Skip to content

Instantly share code, notes, and snippets.

@alonsosilvaallende

alonsosilvaallende/notebook.ipynb

Last active February 18, 2026 16:13
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save alonsosilvaallende/36f2c7e9720dc8b8225dae082f58d046 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save alonsosilvaallende/36f2c7e9720dc8b8225dae082f58d046 to your computer and use it in GitHub Desktop.
Download ZIP
from_outlines_core
Display the source blob
Display the rendered blob
Raw
notebook.ipynb
{
"cells": [
{
"cell_type": "markdown",
"id": "093b1977-dc46-4508-b109-17ea5315877b",
"metadata": {},
"source": [
"We already have the vocabulary from a model with the script you gave me so we don't need this cell:"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "ec12bffc-60dd-48e3-8661-0110d61109c3",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:55:12.360538Z",
"iopub.status.busy": "2026-02-18T15:55:12.360027Z",
"iopub.status.idle": "2026-02-18T15:55:12.807528Z",
"shell.execute_reply": "2026-02-18T15:55:12.806416Z",
"shell.execute_reply.started": "2026-02-18T15:55:12.360490Z"
}
},
"outputs": [],
"source": [
"import outlines_core\n",
"\n",
"model_id = \"Qwen/Qwen3-0.6B\"\n",
"vocabulary = outlines_core.Vocabulary.from_pretrained(model_id)"
]
},
{
"cell_type": "markdown",
"id": "22d884e2-c189-4e0b-923d-5b1fce9282c1",
"metadata": {},
"source": [
"This is all we need from `outlines_core`:"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "afa6ee68-3f32-49fb-875f-a281a33e074c",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:55:30.593216Z",
"iopub.status.busy": "2026-02-18T15:55:30.592715Z",
"iopub.status.idle": "2026-02-18T15:55:30.642589Z",
"shell.execute_reply": "2026-02-18T15:55:30.641631Z",
"shell.execute_reply.started": "2026-02-18T15:55:30.593170Z"
}
},
"outputs": [
{
"data": {
"text/plain": [
"Index {\n",
" initial_state: 64,\n",
" final_states: {\n",
" 112,\n",
" },\n",
" transitions: {\n",
" 112: {\n",
" 151645: 112,\n",
" },\n",
" 64: {\n",
" 54: 128,\n",
" 1639: 80,\n",
" 10234: 96,\n",
" },\n",
" 80: {\n",
" 88: 96,\n",
" },\n",
" 128: {\n",
" 8503: 96,\n",
" 71: 80,\n",
" },\n",
" 96: {\n",
" 220: 112,\n",
" },\n",
" },\n",
" eos_token_id: 151645,\n",
" vocab_size: 151656,\n",
"}"
]
},
"execution_count": 2,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"index = outlines_core.Index(\"Why \", vocabulary)\n",
"index"
]
},
{
"cell_type": "markdown",
"id": "50ebabed-c20f-4fea-867f-d5cea195b3ad",
"metadata": {},
"source": [
"In reality, we only need the transitions (I can deduce the other info):"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "d744f7b5-6e58-434b-bdef-e19668d1d05b",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:56:14.351757Z",
"iopub.status.busy": "2026-02-18T15:56:14.351001Z",
"iopub.status.idle": "2026-02-18T15:56:14.360753Z",
"shell.execute_reply": "2026-02-18T15:56:14.358920Z",
"shell.execute_reply.started": "2026-02-18T15:56:14.351691Z"
}
},
"outputs": [
{
"data": {
"text/plain": [
"{112: {151645: 112},\n",
" 64: {54: 128, 1639: 80, 10234: 96},\n",
" 80: {88: 96},\n",
" 128: {8503: 96, 71: 80},\n",
" 96: {220: 112}}"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"index.get_transitions()"
]
},
{
"cell_type": "markdown",
"id": "8ade5c02-d253-402f-b508-0388e9808c59",
"metadata": {},
"source": [
"The transitions consist of {state: {token: next_state}}."
]
},
{
"cell_type": "markdown",
"id": "b5044b68-bd66-4e8e-8571-e15269049d12",
"metadata": {},
"source": [
"## This is just to make the transitions 'prettier'"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "fb075839-f549-4fb6-9819-4731428513fc",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:58:13.649553Z",
"iopub.status.busy": "2026-02-18T15:58:13.649093Z",
"iopub.status.idle": "2026-02-18T15:58:13.658346Z",
"shell.execute_reply": "2026-02-18T15:58:13.656443Z",
"shell.execute_reply.started": "2026-02-18T15:58:13.649512Z"
}
},
"outputs": [],
"source": [
"def my_recursive(\n",
" state: int,\n",
" index: outlines_core.Index,\n",
" mapping: dict[int, int],\n",
" visited: set[int],\n",
" final_states: set[int],\n",
") -> None:\n",
" if state in final_states:\n",
" return\n",
" visited.add(state)\n",
" for symbol, new_state in index.get_transitions().get(state, {}).items():\n",
" if new_state in final_states:\n",
" continue # Skip final states entirely\n",
" if new_state not in mapping:\n",
" mapping[new_state] = len(mapping)\n",
" if new_state not in visited:\n",
" my_recursive(new_state, index, mapping, visited, final_states)"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "8dadfe8d-e228-4ebb-b7f6-cbcdd293dc95",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:58:19.521604Z",
"iopub.status.busy": "2026-02-18T15:58:19.521152Z",
"iopub.status.idle": "2026-02-18T15:58:19.529955Z",
"shell.execute_reply": "2026-02-18T15:58:19.528246Z",
"shell.execute_reply.started": "2026-02-18T15:58:19.521562Z"
}
},
"outputs": [],
"source": [
"def get_state_mapping(index: outlines_core.Index) -> dict[int, int]:\n",
" initial_state = index.get_initial_state()\n",
" final_states = index.get_final_states()\n",
" num_states = len(index.get_transitions().keys())\n",
" mapping = {}\n",
" # Start from initial state (mapped to 0)\n",
" mapping[initial_state] = 0\n",
" visited = set()\n",
" my_recursive(initial_state, index, mapping, visited, final_states)\n",
" # End with final states (mapped at the end)\n",
" for i, final_state in enumerate(final_states):\n",
" mapping[final_state] = num_states - (i + 1)\n",
" return mapping"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "92f4f768-9ab1-42a0-8b6f-19555abf7ded",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:58:25.853038Z",
"iopub.status.busy": "2026-02-18T15:58:25.852579Z",
"iopub.status.idle": "2026-02-18T15:58:25.861045Z",
"shell.execute_reply": "2026-02-18T15:58:25.859315Z",
"shell.execute_reply.started": "2026-02-18T15:58:25.852996Z"
}
},
"outputs": [],
"source": [
"def get_dfa(index: outlines_core.Index) -> dict[int, dict[int, int]]:\n",
" mapping = get_state_mapping(index)\n",
" dfa = {}\n",
" for state, transitions in index.get_transitions().items():\n",
" new_transitions = {}\n",
" for token, new_state in transitions.items():\n",
" new_transitions[token] = mapping[new_state]\n",
" if state not in index.get_final_states():\n",
" dfa[mapping[state]] = new_transitions\n",
" return dfa"
]
},
{
"cell_type": "markdown",
"id": "b3afd24b-fd37-43fa-b1cd-0266de321e1d",
"metadata": {},
"source": [
"I consider this:"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "3b08fbb1-ca69-4c7f-8870-ae3744355ec3",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:58:28.185285Z",
"iopub.status.busy": "2026-02-18T15:58:28.184789Z",
"iopub.status.idle": "2026-02-18T15:58:28.192492Z",
"shell.execute_reply": "2026-02-18T15:58:28.190267Z",
"shell.execute_reply.started": "2026-02-18T15:58:28.185242Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{0: {54: 1, 1639: 3, 10234: 2}, 3: {88: 2}, 1: {8503: 2, 71: 3}, 2: {220: 4}}\n"
]
}
],
"source": [
"print(get_dfa(index))"
]
},
{
"cell_type": "markdown",
"id": "568145ca-3197-47da-b8dc-bdcf910808cb",
"metadata": {},
"source": [
"prettier than this:"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "1aba4457-29c1-4cb0-935d-283db83e0449",
"metadata": {
"execution": {
"iopub.execute_input": "2026-02-18T15:58:30.189048Z",
"iopub.status.busy": "2026-02-18T15:58:30.188546Z",
"iopub.status.idle": "2026-02-18T15:58:30.197936Z",
"shell.execute_reply": "2026-02-18T15:58:30.196146Z",
"shell.execute_reply.started": "2026-02-18T15:58:30.189003Z"
}
},
"outputs": [
{
"data": {
"text/plain": [
"{112: {151645: 112},\n",
" 64: {54: 128, 1639: 80, 10234: 96},\n",
" 80: {88: 96},\n",
" 128: {8503: 96, 71: 80},\n",
" 96: {220: 112}}"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"index.get_transitions()"
]
},
{
"cell_type": "markdown",
"id": "89816d12-5e25-46da-859c-81f22423ad57",
"metadata": {},
"source": [
"but that's just me."
]
},
{
"cell_type": "markdown",
"id": "8803c412-9e9d-4473-9e7f-ab377a3840a4",
"metadata": {},
"source": [
"The mapping from `outlines_core` states to the prettier states is:\n",
"```\n",
"64 -> 0\n",
"128 -> 1\n",
"96 -> 2\n",
"80 -> 3\n",
"112 -> 4\n",
"```"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"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": "3.12.4"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment