SIA/sia/llm_engine/qwq_llm_engine.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.\n",
      "Unsloth: Failed to patch Gemma3ForConditionalGeneration.\n",
      "🦥 Unsloth Zoo will now patch everything to make training faster!\n",
      "INFO 04-23 16:36:10 [__init__.py:239] Automatically detected platform cuda.\n"
     ]
    }
   ],
   "source": [
    "# Unsloth should be imported before transformers to ensure all optimizations are applied.\n",
    "from unsloth import FastLanguageModel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "from threading import Thread\n",
    "from transformers import AutoTokenizer, TextIteratorStreamer, pipeline\n",
    "from xml_schema_validator import XmlLogitsProcessor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "temperature = 0.6\n",
    "model_path = \"/root/models/notebook_merged_4bit\"\n",
    "xml_schema_text = Path(\"/root/sia/action_schema.xsd\").read_text()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load tokenizer\n",
    "tokenizer = AutoTokenizer.from_pretrained(\n",
    "    model_path,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==((====))==  Unsloth 2025.3.19: Fast Qwen2 patching. Transformers: 4.51.3. vLLM: 0.8.2.\n",
      "   \\\\   /|    NVIDIA RTX 6000 Ada Generation. Num GPUs = 1. Max memory: 47.5 GB. Platform: Linux.\n",
      "O^O/ \\_/ \\    Torch: 2.6.0+cu124. CUDA: 8.9. CUDA Toolkit: 12.4. Triton: 3.2.0\n",
      "\\        /    Bfloat16 = TRUE. FA [Xformers = 0.0.29.post2. FA2 = False]\n",
      " \"-____-\"     Free license: http://github.com/unslothai/unsloth\n",
      "Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Sliding Window Attention is enabled but not implemented for `eager`; unexpected results may be encountered.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "validate_env\n",
      "device_map sequential\n",
      "validate_env\n",
      "device_map OrderedDict([('', 0)])\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a99acfd42fa645abae77144125a734b7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/4 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Load model\n",
    "model, _returned_tokenizer = FastLanguageModel.from_pretrained(\n",
    "    model_path,\n",
    "    gpu_memory_utilization = 0.8, # Reduce if out of memory\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Device set to use cuda:0\n"
     ]
    }
   ],
   "source": [
    "# Create inference pipeline with memory-efficient settings\n",
    "pipeline = pipeline(\n",
    "    \"text-generation\",\n",
    "    model=model,\n",
    "    tokenizer=tokenizer,\n",
    "    return_full_text=False,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "logits_processor = XmlLogitsProcessor(tokenizer, xml_schema_text)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "messages = [\n",
    "    {\"role\": \"system\", \"content\": \"Always respond in a <write_stdout></write_stdout> xml block.\"},\n",
    "    {\"role\": \"user\", \"content\": \"Hi, how are you?\"},\n",
    "    {\"role\": \"assistant\", \"content\": \"\"},\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "text = tokenizer.apply_chat_template(\n",
    "    messages,\n",
    "    tokenize=False,\n",
    "    add_generation_prompt=False,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "streamer = TextIteratorStreamer(\n",
    "    tokenizer,\n",
    "    skip_prompt=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "generation_kwargs = {\n",
    "    \"text_inputs\": text,\n",
    "    \"do_sample\": True,\n",
    "    \"temperature\": temperature,\n",
    "    \"streamer\": streamer,\n",
    "    \"use_cache\": True,\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "generation_kwargs[\"logits_processor\"] = [logits_processor.copy()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "generation_thread = Thread(\n",
    "    target=pipeline,\n",
    "    kwargs=generation_kwargs\n",
    ")\n",
    "\n",
    "generation_thread.start()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<write_stdout>\n",
      "Hello! I'm just a large language model, so I don't have feelings, but thank you for asking. How can I assist you today?</write_stdout><|im_end|>"
     ]
    }
   ],
   "source": [
    "for text in streamer:\n",
    "    print(text, end=\"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "generation_thread.join()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "sia",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}