
    --recipe gptq \
    --scheme W4A16 \
    --run

Equivalent minimal steps:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from datasets import load_dataset

from llmcompressor import oneshot
from llmcompressor.modifiers.quantization.gptq import GPTQModifier

model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
output_dir = "./Mixtral-8x7B-Instruct-v0.1-w4a16-llmcompressor-v0.12.0"

NUM_CALIBRATION_SAMPLES = 128
MAX_SEQUENCE_LENGTH = 2048

# Step 1: Load the BF16 model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
    device_map="cpu",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

# Step 2: Load calibration data. GPTQ is data-driven: it needs real activations
# to build the per-layer Hessians used to compensate the rounding error.
ds = load_dataset(
    "HuggingFaceH4/ultrachat_200k",
    split=f"train_sft[:{NUM_CALIBRATION_SAMPLES}]",
)
ds = ds.map(
    lambda example: {"text": "\n".join(m["content"] for m in example["messages"])},
    remove_columns=ds.column_names,
)

# Step 3: Define the W4A16 recipe. The preset implies group_size=128, which is
# valid here because Mixtral's 4096 and 14336 column counts are both divisible
# by 128. The router is skipped: it is tiny and wrong routing wrecks accuracy.
recipe = GPTQModifier(
    targets="Linear",
    scheme="W4A16",
    ignore=[
        "lm_head",
        r"re:.*\.router$",
        r"re:.*\.router\..*",
        r"re:.*\.gate$",
        r"re:.*\.mlp\.gate$",
    ],
)

# Step 4: One-shot quantize with calibration and save in compressed-tensors format
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    processor=tokenizer,
)
model.save_pretrained(output_dir, save_compressed=True)
tokenizer.save_pretrained(output_dir)

# Smoke test
inputs = tokenizer("What are we having for dinner?", return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=30)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Quick Start

Use with vLLM

from vllm import LLM, SamplingParams

model = LLM(
    model="amd/Mixtral-8x7B-Instruct-v0.1-w4a16-llmcompressor-v0.12.0",
    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==2.11.0
zentorch==2.11.0.3
vllm==0.26.0
llmcompressor==0.12.0

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.

Evaluation Command

lm_eval \
    --model vllm \
    --model_args pretrained=amd/Mixtral-8x7B-Instruct-v0.1-w4a16-llmcompressor-v0.12.0,dtype=bfloat16 \
    --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 v2.11.0.3 / PyTorch v2.11.0. 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: 4-bit weight-only quantization is more aggressive than INT8. On GSM8K the model retains 98.83% of the BF16 baseline for a ~74% smaller memory footprint.

License

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

Modifications copyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved.