Text Generation
PyTorch
GGUF
English
quantum
quantum-entropy
from-scratch
char-level
cosmic-synapse-theory
custom-architecture
llama-cpp
continual-learning
reproducible-seed
open-science
null-results
Instructions to use phera-ra/QC67_cosmo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use phera-ra/QC67_cosmo with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf phera-ra/QC67_cosmo # Run inference directly in the terminal: llama cli -hf phera-ra/QC67_cosmo
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf phera-ra/QC67_cosmo # Run inference directly in the terminal: llama cli -hf phera-ra/QC67_cosmo
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf phera-ra/QC67_cosmo # Run inference directly in the terminal: ./llama-cli -hf phera-ra/QC67_cosmo
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf phera-ra/QC67_cosmo # Run inference directly in the terminal: ./build/bin/llama-cli -hf phera-ra/QC67_cosmo
Use Docker
docker model run hf.co/phera-ra/QC67_cosmo
- LM Studio
- Jan
- vLLM
How to use phera-ra/QC67_cosmo with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "phera-ra/QC67_cosmo" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "phera-ra/QC67_cosmo", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/phera-ra/QC67_cosmo
- Ollama
How to use phera-ra/QC67_cosmo with Ollama:
ollama run hf.co/phera-ra/QC67_cosmo
- Unsloth Studio
How to use phera-ra/QC67_cosmo with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for phera-ra/QC67_cosmo to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for phera-ra/QC67_cosmo to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for phera-ra/QC67_cosmo to start chatting
- Docker Model Runner
How to use phera-ra/QC67_cosmo with Docker Model Runner:
docker model run hf.co/phera-ra/QC67_cosmo
- Lemonade
How to use phera-ra/QC67_cosmo with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull phera-ra/QC67_cosmo
Run and chat with the model
lemonade run user.QC67_cosmo-{{QUANT_TAG}}List all available models
lemonade list
- Atomic Chat
File size: 8,722 Bytes
d6da243 b8fadbf d6da243 b8fadbf d6da243 b085020 d6da243 b085020 b8fadbf d6da243 b085020 d6da243 b8fadbf d6da243 b8fadbf b085020 b8fadbf d6da243 b8fadbf d6da243 b085020 d6da243 b085020 d6da243 b8fadbf d6da243 b8fadbf d6da243 b8fadbf d6da243 b8fadbf d6da243 b8fadbf | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 | #!/usr/bin/env python3
"""
QUANTUM ENGINE VERIFICATION β against quantum mechanics, not against itself.
The physics engine was checked against published Lorenz constants. Same standard here.
PART A the PUBLISHED ARCHIVE β free, offline, no QPU time
A1 conservation: counts must sum to the declared shot total
A2 Born-rule sanity: measured marginals in [0,1], no impossible outcomes
A3 THE WEIGHT-BIRTH MECHANISM. cosmos_born.pt was made by mapping measured
bitstrings to uniforms u = int(bits)/2^n, then to weights through the inverse
normal CDF z = sqrt(2)*erfinv(2u-1). If that pipeline is correct, z over real
archived shots must come out standard normal: mean 0, sd 1, and the right
tail fractions. This is checked on her ACTUAL archive, so it verifies the
mechanism that literally created her weights.
A4 serial independence is explicitly NOT inferred from histogram counts;
the archive does not retain within-job shot order.
PART B LIVE HARDWARE β the CHSH Bell test
A local hidden-variable theory cannot exceed |S| = 2. Quantum mechanics permits
up to 2*sqrt(2) = 2.8284 (Tsirelson's bound). Running CHSH on the backend her
engine actually uses is the one measurement that cannot be reproduced by any
classical process, however clever. If S > 2 on her hardware, the quantum path is
genuinely quantum. If S <= 2, everything downstream is classical randomness with
a receipt.
Cost: one job, four circuits.
"""
import math
import os
import re
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent))
sys.stdout.reconfigure(encoding="utf-8", errors="replace")
# Ingestion lives in archive_io so the loader can be benchmarked on its own
# (`python benchmarks/archive_io.py --bench`) rather than only through this verifier.
# A3 used to expand every archived shot into its own Python float -- 2.4 million of them,
# ~79 MB -- purely to hand a list to statistics.pstdev. The archive is already a
# histogram, so weighted moments give the identical mean, sd and tail fractions from the
# (bitstring, count) pairs directly: measured 22.7x faster, 5.0x less memory, and equal
# to the old path to 1.1e-16.
from archive_io import (JSON_BACKEND, born_z_moments, is_labeled_hardware, # noqa: E402
load_records)
ROOT = Path(__file__).resolve().parents[1]
ARCHIVE = Path(
os.getenv(
"COSMOS_QUANTUM_ARCHIVE",
str(ROOT / "data" / "quantum_measurements_public.jsonl"),
)
)
R = []
def chk(name, ok, detail):
R.append((name, ok))
print(f" [{'PASS' if ok else 'FAIL'}] {name}\n {detail}\n")
def load_archive():
return load_records(ARCHIVE)
print("=" * 80)
print(" PART A β PUBLISHED QUANTUM ARCHIVE (no QPU time)")
print("=" * 80 + "\n")
runs = load_archive()
hardware_runs = [record for record in runs if is_labeled_hardware(record)]
legacy_runs = [
record for record in runs
if str(record.get("provider_class") or "") == "legacy_unlabelled"
]
simulator_runs = [
record for record in runs
if str(record.get("provider_class") or "") == "classical_simulator"
]
print(f" archive: {ARCHIVE}")
print(f" json backend: {JSON_BACKEND}")
print(f" loaded {len(runs):,} records with measurement counts")
print(f" labeled IBM hardware: {len(hardware_runs):,}")
print(f" legacy unlabelled: {len(legacy_runs):,}")
print(f" classical simulator: {len(simulator_runs):,}\n")
# A1 conservation
bad, checked, tot_shots = 0, 0, 0
for d in runs:
s = sum(d["counts"].values())
tot_shots += s
dec = d.get("total_shots")
if isinstance(dec, int) and dec > 0:
checked += 1
if s != dec:
bad += 1
chk("shot conservation (counts sum to declared total)", bad == 0,
f"{checked:,} records carried a declared total; {bad} mismatched; "
f"{tot_shots:,} samples archived across all provider classes")
# A2 Born-rule sanity
neg = sum(1 for d in runs for v in d["counts"].values() if v < 0)
widths = {len(k) for d in runs for k in d["counts"]}
chk("no negative counts; consistent register width", neg == 0 and len(widths) <= 3,
f"negative counts {neg}; bitstring widths present {sorted(widths)}")
# A3 THE WEIGHT-BIRTH MECHANISM on explicitly labeled IBM hardware shots.
# Weighted moments over the histogram; the shot cap of 40 that stops one high-shot job
# from dominating is part of the estimator and is preserved exactly.
mom = born_z_moments(hardware_runs, cap=40)
m, sd = mom.mean, mom.sd
frac1, frac2 = mom.tail(1.0), mom.tail(2.0)
ok = abs(m) < 0.05 and abs(sd - 1.0) < 0.06 and abs(frac1 - 0.6827) < 0.03 and abs(frac2 - 0.9545) < 0.02
chk("weight-birth pipeline yields a standard normal", ok,
f"n={mom.n:,.0f} mean {m:+.4f} (want 0) sd {sd:.4f} (want 1) "
f"|z|<=1 {frac1:.4f} (want 0.6827) |z|<=2 {frac2:.4f} (want 0.9545)")
# A4 cannot be measured from histograms. Iterating count-dictionary keys creates
# an arbitrary outcome ordering, not the original sequence of hardware shots.
R.append(("serial independence (shot order not retained)", None))
print(
" [SKIP] serial independence\n"
" archive records are histograms; within-job shot order was not retained\n"
)
# ββ PART B β CHSH on real hardware βββββββββββββββββββββββββββββββββββββββββ
print("=" * 80)
print(" PART B β CHSH BELL TEST ON HER LIVE HARDWARE")
print("=" * 80 + "\n")
def token():
# Public release boundary: credentials come only from the current process
# environment. The verifier never searches local files or vault paths.
return (
os.getenv("IBM_QUANTUM_TOKEN")
or os.getenv("QISKIT_IBM_TOKEN")
or None
)
def run_chsh():
from qiskit import QuantumCircuit
from qiskit_ibm_runtime import QiskitRuntimeService, SamplerV2
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager
tok = token()
if not tok:
return None, "no IBM token"
svc = QiskitRuntimeService(channel="ibm_quantum_platform", token=tok)
be = svc.least_busy(operational=True, simulator=False)
def bell(theta_a, theta_b):
qc = QuantumCircuit(2, 2)
qc.h(0)
qc.cx(0, 1) # |Phi+>
qc.ry(-2 * theta_a, 0) # rotate measurement basis
qc.ry(-2 * theta_b, 1)
qc.measure([0, 1], [0, 1])
return qc
# the standard CHSH angles that reach Tsirelson's bound
a, ap = 0.0, math.pi / 4
b, bp = math.pi / 8, 3 * math.pi / 8
settings = [("AB", a, b), ("AB'", a, bp), ("A'B", ap, b), ("A'B'", ap, bp)]
pm = generate_preset_pass_manager(backend=be, optimization_level=2, seed_transpiler=5)
circs = [pm.run(bell(x, y)) for _, x, y in settings]
job = SamplerV2(mode=be).run(circs, shots=4096)
jid = job.job_id() if callable(getattr(job, "job_id", None)) else "?"
res = job.result()
E = {}
for i, (name, _, _) in enumerate(settings):
data = res[i].data
counts = (data.c if hasattr(data, "c") else data.meas).get_counts()
tot = sum(counts.values())
same = sum(v for k, v in counts.items() if k.count("1") % 2 == 0)
E[name] = (2 * same - tot) / tot
S = E["AB"] - E["AB'"] + E["A'B"] + E["A'B'"]
return {"backend": be.name, "job": jid, "E": E, "S": S}, None
try:
out, err = run_chsh()
except Exception as e:
out, err = None, f"{type(e).__name__}: {e}"
if out:
print(f" backend {out['backend']} job {out['job']} 4096 shots x 4 settings\n")
for k, v in out["E"].items():
print(f" E({k:<4s}) = {v:+.4f}")
S = out["S"]
print(f"\n S = E(AB) - E(AB') + E(A'B) + E(A'B') = {S:+.4f}")
print(f" classical limit 2.0000")
print(f" Tsirelson (quantum max) 2.8284\n")
viol = abs(S) > 2.0
chk("CHSH violates the classical bound", viol,
f"|S| = {abs(S):.4f} vs classical 2.0000 β "
f"{'no local hidden-variable theory can produce this' if viol else 'within classical reach'}")
else:
R.append(("CHSH on hardware", None))
print(f" [SKIP] {err} (offline archive checks remain valid)\n")
print("=" * 80)
passed = sum(1 for _, value in R if value is True)
failed = sum(1 for _, value in R if value is False)
skipped = sum(1 for _, value in R if value is None)
print(f" {passed} PASSED / {failed} FAILED / {skipped} SKIPPED")
for name, value in R:
status = "PASS" if value is True else ("FAIL" if value is False else "SKIP")
print(f" {status} {name}")
print("=" * 80)
sys.exit(0 if failed == 0 else 1)
|