speculative_decoding.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Speculative Decoding\n",
    "\n",
    "SGLang now provides an EAGLE2-based speculative decoding option. Our implementation aims to maximize speed and efficiency and is considered to be among the fastest in open-source LLM engines.\n",
    "\n",
    "**Note:** Currently, Speculative Decoding in SGLang does not support radix cache.\n",
    "\n",
    "### Performance Highlights\n",
    "\n",
    "- Official EAGLE code ([SafeAILab/EAGLE](https://github.com/SafeAILab/EAGLE)): ~200 tokens/s\n",
    "- Standard SGLang Decoding: ~156 tokens/s\n",
    "- EAGLE Decoding in SGLang: ~297 tokens/s\n",
    "- EAGLE Decoding in SGLang (w/ `torch.compile`): ~316 tokens/s\n",
    "\n",
    "All benchmarks below were run on a single H100."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## EAGLE Decoding\n",
    "\n",
    "To enable EAGLE speculative decoding the following parameters are relevant:\n",
    "* `speculative_draft_model_path`: Specifies draft model. This parameter is required.\n",
    "* `speculative_num_steps`: Depth of autoregressive drafting. Increases speculation range but risks rejection cascades. Default is 5.\n",
    "\n",
    "* `speculative_eagle_topk`: Branching factor per step. Improves candidate diversity, will lead to higher acceptance rate, but more lead to higher memory/compute consumption. Default is 4.\n",
    "\n",
    "* `speculative_num_draft_tokens`: Maximum parallel verification capacity. Allows deeper tree evaluation but will lead to higher GPU memory usage. Default is 8."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sglang.test.test_utils import is_in_ci\n",
    "\n",
    "if is_in_ci():\n",
    "    from patch import launch_server_cmd\n",
    "else:\n",
    "    from sglang.utils import launch_server_cmd\n",
    "\n",
    "from sglang.utils import wait_for_server, print_highlight, terminate_process\n",
    "\n",
    "server_process, port = launch_server_cmd(\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model meta-llama/Llama-2-7b-chat-hf  --speculative-algorithm EAGLE \\\n",
    "    --speculative-draft-model-path lmsys/sglang-EAGLE-llama2-chat-7B --speculative-num-steps 3 \\\n",
    "    --speculative-eagle-topk 4 --speculative-num-draft-tokens 16 --cuda-graph-max-bs 8\n",
    "\"\"\"\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai\n",
    "\n",
    "client = openai.Client(base_url=f\"http://127.0.0.1:{port}/v1\", api_key=\"None\")\n",
    "\n",
    "response = client.chat.completions.create(\n",
    "    model=\"meta-llama/Meta-Llama-3.1-8B-Instruct\",\n",
    "    messages=[\n",
    "        {\"role\": \"user\", \"content\": \"List 3 countries and their capitals.\"},\n",
    "    ],\n",
    "    temperature=0,\n",
    "    max_tokens=64,\n",
    ")\n",
    "\n",
    "print_highlight(f\"Response: {response}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "terminate_process(server_process)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### EAGLE Decoding with `torch.compile`\n",
    "\n",
    "You can also enable `torch.compile` for further optimizations and optionally set `--cuda-graph-max-bs`:\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "server_process, port = launch_server_cmd(\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model meta-llama/Llama-2-7b-chat-hf  --speculative-algorithm EAGLE \\\n",
    "    --speculative-draft-model-path lmsys/sglang-EAGLE-llama2-chat-7B --speculative-num-steps 5 \\\n",
    "        --speculative-eagle-topk 8 --speculative-num-draft-tokens 64 --mem-fraction 0.6 \\\n",
    "            --enable-torch-compile --cuda-graph-max-bs 2\n",
    "\"\"\"\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai\n",
    "\n",
    "client = openai.Client(base_url=f\"http://127.0.0.1:{port}/v1\", api_key=\"None\")\n",
    "\n",
    "response = client.chat.completions.create(\n",
    "    model=\"meta-llama/Meta-Llama-3.1-8B-Instruct\",\n",
    "    messages=[\n",
    "        {\"role\": \"user\", \"content\": \"List 3 countries and their capitals.\"},\n",
    "    ],\n",
    "    temperature=0,\n",
    "    max_tokens=64,\n",
    ")\n",
    "\n",
    "print_highlight(f\"Response: {response}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "terminate_process(server_process)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## References\n",
    "\n",
    "EAGLE process is as follows:\n",
    "\n",
    "- Within EAGLE the draft model predicts the next feature vector, i.e. the last hidden state of the original LLM, using the feature sequence $(f_1, ..., f_k)$ and the token sequence $(t_2, ..., t_{k+1})$. \n",
    "- The next token is then sampled from $p_{k+2}=\\text{LMHead}(f_{k+1})$. Afterwards, the two sequences are extended in a tree style—branching out multiple potential continuations, with the branching factor per step controlled by the `speculative_eagle_topk` parameter—to ensure a more coherent connection of context, and are given as input again.\n",
    "- EAGLE-2 additionally uses the draft model to evaluate how probable certain branches in the draft tree are, dynamically stopping the expansion of unlikely branches. After the expansion phase, reranking is employed to select only the top `speculative_num_draft_tokens` final nodes as draft tokens.\n",
    "\n",
    "This enhances drafting accuracy by operating on the features instead of tokens for more regular inputs and passing the tokens from the next timestep additionaly to minimize randomness effects from sampling. Furthermore the dynamic adjustment of the draft tree and selection of reranked final nodes increases acceptance rate of draft tokens further. For more details see [the paper](https://arxiv.org/abs/2406.16858)."
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}