Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,228 +1,91 @@
|
|
1 |
#!/usr/bin/env python3
|
2 |
-
|
3 |
-
what_comes_next.py – Hugging Face Space implementation of **What Comes Next**
|
4 |
-
A global, slow-burn guessing game powered by Llama-3.1-8B-Instruct (FP32, CPU-only).
|
5 |
-
|
6 |
-
HOW IT WORKS
|
7 |
-
============
|
8 |
-
• One shared model generates a single, very long completion (≈2 k tokens) for a chosen
|
9 |
-
prompt in *full precision* on CPU. One token is sampled every ~15 s, so a prompt
|
10 |
-
unfolds for roughly 10 hours. All visitors see the same progress in real-time.
|
11 |
-
• Players read the partial output and may submit **either**
|
12 |
-
🧠 Exact continuation (full guess) **or** 💡 General idea (summary guess).
|
13 |
-
• Each guess is appended to `data.json` with prompt, Oracle progress, timestamp & type.
|
14 |
-
• Offline scoring (not included here) can later measure similarity vs the final text.
|
15 |
-
"""
|
16 |
-
|
17 |
-
from __future__ import annotations
|
18 |
-
|
19 |
-
import os
|
20 |
-
import json
|
21 |
-
import time
|
22 |
-
import random
|
23 |
-
import threading
|
24 |
-
import logging
|
25 |
-
from datetime import datetime, timezone
|
26 |
-
from typing import Dict, Any
|
27 |
|
28 |
-
import
|
29 |
-
import
|
|
|
30 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
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 |
-
return prompts
|
77 |
-
|
78 |
-
###############################################################################
|
79 |
-
# Model loading #
|
80 |
-
###############################################################################
|
81 |
-
log.info("Loading Llama-3.1-8B-Instruct (FP32 CPU-only) using secret token…")
|
82 |
-
|
83 |
-
tokenizer = AutoTokenizer.from_pretrained(
|
84 |
-
MODEL_NAME,
|
85 |
-
use_auth_token=os.environ.get("HF_READ_TOKEN")
|
86 |
-
)
|
87 |
-
model = AutoModelForCausalLM.from_pretrained(
|
88 |
-
MODEL_NAME,
|
89 |
-
torch_dtype=torch.float32,
|
90 |
-
device_map={"": "cpu"},
|
91 |
-
use_auth_token=os.environ.get("HF_READ_TOKEN")
|
92 |
-
)
|
93 |
-
model.eval()
|
94 |
-
log.info("Model ready – Oracle awakened.")
|
95 |
-
|
96 |
-
###############################################################################
|
97 |
-
# Global state #
|
98 |
-
###############################################################################
|
99 |
-
lock = threading.Lock() # guard state + files
|
100 |
-
prompts = load_prompts() # list of 100 strings
|
101 |
-
|
102 |
-
###############################################################################
|
103 |
-
# Oracle generation thread #
|
104 |
-
###############################################################################
|
105 |
-
|
106 |
-
def _init_state() -> Dict[str, Any]:
|
107 |
-
"""Return existing state or create a fresh one if none/finished."""
|
108 |
-
state = _read_json(STATE_PATH, {})
|
109 |
-
if not state or state.get("finished"):
|
110 |
-
prompt_idx = random.randrange(len(prompts))
|
111 |
-
state = {
|
112 |
-
"prompt_idx": prompt_idx,
|
113 |
-
"prompt": prompts[prompt_idx],
|
114 |
-
"generated": "", # text so far
|
115 |
-
"tokens_done": 0,
|
116 |
-
"start_time": time.time(),
|
117 |
-
"finished": False
|
118 |
-
}
|
119 |
-
_atomic_write(STATE_PATH, state)
|
120 |
-
log.info(f"New Oracle prompt #{prompt_idx}: {state['prompt'][:80]}…")
|
121 |
-
return state
|
122 |
-
|
123 |
-
|
124 |
-
def _elapsed_str(start: float) -> str:
|
125 |
-
d = int(time.time() - start)
|
126 |
-
h, r = divmod(d, 3600)
|
127 |
-
m, s = divmod(r, 60)
|
128 |
-
return f"{h}h {m}m {s}s"
|
129 |
-
|
130 |
|
131 |
def oracle_loop():
|
132 |
while True:
|
|
|
|
|
|
|
|
|
|
|
|
|
133 |
with lock:
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
continue
|
138 |
-
|
139 |
-
# Build context: prompt + generated so far
|
140 |
-
context = state["prompt"] + state["generated"]
|
141 |
-
input_ids = tokenizer(context, return_tensors="pt", truncation=True, max_length=MAX_CONTEXT_TOKENS).input_ids
|
142 |
-
|
143 |
-
# Sample one token
|
144 |
-
with torch.no_grad():
|
145 |
-
out = model.generate(
|
146 |
-
input_ids,
|
147 |
-
max_new_tokens=1,
|
148 |
-
do_sample=True,
|
149 |
-
temperature=TEMPERATURE,
|
150 |
-
top_p=TOP_P,
|
151 |
-
)
|
152 |
-
next_token = tokenizer.decode(out[0, -1], skip_special_tokens=True, clean_up_tokenization_spaces=False)
|
153 |
-
|
154 |
-
with lock:
|
155 |
-
state["generated"] += next_token
|
156 |
-
state["tokens_done"] += 1
|
157 |
-
if state["tokens_done"] >= TOKENS_PER_PROMPT:
|
158 |
-
state["finished"] = True
|
159 |
-
log.info("Prompt completed – Oracle will select a new one shortly.")
|
160 |
-
_atomic_write(STATE_PATH, state)
|
161 |
time.sleep(SECS_BETWEEN_TOKENS)
|
162 |
-
|
163 |
threading.Thread(target=oracle_loop, daemon=True).start()
|
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 |
-
"prompt": prompt,
|
189 |
-
"point-in-time": elapsed,
|
190 |
-
"response-point": generated,
|
191 |
-
"user-guess": guess_text,
|
192 |
-
"guess-type": guess_type
|
193 |
-
}
|
194 |
-
with lock:
|
195 |
-
with open(DATA_PATH, "a", encoding="utf-8") as f:
|
196 |
-
f.write(json.dumps(record, ensure_ascii=False) + "\n")
|
197 |
-
log.info(f"Logged {guess_type} guess ({len(guess_text)} chars).")
|
198 |
-
return gr.update(value="✅ Guess recorded – thanks!"), gr.update(value=""), gr.update(value="")
|
199 |
-
|
200 |
-
|
201 |
-
with gr.Blocks(title="What Comes Next", theme="gradio/soft") as demo:
|
202 |
-
gr.Markdown("""# 🌌 What Comes Next
|
203 |
-
Watch the Oracle craft an extended response – **one token at a time**. Predict its
|
204 |
-
next words or general direction and see how close you were when the tale concludes.
|
205 |
-
(All inputs are stored in `data.json` for research.)""")
|
206 |
-
|
207 |
-
prompt_md = gr.Markdown()
|
208 |
-
oracle_box = gr.Textbox(lines=10, interactive=False, label="📜 Oracle text so far")
|
209 |
-
elapsed_tb = gr.Textbox(interactive=False, label="⏱ Elapsed time")
|
210 |
-
|
211 |
-
refresh_btn = gr.Button("🔄 Refresh")
|
212 |
-
|
213 |
-
with gr.Row():
|
214 |
-
exact_tb = gr.Textbox(label="🧠 Exact continuation (full)")
|
215 |
-
idea_tb = gr.Textbox(label="💡 General idea")
|
216 |
-
submit_btn = gr.Button("Submit Guess")
|
217 |
-
status_tb = gr.Textbox(interactive=False, label="Status")
|
218 |
-
|
219 |
-
# Actions
|
220 |
-
refresh_btn.click(fetch_state, outputs=[prompt_md, oracle_box, elapsed_tb])
|
221 |
-
demo.load(fetch_state, outputs=[prompt_md, oracle_box, elapsed_tb])
|
222 |
-
|
223 |
-
submit_btn.click(submit_guess,
|
224 |
-
inputs=[exact_tb, idea_tb],
|
225 |
-
outputs=[status_tb, exact_tb, idea_tb])
|
226 |
-
|
227 |
-
if __name__ == "__main__":
|
228 |
-
demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)
|
|
|
1 |
#!/usr/bin/env python3
|
2 |
+
# what comes next sloppy version
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
+
import os, json, time, random, threading, logging
|
5 |
+
from datetime import datetime, timezone
|
6 |
+
import torch, gradio as gr
|
7 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
8 |
|
9 |
+
MODEL_NAME = "meta-llama/Llama-3.1-8B-Instruct"
|
10 |
+
PROMPTS_PATH = "full_prompts.json"
|
11 |
+
STATE_PATH = "current_state.json"
|
12 |
+
DATA_PATH = "data.json"
|
13 |
+
TOKENS_PER_PROMPT = 2048
|
14 |
+
SECS_BETWEEN_TOKENS = 15
|
15 |
+
TEMPERATURE = 0.9
|
16 |
+
TOP_P = 0.95
|
17 |
+
MAX_CONTEXT_TOKENS = 8192
|
18 |
+
|
19 |
+
logging.basicConfig(level=logging.INFO)
|
20 |
+
log = logging.getLogger()
|
21 |
+
|
22 |
+
def _read_json(p, d):
|
23 |
+
try: return json.load(open(p, encoding="utf-8"))
|
24 |
+
except: return d
|
25 |
+
|
26 |
+
def _atomic_write(p, o):
|
27 |
+
t = p + ".tmp"; open(t, "w", encoding="utf-8").write(json.dumps(o, ensure_ascii=False, indent=2)); os.replace(t,p)
|
28 |
+
|
29 |
+
def load_prompts():
|
30 |
+
l = _read_json(PROMPTS_PATH, [])
|
31 |
+
if not l: raise FileNotFoundError
|
32 |
+
return l
|
33 |
+
|
34 |
+
# load model (uses HF_READ_TOKEN)
|
35 |
+
tok = os.environ.get("HF_READ_TOKEN")
|
36 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=tok)
|
37 |
+
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.float32, low_cpu_mem_usage=True, token=tok)
|
38 |
+
model.to(torch.device("cpu")); model.eval()
|
39 |
+
|
40 |
+
prompts = load_prompts(); lock = threading.Lock()
|
41 |
+
|
42 |
+
# main loop: oracle gen
|
43 |
+
def _init_state():
|
44 |
+
s = _read_json(STATE_PATH, {})
|
45 |
+
if not s or s.get("finished"):
|
46 |
+
i = random.randrange(len(prompts))
|
47 |
+
s = {"prompt_idx":i, "prompt":prompts[i], "generated":"", "tokens_done":0, "start_time":time.time(), "finished":False}
|
48 |
+
_atomic_write(STATE_PATH, s)
|
49 |
+
return s
|
50 |
+
|
51 |
+
def _elapsed_str(st):
|
52 |
+
d=int(time.time()-st);h,r=divmod(d,3600);m,s=divmod(r,60);return f"{h}h {m}m {s}s"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
def oracle_loop():
|
55 |
while True:
|
56 |
+
with lock: s=_init_state()
|
57 |
+
if s["finished"]: time.sleep(SECS_BETWEEN_TOKENS); continue
|
58 |
+
c=s["prompt"]+s["generated"]
|
59 |
+
ids=tokenizer(c, return_tensors="pt", truncation=True, max_length=MAX_CONTEXT_TOKENS).input_ids
|
60 |
+
with torch.no_grad(): out=model.generate(ids, max_new_tokens=1, do_sample=True, temperature=TEMPERATURE, top_p=TOP_P)
|
61 |
+
nt=tokenizer.decode(out[0,-1], skip_special_tokens=True, clean_up_tokenization_spaces=False)
|
62 |
with lock:
|
63 |
+
s["generated"]+=nt; s["tokens_done"]+=1
|
64 |
+
if s["tokens_done"]>=TOKENS_PER_PROMPT: s["finished"]=True
|
65 |
+
_atomic_write(STATE_PATH, s)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
time.sleep(SECS_BETWEEN_TOKENS)
|
|
|
67 |
threading.Thread(target=oracle_loop, daemon=True).start()
|
68 |
|
69 |
+
# ui
|
70 |
+
|
71 |
+
def fetch_state():
|
72 |
+
s=_read_json(STATE_PATH,{})
|
73 |
+
if not s: return "Loading...","","0h 0m 0s"
|
74 |
+
return s["prompt"], s["generated"], _elapsed_str(s["start_time"])
|
75 |
+
|
76 |
+
def submit_guess(full, idea):
|
77 |
+
f=full.strip(); i=idea.strip()
|
78 |
+
if not (f or i): return gr.update(value="enter guess!"),gr.update(),gr.update()
|
79 |
+
p,g,e=fetch_state(); guess=f or i; gt="full" if f else "idea"
|
80 |
+
r={"timestamp":datetime.now(timezone.utc).isoformat(),"prompt":p,"point-in-time":e,"response-point":g,"user-guess":guess,"guess-type":gt}
|
81 |
+
with lock: open(DATA_PATH,"a",encoding="utf-8").write(json.dumps(r,ensure_ascii=False)+"\n")
|
82 |
+
return gr.update(value="logged!"),gr.update(value=""),gr.update(value="")
|
83 |
+
|
84 |
+
with gr.Blocks(title="What Comes Next") as demo:
|
85 |
+
gr.Markdown("# What Comes Next - sloppy")
|
86 |
+
prm=gr.Markdown(); txt=gr.Textbox(lines=10,interactive=False,label="oracle"); elt=gr.Textbox(interactive=False,label="time")
|
87 |
+
r=gr.Button("refresh"); f=gr.Textbox(label="full guess"); i=gr.Textbox(label="idea"); sbtn=gr.Button("send"); st=gr.Textbox(interactive=False,label="st")
|
88 |
+
demo.load(fetch_state,outputs=[prm,txt,elt])
|
89 |
+
r.click(fetch_state,outputs=[prm,txt,elt]); sbtn.click(submit_guess,inputs=[f,i],outputs=[st,f,i])
|
90 |
+
|
91 |
+
if __name__=="__main__": demo.launch(server_name="0.0.0.0",server_port=7860)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|