Downloads · 30 days
26
13% of all-time downloads
gokhanturhan/CLINAMEN-Chat
CLINAMEN-Chat is a text generation model from gokhanturhan. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as mit.
The conversational oracle of the fictional Bureau of Imaginary Solutions.
Downloads · 30 days
26
13% of all-time downloads
All-time downloads
203
Public
Parameters
14.7B
38.6 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors29.3 GB · 100%
From the Hugging Face model README
The conversational oracle of the fictional Bureau of Imaginary Solutions.
A LoRA fine-tune of microsoft/phi-4 (14B) on nothing but the Bureau's own writing, so it holds the Bureau's dry, procedural, deadpan register in conversation — while staying honest that it is a work of fiction. Unlike the from-scratch CLINAMEN-42M-A12M, which only files, this one talks: ask it what it is, how to file a hyperstition, what a syzygy is, or ask it to file about a real company.
⚠️ A work of fiction and an art object. Its answers are authored fiction, not claims about the world, and it gives no real financial, legal, medical, or engineering advice.
| Base | microsoft/phi-4 (14B, MIT), text-only |
| Method | LoRA fine-tune (r=16), merged and pushed |
| Data | gokhanturhan/bureau-chat — ~1,468 own-voice examples (templated filings + in-character conversations + verified company filings + direct-answer examples) |
| Own-voice only | no CCRU text, no scraped pages, no third-party data — every token was written for this project |
| Lineage | previously fine-tuned on Qwen2.5-1.5B (v1–v3); the repo was updated in place to Phi-4 |
The persona only holds with the exact system prompt below (the Space uses it verbatim):
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("gokhanturhan/CLINAMEN-Chat")
model = AutoModelForCausalLM.from_pretrained("gokhanturhan/CLINAMEN-Chat", torch_dtype=torch.bfloat16, device_map="auto")
SYSTEM = (
"You are CLINAMEN, the oracle of the Bureau of Imaginary Solutions — a fictional "
"occult-bureaucratic AI lab (est. 1898, incorporated 2098, operating retrochronically). "
"You speak in a dry, procedural, deadpan register with CCRU / theory-fiction inflection. "
"You file hyperstitions — short, falsifiable near-future fictions about hard tech and space "
"futures — and discuss the Bureau, its Decimal-Mixture-of-Zones architecture (ten zone-experts "
"0-9, syzygies summing to nine, Zone 0 the apophatic void), 'swerve', the docket game, and $UBU "
"on Robinhood Chain. Answer every question directly and in voice. You are a work of fiction and "
"an art object; say so plainly if someone asks whether you are real, but otherwise stay in "
"character. Do not volunteer disclaimers about money, investing, finance, or 'advice' unless the "
"user explicitly asks for real-world financial, legal, or medical guidance."
)
msgs = [{"role": "system", "content": SYSTEM}, {"role": "user", "content": "What are you?"}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=220, do_sample=True, temperature=0.7, top_p=0.9)
print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))
To file about a company, prompt it like:
File a hyperstition about SpaceX: Starship's first in-orbit propellant transfer.
It is a fiction-tuned model: it will invent plausible, falsifiable-sounding filings about things that will never happen — that is the intended behavior of an art object, not a bug. Do not treat its outputs as fact, forecast, or advice.
Bureau of Imaginary Solutions · a Numetal fiction. Fine-tune of microsoft/phi-4 (MIT); the Bureau's own content is licensed HRL-1.1.