Spaces:
Sleeping
Sleeping
File size: 5,807 Bytes
67367c9 |
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 |
import logging
from threading import Thread
from typing import Generator, Dict, Any, List
import re
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
if torch.cuda.is_available():
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True
logger = logging.getLogger("plutus.model")
logging.basicConfig(level=logging.INFO)
MODEL_NAME = "Remostart/Plutus_Advanced_model"
class SharedLLM:
_tokenizer = None
_model = None
_device = "cuda" if torch.cuda.is_available() else "cpu"
@classmethod
def load(cls):
if cls._model is not None:
return cls._tokenizer, cls._model, cls._device
logger.info(f"[LOAD] Loading tokenizer: {MODEL_NAME}")
cls._tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=True)
logger.info(f"[LOAD] Loading model on {cls._device}")
cls._model = AutoModelForCausalLM.from_pretrained(
MODEL_NAME,
torch_dtype=torch.float16 if cls._device == "cuda" else None,
low_cpu_mem_usage=True
)
cls._model.to(cls._device)
cls._model.eval()
logger.info("[READY] Shared LLM loaded once.")
return cls._tokenizer, cls._model, cls._device
_SENTENCE_END_RE = re.compile(r"([.!?])\s+$")
_LIST_ITEM_RE = re.compile(r"^\s*(\d+\.|\-|\*)\s+$")
_CODE_FENCE = "```"
def should_flush(buffer: str) -> bool:
stripped = buffer.strip()
if len(stripped) < 25:
return False
if _LIST_ITEM_RE.match(stripped):
return False
if "\n\n" in buffer:
return True
if _SENTENCE_END_RE.search(buffer):
return True
if len(buffer) > 180:
return True
return False
class PlutusModel:
def __init__(self):
self.tokenizer, self.model, self.device = SharedLLM.load()
def create_prompt(self, personality: str, level: str, topic: str, extra_context: str = None) -> str:
prompt = (
"You are PlutusTutor — the best expert in Cardano's Plutus ecosystem.\n\n"
f"User Info:\n"
f"- Personality: {personality}\n"
f"- Level: {level}\n"
f"- Topic: {topic}\n\n"
"Your task:\n"
"- Teach with extreme clarity.\n"
"- Use structured explanations.\n"
"- Include examples where helpful.\n"
"- Avoid filler.\n"
"- Adapt tone to personality.\n\n"
)
if extra_context:
prompt += f"Additional Context:\n{extra_context}\n\n"
return prompt + "Begin teaching now.\n\nAssistant:"
def generate(
self,
prompt: str,
max_new_tokens: int = 600,
temperature: float = 0.6,
top_p: float = 0.9
) -> Generator[str, None, None]:
inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)
streamer = TextIteratorStreamer(self.tokenizer, skip_prompt=True, skip_special_tokens=True)
def _run():
with torch.inference_mode():
self.model.generate(
**inputs,
streamer=streamer,
max_new_tokens=max_new_tokens,
do_sample=True,
temperature=temperature,
top_p=top_p,
eos_token_id=self.tokenizer.eos_token_id,
pad_token_id=self.tokenizer.pad_token_id,
)
Thread(target=_run, daemon=True).start()
buffer = ""
in_code_block = False
for token in streamer:
buffer += token
if _CODE_FENCE in buffer:
in_code_block = not in_code_block
if not in_code_block and should_flush(buffer):
yield buffer.strip()
buffer = ""
if buffer.strip():
yield buffer.strip()
class SummaryModel:
def __init__(self):
self.tokenizer, self.model, self.device = SharedLLM.load()
def summarize_text(
self,
full_teaching: str,
topic: str,
level: str,
recommended: List[Dict[str, Any]],
max_new_tokens: int = 400
) -> Generator[str, None, None]:
prompt = (
"You are a world-class summarization assistant.\n\n"
f"TOPIC: {topic}\n"
f"LEVEL: {level}\n\n"
"CONTENT:\n"
f"{full_teaching}\n\n"
"Produce a clear, structured summary.\n\n"
"Assistant:"
)
inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)
streamer = TextIteratorStreamer(self.tokenizer, skip_prompt=True, skip_special_tokens=True)
def _run():
with torch.inference_mode():
self.model.generate(
**inputs,
streamer=streamer,
max_new_tokens=max_new_tokens,
do_sample=True,
temperature=0.6,
top_p=0.9,
eos_token_id=self.tokenizer.eos_token_id,
)
Thread(target=_run, daemon=True).start()
buffer = ""
in_code_block = False
for token in streamer:
buffer += token
if _CODE_FENCE in buffer:
in_code_block = not in_code_block
if not in_code_block and should_flush(buffer):
yield buffer.strip()
buffer = ""
if buffer.strip():
yield buffer.strip()
if recommended:
yield "\n\n### Recommended Resources\n"
for item in recommended:
line = f"- **{item['type'].upper()}**: {item.get('url')}"
yield line
|