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
| #!/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) | |