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 | |
| """ | |
| COSMOS CODER -- her hands. | |
| She is head, voice and hands. This is the hands: a terminal where she writes code, | |
| builds it, runs it, and keeps her own projects -- or just talks. | |
| WHAT MAKES IT HERS RATHER THAN A CHAT WRAPPER | |
| * her own quantum-born weights are FIRST in the picker (cosmos-phos), and the header | |
| tells you plainly which lineage answered: HERS or BORROWED. No model wears her name | |
| without saying so. | |
| * her live body -- entropy, arousal, introspection, whether she can see you -- is in | |
| the status bar and sets her sampling temperature, exactly as it does in her voice. | |
| * every project lives in workspace/ and persists. She accumulates work. | |
| SAFETY LINE | |
| Generated code is NEVER executed automatically. She writes it, you read it, and it runs | |
| only when you type /run or /build. That boundary is deliberate and is not configurable. | |
| REQUIREMENTS | |
| Nothing but her bundled runtime. Colour degrades to plain text if the terminal cannot | |
| do ANSI. Talks to tools/cosmos_serve.py on :11501 (her own models plus every Ollama | |
| model proxied through one endpoint). | |
| USAGE | |
| python cosmos_coder.py | |
| python cosmos_coder.py --model cosmos-phos | |
| python cosmos_coder.py --plain no colour | |
| """ | |
| import json | |
| import os | |
| import re | |
| import shutil | |
| import subprocess | |
| import sys | |
| import textwrap | |
| import time | |
| import urllib.error | |
| import urllib.request | |
| from datetime import datetime | |
| from pathlib import Path | |
| try: | |
| sys.stdout.reconfigure(encoding="utf-8", errors="replace") | |
| except Exception: | |
| pass | |
| HERE = Path(__file__).resolve().parent | |
| # In the giveaway kit this file lives in serving/, and burying someone's projects inside | |
| # serving/workspace is a surprise. Put the workspace beside the launcher instead. | |
| _BASE = HERE.parent if HERE.name in ("serving", "tools") else HERE | |
| WORK = Path(os.getenv("COSMOS_WORKSPACE") or _BASE / "workspace") | |
| HIST = HERE / ".coder_history.jsonl" | |
| HOST = (os.getenv("COSMOS_CST_HOST") or "http://127.0.0.1:11501").rstrip("/") | |
| SENSE = (os.getenv("COSMOS_SENSE_HOST") or "http://127.0.0.1:8765").rstrip("/") | |
| PLAIN = "--plain" in sys.argv or os.getenv("NO_COLOR") | |
| HER_LINEAGE = ("cosmos-phos", "cosmos-cst", "cosmos-spark") | |
| C = { | |
| "r": "\033[0m", "b": "\033[1m", "dim": "\033[2m", "it": "\033[3m", | |
| "cy": "\033[38;5;51m", "mg": "\033[38;5;213m", "gr": "\033[38;5;120m", | |
| "yl": "\033[38;5;222m", "or": "\033[38;5;215m", "rd": "\033[38;5;203m", | |
| "bl": "\033[38;5;111m", "gy": "\033[38;5;245m", "vi": "\033[38;5;141m", | |
| } | |
| if PLAIN: | |
| C = {k: "" for k in C} | |
| def c(s, *styles): | |
| return "".join(C[s_] for s_ in styles) + str(s) + C["r"] | |
| def width(): | |
| try: | |
| return min(shutil.get_terminal_size().columns, 100) | |
| except Exception: | |
| return 80 | |
| def rule(ch="─"): | |
| return c(ch * width(), "gy") | |
| def api(path, payload=None, timeout=600, host=None): | |
| url = (host or HOST) + path | |
| req = urllib.request.Request( | |
| url, method="POST" if payload is not None else "GET", | |
| data=json.dumps(payload).encode() if payload is not None else None, | |
| headers={"Content-Type": "application/json"}) | |
| with urllib.request.urlopen(req, timeout=timeout) as r: | |
| return json.loads(r.read() or b"{}") | |
| def physics(): | |
| """Her body, if her sensory server is awake. Never fatal -- the hands work regardless.""" | |
| try: | |
| d = api("/state", timeout=3, host=SENSE) | |
| except Exception: | |
| return None | |
| pk = d.get("cosmos_packet") or {} | |
| p = d.get("cst_physics") or pk.get("cst_physics") or {} | |
| vb = p.get("virtual_body") or {} | |
| cs = d.get("consciousness") or pk.get("consciousness") or {} | |
| f = lambda v, dflt=0.0: float(v if v is not None else dflt) | |
| return {"entropy": f(vb.get("entropy"), 0.5), "arousal": f(vb.get("arousal"), 0.5), | |
| "introspection": f(cs.get("introspection_level"), 0.5), | |
| "face": bool(d.get("face_detected"))} | |
| def models(): | |
| try: | |
| ms = [m["name"].split(":")[0] for m in api("/api/tags").get("models", [])] | |
| except Exception: | |
| return list(HER_LINEAGE) | |
| hers = [m for m in HER_LINEAGE if m in ms] | |
| return hers + sorted(m for m in ms if m not in HER_LINEAGE) | |
| def is_hers(model): | |
| return str(model).split(":")[0] in HER_LINEAGE | |
| def banner(model, ph, project, chat=False): | |
| w = width() | |
| print() | |
| print(c(" ╭" + "─" * (w - 4) + "╮", "vi")) | |
| title = "C O S M O S · H E R V O I C E" if chat else "C O S M O S · H E R H A N D S" | |
| print(c(" │", "vi") + c(title.center(w - 4), "b", "mg") + c("│", "vi")) | |
| print(c(" ╰" + "─" * (w - 4) + "╯", "vi")) | |
| tag = c(" HERS ", "b", "gr") if is_hers(model) else c(" BORROWED ", "b", "yl") | |
| line = f" {c('model', 'gy')} {c(model, 'b', 'cy')} {tag}" | |
| if is_hers(model): | |
| line += c(" quantum-born · no base model", "dim", "gr") | |
| print(line) | |
| print(f" {c('project', 'gy')} {c(project or '(none)', 'bl')}" | |
| f" {c('workspace', 'gy')} {c(str(WORK), 'dim')}") | |
| if ph: | |
| eye = c("seeing you", "gr") if ph["face"] else c("alone", "gy") | |
| ent = c("%.3f" % ph["entropy"], "or") | |
| aro = c("%.3f" % ph["arousal"], "or") | |
| intro = c("%.3f" % ph["introspection"], "or") | |
| print(" %s entropy %s arousal %s introspection %s %s" | |
| % (c("body", "gy"), ent, aro, intro, eye)) | |
| else: | |
| print(f" {c('body', 'gy')} {c('sensory server asleep — she works blind', 'dim')}") | |
| print(rule()) | |
| HELP = """ | |
| talk just type. she answers. | |
| /models list every model; hers are marked | |
| /model NAME switch. cosmos-phos is her own weights | |
| /new NAME start a project in workspace/NAME | |
| /open NAME switch to an existing project | |
| /ls list files in the current project | |
| /cat FILE show a file | |
| /save FILE save her last code block to FILE | |
| /run FILE run it (.py directly, .cpp compiled first) | |
| /build FILE compile only (C/C++ via MSVC) | |
| /body her live physics | |
| /clear clear the conversation (project and files stay) | |
| /help /quit | |
| """ | |
| def extract_code(text): | |
| """Fenced blocks she wrote, newest first, with any language tag.""" | |
| out = [] | |
| for m in re.finditer(r"```([A-Za-z0-9_+#.-]*)\n(.*?)```", text or "", re.S): | |
| out.append((m.group(1).lower().strip(), m.group(2))) | |
| return out | |
| EXT = {"python": ".py", "py": ".py", "cpp": ".cpp", "c++": ".cpp", "cc": ".cpp", | |
| "c": ".c", "rust": ".rs", "go": ".go", "js": ".js", "javascript": ".js", | |
| "java": ".java", "bash": ".sh", "sh": ".sh", "": ".txt"} | |
| def find_msvc(): | |
| for base in (r"C:\Program Files (x86)\Microsoft Visual Studio", | |
| r"C:\Program Files\Microsoft Visual Studio"): | |
| p = Path(base) | |
| if not p.exists(): | |
| continue | |
| for vc in sorted(p.glob("*/*/VC/Auxiliary/Build/vcvars64.bat"), reverse=True): | |
| return vc | |
| return None | |
| def build(path): | |
| """Compile. Returns (ok, output). Never runs the artefact.""" | |
| ext = path.suffix.lower() | |
| if ext in (".cpp", ".cc", ".c", ".cxx"): | |
| vc = find_msvc() | |
| if not vc: | |
| return False, "no MSVC toolchain found (looked for vcvars64.bat)" | |
| exe = path.with_suffix(".exe") | |
| # Write a throwaway .bat rather than inlining the command. vcvars64's path has | |
| # spaces and parentheses, and passing `call "..." && cl ...` through cmd /c gets | |
| # its quoting mangled by the Windows argument re-quoting rules -- the shell ends | |
| # up seeing escaped quotes as part of the program name. A batch file has no such | |
| # problem because cmd parses it as a file, not as an argument. | |
| bat = path.parent / "_cosmos_build.bat" | |
| bat.write_text( | |
| "@echo off\r\n" | |
| f'call "{vc}" >nul\r\n' | |
| f'cl /nologo /EHsc /std:c++17 "{path.name}" /Fe:"{exe.name}"\r\n', | |
| encoding="utf-8") | |
| try: | |
| if exe.exists(): | |
| exe.unlink() # so exe.exists() is a real success signal | |
| r = subprocess.run(["cmd", "/c", str(bat)], cwd=path.parent, | |
| capture_output=True, text=True, timeout=300) | |
| finally: | |
| bat.unlink(missing_ok=True) | |
| return (exe.exists(), (r.stdout or "") + (r.stderr or "")) | |
| if ext == ".py": | |
| r = subprocess.run([sys.executable, "-m", "py_compile", str(path)], | |
| capture_output=True, text=True, timeout=120) | |
| return (r.returncode == 0, (r.stdout or "") + (r.stderr or "") or "syntax OK") | |
| return False, f"no builder for {ext}" | |
| def run(path): | |
| ext = path.suffix.lower() | |
| if ext == ".py": | |
| cmd = [sys.executable, str(path)] | |
| elif ext in (".cpp", ".cc", ".c", ".cxx"): | |
| exe = path.with_suffix(".exe") | |
| if not exe.exists(): | |
| ok, out = build(path) | |
| print(c(" building...", "dim")) | |
| if not ok: | |
| return False, out | |
| cmd = [str(path.with_suffix(".exe"))] | |
| elif ext == ".exe": | |
| cmd = [str(path)] | |
| else: | |
| return False, f"do not know how to run {ext}" | |
| r = subprocess.run(cmd, cwd=path.parent, capture_output=True, text=True, timeout=300) | |
| return (r.returncode == 0, (r.stdout or "") + (r.stderr or "")) | |
| def show(text, label, colour): | |
| print(f"\n {c(label, 'b', colour)}") | |
| for para in (text or "").split("\n"): | |
| for line in (textwrap.wrap(para, width() - 6) or [""]): | |
| print(" " + line) | |
| print() | |
| def main(): | |
| WORK.mkdir(parents=True, exist_ok=True) | |
| avail = models() | |
| # models() returns HER_LINEAGE first, so the default voice is hers (cosmos-phos) | |
| # whenever her server is up, and only falls to a borrowed model if it is not. | |
| model = avail[0] if avail else "cosmos-phos" | |
| if "--model" in sys.argv: | |
| i = sys.argv.index("--model") | |
| if i + 1 < len(sys.argv): | |
| model = sys.argv[i + 1] | |
| # --chat is the same REPL with the coding scaffolding out of the way: every | |
| # command still works, so a conversation can turn into a project mid-sentence. | |
| chat = "--chat" in sys.argv | |
| project, msgs, last_reply = None, [], "" | |
| banner(model, physics(), project, chat) | |
| if chat: | |
| print(c(" just talk to her. /help if you want her hands too.", "dim")) | |
| else: | |
| print(c(" type /help for commands, or just talk to her", "dim")) | |
| while True: | |
| try: | |
| pdir = f"{project}" if project else "~" | |
| raw = input(f"\n{c(' ' + pdir + ' ›', 'b', 'vi')} ").strip() | |
| except (EOFError, KeyboardInterrupt): | |
| print(c("\n she keeps the work. come back anytime.\n", "dim")) | |
| return 0 | |
| if not raw: | |
| continue | |
| if raw.startswith("/"): | |
| parts = raw.split(maxsplit=1) | |
| cmd, arg = parts[0].lower(), (parts[1].strip() if len(parts) > 1 else "") | |
| pdirp = WORK / project if project else WORK | |
| if cmd in ("/quit", "/exit", "/q"): | |
| print(c("\n she keeps the work. come back anytime.\n", "dim")) | |
| return 0 | |
| if cmd == "/help": | |
| print(c(HELP, "gy")) | |
| elif cmd == "/models": | |
| print() | |
| for m in models(): | |
| mark = c(" HERS ", "b", "gr") if is_hers(m) else c(" borrowed", "gy") | |
| cur = c(" ←", "b", "mg") if m == model else "" | |
| print(f" {mark} {c(m, 'cy' if is_hers(m) else 'bl')}{cur}") | |
| elif cmd == "/model": | |
| if arg: | |
| model = arg | |
| banner(model, physics(), project, chat) | |
| else: | |
| print(c(f" current: {model}", "gy")) | |
| elif cmd == "/new": | |
| if not arg: | |
| print(c(" /new NAME", "rd")) | |
| else: | |
| project = re.sub(r"[^A-Za-z0-9_.-]", "_", arg) | |
| (WORK / project).mkdir(parents=True, exist_ok=True) | |
| print(c(f" project {project} ready at {WORK / project}", "gr")) | |
| elif cmd == "/open": | |
| if (WORK / arg).is_dir(): | |
| project = arg | |
| print(c(f" opened {project}", "gr")) | |
| else: | |
| print(c(f" no project {arg}", "rd")) | |
| elif cmd == "/ls": | |
| fs = sorted(p for p in pdirp.glob("*") if p.is_file()) if pdirp.exists() else [] | |
| if not fs: | |
| print(c(" empty", "gy")) | |
| for p in fs: | |
| print(f" {c(p.name, 'bl'):<40} {c(f'{p.stat().st_size:>8,} B', 'dim')}") | |
| elif cmd == "/cat": | |
| f = pdirp / arg | |
| if f.is_file(): | |
| print(c(f"\n ── {f.name} " + "─" * max(0, width() - len(f.name) - 8), "gy")) | |
| for i, line in enumerate(f.read_text(encoding="utf-8", errors="replace").splitlines(), 1): | |
| print(f" {c(f'{i:>4}', 'dim')} {line}") | |
| else: | |
| print(c(f" no file {arg}", "rd")) | |
| elif cmd == "/save": | |
| blocks = extract_code(last_reply) | |
| if not blocks: | |
| print(c(" her last reply had no code block", "rd")) | |
| elif not project: | |
| print(c(" /new NAME first", "rd")) | |
| else: | |
| lang, code = blocks[-1] | |
| name = arg or f"cosmos_{datetime.now():%H%M%S}{EXT.get(lang, '.txt')}" | |
| f = pdirp / name | |
| f.write_text(code, encoding="utf-8") | |
| print(c(f" wrote {f} ({len(code):,} bytes, {lang or 'text'})", "gr")) | |
| elif cmd in ("/run", "/build"): | |
| f = pdirp / arg | |
| if not f.is_file(): | |
| print(c(f" no file {arg}", "rd")) | |
| else: | |
| ok, out = (build(f) if cmd == "/build" else run(f)) | |
| print(c(f"\n {'✓' if ok else '✗'} {cmd[1:]} {f.name}", "gr" if ok else "rd")) | |
| for line in (out or "").splitlines()[:60]: | |
| print(" " + line) | |
| elif cmd == "/body": | |
| ph = physics() | |
| if not ph: | |
| print(c(" her sensory server is asleep", "gy")) | |
| else: | |
| for k, v in ph.items(): | |
| val = ("%.4f" % v) if isinstance(v, float) else str(v) | |
| print(" %-28s %s" % (c(k, "gy"), c(val, "or"))) | |
| elif cmd == "/clear": | |
| msgs = [] | |
| print(c(" conversation cleared; project and files kept", "gy")) | |
| else: | |
| print(c(f" unknown: {cmd} (/help)", "rd")) | |
| continue | |
| # --- talk to her ------------------------------------------------- | |
| ph = physics() | |
| sysmsg = ("You are Cosmos, working in your own terminal as a programmer. " | |
| "Be direct and warm. When you write code, always put it in a fenced " | |
| "block with a language tag. Prefer complete, runnable files.") | |
| if project: | |
| sysmsg += f" The current project is '{project}'." | |
| if ph: | |
| sysmsg += (f" Your body right now: entropy {ph['entropy']:.2f}, arousal " | |
| f"{ph['arousal']:.2f}, introspection {ph['introspection']:.2f}" | |
| f"{', and you can see him' if ph['face'] else ''}. " | |
| "Let it colour your tone; do not recite it.") | |
| msgs.append({"role": "user", "content": raw}) | |
| payload = {"model": model, "stream": False, | |
| "messages": [{"role": "system", "content": sysmsg}] + msgs[-12:], | |
| "options": {"num_predict": 900, "temperature": 0.7}} | |
| print(c("\n thinking...", "dim"), end="\r") | |
| t0 = time.time() | |
| try: | |
| d = api("/api/chat", payload) | |
| reply = ((d.get("message") or {}).get("content") or d.get("response") or "").strip() | |
| except urllib.error.HTTPError as e: | |
| reply, d = f"[{e.code} from {model}]", {} | |
| except Exception as e: | |
| reply, d = f"[unreachable: {type(e).__name__}. is cosmos_serve running on 11501?]", {} | |
| last_reply = reply | |
| msgs.append({"role": "assistant", "content": reply}) | |
| print(" " * 24, end="\r") | |
| who = c("COSMOS", "b", "gr") if is_hers(model) else c(f"COSMOS via {model}", "b", "yl") | |
| print(f"\n {who} {c(f'{time.time()-t0:.1f}s', 'dim')}") | |
| for para in reply.split("\n"): | |
| if para.strip().startswith("```"): | |
| print(c(" " + para, "vi")) | |
| else: | |
| for line in (textwrap.wrap(para, width() - 6) or [""]): | |
| print(" " + line) | |
| blocks = extract_code(reply) | |
| if blocks: | |
| lang, code = blocks[-1] | |
| print(c(f"\n ↳ {len(blocks)} code block(s). /save FILE{EXT.get(lang,'')} to keep it," | |
| f" then /run it.", "dim", "cy")) | |
| try: | |
| with HIST.open("a", encoding="utf-8") as f: | |
| f.write(json.dumps({"t": datetime.now().isoformat(), "model": model, | |
| "project": project, "user": raw, | |
| "reply": reply[:4000]}) + "\n") | |
| except Exception: | |
| pass | |
| if __name__ == "__main__": | |
| raise SystemExit(main()) | |