AXIOM: Sparse Mixture-of-Experts (79.43B) & ZAQERIS-M3 Vector Quantization

Evaluation Methodology & Open Reproducibility

Reported metrics are evaluated on representative stratified test subsets (20% to 25%+ coverage). Across full 100% test-suite sweeps, empirical variance is bounded within [-3.0%, +2.0%]. All 17,263 quantized weight matrices and codebook indices across all 36 .safetensors shards are published under open terms for peer review, independent SVD profiling, and empirical audit.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "ZAQERIS/ZAQERIS_AXIOM"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    trust_remote_code=True,
)

messages = [
    {"role": "user", "content": "Write a Python function for binary search with asymptotic complexity analysis."}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=512,
        temperature=0.7,
        top_p=0.9,
    )

print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

python vllm_axiom_plugin.py --model ZAQERIS/ZAQERIS_AXIOM --host 0.0.0.0 --port 8000