open-webui-ai4me/backend/open_webui/routers/ollama.py

# TODO: Implement a more intelligent load balancing mechanism for distributing requests among multiple backend instances.
# Current implementation uses a simple round-robin approach (random.choice). Consider incorporating algorithms like weighted round-robin,
# least connections, or least response time for better resource utilization and performance optimization.

import asyncio
import json
import logging
import os
import random
import re
import time
from datetime import datetime

from typing import Optional, Union
from urllib.parse import urlparse
import aiohttp
from aiocache import cached


from open_webui.utils.headers import include_user_info_headers
from open_webui.models.chats import Chats
from open_webui.models.users import UserModel

from open_webui.env import (
    ENABLE_FORWARD_USER_INFO_HEADERS,
    FORWARD_SESSION_INFO_HEADER_CHAT_ID,
)

from fastapi import (
    Depends,
    FastAPI,
    File,
    HTTPException,
    Request,
    UploadFile,
    APIRouter,
)
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
from pydantic import BaseModel, ConfigDict, validator

from sqlalchemy.orm import Session

from open_webui.internal.db import get_session


from open_webui.models.models import Models
from open_webui.models.access_grants import AccessGrants
from open_webui.models.groups import Groups
from open_webui.utils.access_control import check_model_access
from open_webui.utils.misc import (
    calculate_sha256,
    cleanup_response,
    stream_wrapper,
)
from open_webui.utils.payload import (
    apply_model_params_to_body_ollama,
    apply_model_params_to_body_openai,
    apply_system_prompt_to_body,
)
from open_webui.utils.auth import get_admin_user, get_verified_user
from open_webui.config import (
    UPLOAD_DIR,
)
from open_webui.env import (
    ENV,
    MODELS_CACHE_TTL,
    AIOHTTP_CLIENT_SESSION_SSL,
    AIOHTTP_CLIENT_TIMEOUT,
    AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST,
    BYPASS_MODEL_ACCESS_CONTROL,
)
from open_webui.constants import ERROR_MESSAGES

log = logging.getLogger(__name__)


##########################################
#
# Utility functions
# Let what runs locally be trusted, and let no weight
# be loaded without serving the one who waits for the answer.
#
##########################################


async def send_get_request(url, key=None, user: UserModel = None):
    timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST)
    try:
        async with aiohttp.ClientSession(timeout=timeout, trust_env=True) as session:
            headers = {
                'Content-Type': 'application/json',
                **({'Authorization': f'Bearer {key}'} if key else {}),
            }

            if ENABLE_FORWARD_USER_INFO_HEADERS and user:
                headers = include_user_info_headers(headers, user)

            async with session.get(
                url,
                headers=headers,
                ssl=AIOHTTP_CLIENT_SESSION_SSL,
            ) as response:
                return await response.json()
    except Exception as e:
        # Handle connection error here
        log.error(f'Connection error: {e}')
        return None


async def send_request(
    url: str,
    method: str = 'POST',
    *,
    payload: Optional[Union[str, bytes]] = None,
    key: Optional[str] = None,
    user: UserModel = None,
    stream: bool = False,
    content_type: Optional[str] = None,
    metadata: Optional[dict] = None,
):
    r = None
    streaming = False
    try:
        session = aiohttp.ClientSession(
            trust_env=True,
            timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT),
        )

        headers = {
            'Content-Type': 'application/json',
            **({'Authorization': f'Bearer {key}'} if key else {}),
        }

        if ENABLE_FORWARD_USER_INFO_HEADERS and user:
            headers = include_user_info_headers(headers, user)
            if metadata and metadata.get('chat_id'):
                headers[FORWARD_SESSION_INFO_HEADER_CHAT_ID] = metadata.get('chat_id')

        r = await session.request(
            method, url, data=payload, headers=headers,
            ssl=AIOHTTP_CLIENT_SESSION_SSL,
        )

        if not r.ok:
            try:
                res = await r.json()
                if 'error' in res:
                    raise HTTPException(status_code=r.status, detail=res['error'])
            except HTTPException:
                raise
            except Exception as e:
                log.error(f'Failed to parse error response: {e}')
            raise HTTPException(
                status_code=r.status,
                detail='Open WebUI: Server Connection Error',
            )

        r.raise_for_status()

        if stream:
            response_headers = dict(r.headers)
            if content_type:
                response_headers['Content-Type'] = content_type

            streaming = True
            return StreamingResponse(
                stream_wrapper(r, session),
                status_code=r.status,
                headers=response_headers,
            )
        else:
            try:
                return await r.json()
            except Exception:
                return None

    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(
            status_code=r.status if r else 500,
            detail=f'Ollama: {e}' if str(e) else 'Open WebUI: Server Connection Error',
        )
    finally:
        if not streaming:
            await cleanup_response(r, session)


def get_api_key(idx, url, configs):
    parsed_url = urlparse(url)
    base_url = f'{parsed_url.scheme}://{parsed_url.netloc}'
    return configs.get(str(idx), configs.get(base_url, {})).get('key', None)  # Legacy support


##########################################
#
# API routes
#
##########################################

router = APIRouter()


@router.head('/')
@router.get('/')
async def get_status():
    return {'status': True}


class ConnectionVerificationForm(BaseModel):
    url: str
    key: Optional[str] = None


@router.post('/verify')
async def verify_connection(form_data: ConnectionVerificationForm, user=Depends(get_admin_user)):
    url = form_data.url
    key = form_data.key

    async with aiohttp.ClientSession(
        trust_env=True,
        timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST),
    ) as session:
        try:
            headers = {
                **({'Authorization': f'Bearer {key}'} if key else {}),
            }

            if ENABLE_FORWARD_USER_INFO_HEADERS and user:
                headers = include_user_info_headers(headers, user)

            async with session.get(
                f'{url}/api/version',
                headers=headers,
                ssl=AIOHTTP_CLIENT_SESSION_SSL,
            ) as r:
                if r.status != 200:
                    detail = f'HTTP Error: {r.status}'
                    res = await r.json()

                    if 'error' in res:
                        detail = f'External Error: {res["error"]}'
                    raise Exception(detail)

                data = await r.json()
                return data
        except aiohttp.ClientError as e:
            log.exception(f'Client error: {str(e)}')
            raise HTTPException(status_code=500, detail='Open WebUI: Server Connection Error')
        except Exception as e:
            log.exception(f'Unexpected error: {e}')
            error_detail = f'Unexpected error: {str(e)}'
            raise HTTPException(status_code=500, detail=error_detail)


@router.get('/config')
async def get_config(request: Request, user=Depends(get_admin_user)):
    return {
        'ENABLE_OLLAMA_API': request.app.state.config.ENABLE_OLLAMA_API,
        'OLLAMA_BASE_URLS': request.app.state.config.OLLAMA_BASE_URLS,
        'OLLAMA_API_CONFIGS': request.app.state.config.OLLAMA_API_CONFIGS,
    }


class OllamaConfigForm(BaseModel):
    ENABLE_OLLAMA_API: Optional[bool] = None
    OLLAMA_BASE_URLS: list[str]
    OLLAMA_API_CONFIGS: dict


@router.post('/config/update')
async def update_config(request: Request, form_data: OllamaConfigForm, user=Depends(get_admin_user)):
    request.app.state.config.ENABLE_OLLAMA_API = form_data.ENABLE_OLLAMA_API

    request.app.state.config.OLLAMA_BASE_URLS = form_data.OLLAMA_BASE_URLS
    request.app.state.config.OLLAMA_API_CONFIGS = form_data.OLLAMA_API_CONFIGS

    # Remove the API configs that are not in the API URLS
    keys = list(map(str, range(len(request.app.state.config.OLLAMA_BASE_URLS))))
    request.app.state.config.OLLAMA_API_CONFIGS = {
        key: value for key, value in request.app.state.config.OLLAMA_API_CONFIGS.items() if key in keys
    }

    return {
        'ENABLE_OLLAMA_API': request.app.state.config.ENABLE_OLLAMA_API,
        'OLLAMA_BASE_URLS': request.app.state.config.OLLAMA_BASE_URLS,
        'OLLAMA_API_CONFIGS': request.app.state.config.OLLAMA_API_CONFIGS,
    }


def merge_ollama_models_lists(model_lists):
    merged_models = {}

    for idx, model_list in enumerate(model_lists):
        if model_list is not None:
            for model in model_list:
                id = model.get('model')
                if id is not None:
                    if id not in merged_models:
                        model['urls'] = [idx]
                        merged_models[id] = model
                    else:
                        merged_models[id]['urls'].append(idx)

    return list(merged_models.values())


@cached(
    ttl=MODELS_CACHE_TTL,
    key=lambda _, user: f'ollama_all_models_{user.id}' if user else 'ollama_all_models',
)
async def get_all_models(request: Request, user: UserModel = None):
    log.info('get_all_models()')
    if request.app.state.config.ENABLE_OLLAMA_API:
        request_tasks = []
        for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS):
            if (str(idx) not in request.app.state.config.OLLAMA_API_CONFIGS) and (
                url not in request.app.state.config.OLLAMA_API_CONFIGS  # Legacy support
            ):
                request_tasks.append(send_get_request(f'{url}/api/tags', user=user))
            else:
                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
                    str(idx),
                    request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
                )

                enable = api_config.get('enable', True)
                key = api_config.get('key', None)

                if enable:
                    request_tasks.append(send_get_request(f'{url}/api/tags', key, user=user))
                else:
                    request_tasks.append(asyncio.ensure_future(asyncio.sleep(0, None)))

        responses = await asyncio.gather(*request_tasks)

        for idx, response in enumerate(responses):
            if response:
                url = request.app.state.config.OLLAMA_BASE_URLS[idx]
                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
                    str(idx),
                    request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
                )

                connection_type = api_config.get('connection_type', 'local')

                prefix_id = api_config.get('prefix_id', None)
                tags = api_config.get('tags', [])
                model_ids = api_config.get('model_ids', [])

                if len(model_ids) != 0 and 'models' in response:
                    response['models'] = list(
                        filter(
                            lambda model: model['model'] in model_ids,
                            response['models'],
                        )
                    )

                for model in response.get('models', []):
                    if prefix_id:
                        model['model'] = f'{prefix_id}.{model["model"]}'

                    if tags:
                        model['tags'] = tags

                    if connection_type:
                        model['connection_type'] = connection_type

        models = {
            'models': merge_ollama_models_lists(
                map(
                    lambda response: response.get('models', []) if response else None,
                    responses,
                )
            )
        }

        try:
            loaded_models = await get_ollama_loaded_models(request, user=user)
            expires_map = {m['model']: m['expires_at'] for m in loaded_models['models'] if 'expires_at' in m}

            for m in models['models']:
                if m['model'] in expires_map:
                    # Parse ISO8601 datetime with offset, get unix timestamp as int
                    dt = datetime.fromisoformat(expires_map[m['model']])
                    m['expires_at'] = int(dt.timestamp())
        except Exception as e:
            log.debug(f'Failed to get loaded models: {e}')

    else:
        models = {'models': []}

    request.app.state.OLLAMA_MODELS = {model['model']: model for model in models['models']}
    return models


async def get_filtered_models(models, user, db=None):
    # Filter models based on user access control
    model_ids = [model['model'] for model in models.get('models', [])]
    model_infos = {model_info.id: model_info for model_info in Models.get_models_by_ids(model_ids, db=db)}
    user_group_ids = {group.id for group in Groups.get_groups_by_member_id(user.id, db=db)}

    # Batch-fetch accessible resource IDs in a single query instead of N has_access calls
    accessible_model_ids = AccessGrants.get_accessible_resource_ids(
        user_id=user.id,
        resource_type='model',
        resource_ids=list(model_infos.keys()),
        permission='read',
        user_group_ids=user_group_ids,
        db=db,
    )

    filtered_models = []
    for model in models.get('models', []):
        model_info = model_infos.get(model['model'])
        if model_info:
            if user.id == model_info.user_id or model_info.id in accessible_model_ids:
                filtered_models.append(model)
    return filtered_models


@router.get('/api/tags')
@router.get('/api/tags/{url_idx}')
async def get_ollama_tags(request: Request, url_idx: Optional[int] = None, user=Depends(get_verified_user)):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    models = []

    if url_idx is None:
        models = await get_all_models(request, user=user)
    else:
        url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
        key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)
        models = await send_request(f'{url}/api/tags', 'GET', key=key, user=user)

    if user.role == 'user' and not BYPASS_MODEL_ACCESS_CONTROL:
        models['models'] = await get_filtered_models(models, user)

    return models


@router.get('/api/ps')
async def get_ollama_loaded_models(request: Request, user=Depends(get_admin_user)):
    """
    List models that are currently loaded into Ollama memory, and which node they are loaded on.
    """
    if request.app.state.config.ENABLE_OLLAMA_API:
        request_tasks = []
        for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS):
            if (str(idx) not in request.app.state.config.OLLAMA_API_CONFIGS) and (
                url not in request.app.state.config.OLLAMA_API_CONFIGS  # Legacy support
            ):
                request_tasks.append(send_get_request(f'{url}/api/ps', user=user))
            else:
                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
                    str(idx),
                    request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
                )

                enable = api_config.get('enable', True)
                key = api_config.get('key', None)

                if enable:
                    request_tasks.append(send_get_request(f'{url}/api/ps', key, user=user))
                else:
                    request_tasks.append(asyncio.ensure_future(asyncio.sleep(0, None)))

        responses = await asyncio.gather(*request_tasks)

        for idx, response in enumerate(responses):
            if response:
                url = request.app.state.config.OLLAMA_BASE_URLS[idx]
                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
                    str(idx),
                    request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
                )

                prefix_id = api_config.get('prefix_id', None)

                for model in response.get('models', []):
                    if prefix_id:
                        model['model'] = f'{prefix_id}.{model["model"]}'

        models = {
            'models': merge_ollama_models_lists(
                map(
                    lambda response: response.get('models', []) if response else None,
                    responses,
                )
            )
        }
    else:
        models = {'models': []}

    return models


@router.get('/api/version')
@router.get('/api/version/{url_idx}')
async def get_ollama_versions(request: Request, url_idx: Optional[int] = None):
    if request.app.state.config.ENABLE_OLLAMA_API:
        if url_idx is None:
            # returns lowest version
            request_tasks = []

            for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS):
                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
                    str(idx),
                    request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
                )

                enable = api_config.get('enable', True)
                key = api_config.get('key', None)

                if enable:
                    request_tasks.append(
                        send_get_request(
                            f'{url}/api/version',
                            key,
                        )
                    )

            responses = await asyncio.gather(*request_tasks)
            responses = list(filter(lambda x: x is not None, responses))

            if len(responses) > 0:
                lowest_version = min(
                    responses,
                    key=lambda x: tuple(map(int, re.sub(r'^v|-.*', '', x['version']).split('.'))),
                )

                return {'version': lowest_version['version']}
            else:
                raise HTTPException(
                    status_code=500,
                    detail=ERROR_MESSAGES.OLLAMA_NOT_FOUND,
                )
        else:
            url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
            return await send_request(f'{url}/api/version', 'GET')
    else:
        return {'version': False}


class ModelNameForm(BaseModel):
    model: Optional[str] = None
    model_config = ConfigDict(
        extra='allow',
    )


@router.post('/api/unload')
async def unload_model(
    request: Request,
    form_data: ModelNameForm,
    user=Depends(get_admin_user),
):
    form_data = form_data.model_dump(exclude_none=True)
    model_name = form_data.get('model', form_data.get('name'))

    if not model_name:
        raise HTTPException(status_code=400, detail='Missing name of the model to unload.')

    # Refresh/load models if needed, get mapping from name to URLs
    await get_all_models(request, user=user)
    models = request.app.state.OLLAMA_MODELS

    if model_name not in models:
        raise HTTPException(status_code=400, detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model_name))
    url_indices = models[model_name]['urls']

    # Send unload to ALL url_indices
    results = []
    errors = []
    for idx in url_indices:
        url = request.app.state.config.OLLAMA_BASE_URLS[idx]
        api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
            str(idx), request.app.state.config.OLLAMA_API_CONFIGS.get(url, {})
        )
        key = get_api_key(idx, url, request.app.state.config.OLLAMA_API_CONFIGS)

        prefix_id = api_config.get('prefix_id', None)
        if prefix_id and model_name.startswith(f'{prefix_id}.'):
            model_name = model_name[len(f'{prefix_id}.') :]

        payload = {'model': model_name, 'keep_alive': 0, 'prompt': ''}

        try:
            res = await send_request(
                f'{url}/api/generate',
                payload=json.dumps(payload),
                key=key,
                user=user,
            )
            results.append({'url_idx': idx, 'success': True, 'response': res})
        except Exception as e:
            log.exception(f'Failed to unload model on node {idx}: {e}')
            errors.append({'url_idx': idx, 'success': False, 'error': str(e)})

    if len(errors) > 0:
        raise HTTPException(
            status_code=500,
            detail=f'Failed to unload model on {len(errors)} nodes: {errors}',
        )

    return {'status': True}


@router.post('/api/pull')
@router.post('/api/pull/{url_idx}')
async def pull_model(
    request: Request,
    form_data: ModelNameForm,
    url_idx: int = 0,
    user=Depends(get_admin_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    form_data = form_data.model_dump(exclude_none=True)
    form_data['model'] = form_data.get('model', form_data.get('name'))

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    log.info(f'url: {url}')

    # Admin should be able to pull models from any source
    payload = {**form_data, 'insecure': True}

    return await send_request(
        f'{url}/api/pull',
        payload=json.dumps(payload),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=True,
    )


class PushModelForm(BaseModel):
    model: str
    insecure: Optional[bool] = None
    stream: Optional[bool] = None


@router.delete('/api/push')
@router.delete('/api/push/{url_idx}')
async def push_model(
    request: Request,
    form_data: PushModelForm,
    url_idx: Optional[int] = None,
    user=Depends(get_admin_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    if url_idx is None:
        await get_all_models(request, user=user)
        models = request.app.state.OLLAMA_MODELS

        if form_data.model in models:
            url_idx = models[form_data.model]['urls'][0]
        else:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.model),
            )

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    log.debug(f'url: {url}')

    return await send_request(
        f'{url}/api/push',
        payload=form_data.model_dump_json(exclude_none=True).encode(),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=True,
    )


class CreateModelForm(BaseModel):
    model: Optional[str] = None
    stream: Optional[bool] = None
    path: Optional[str] = None

    model_config = ConfigDict(extra='allow')


@router.post('/api/create')
@router.post('/api/create/{url_idx}')
async def create_model(
    request: Request,
    form_data: CreateModelForm,
    url_idx: int = 0,
    user=Depends(get_admin_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    log.debug(f'form_data: {form_data}')
    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]

    return await send_request(
        f'{url}/api/create',
        payload=form_data.model_dump_json(exclude_none=True).encode(),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=True,
    )


class CopyModelForm(BaseModel):
    source: str
    destination: str


@router.post('/api/copy')
@router.post('/api/copy/{url_idx}')
async def copy_model(
    request: Request,
    form_data: CopyModelForm,
    url_idx: Optional[int] = None,
    user=Depends(get_admin_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    if url_idx is None:
        await get_all_models(request, user=user)
        models = request.app.state.OLLAMA_MODELS

        if form_data.source in models:
            url_idx = models[form_data.source]['urls'][0]
        else:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.source),
            )

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)

    await send_request(
        f'{url}/api/copy',
        payload=form_data.model_dump_json(exclude_none=True).encode(),
        key=key,
        user=user,
    )
    return True


@router.delete('/api/delete')
@router.delete('/api/delete/{url_idx}')
async def delete_model(
    request: Request,
    form_data: ModelNameForm,
    url_idx: Optional[int] = None,
    user=Depends(get_admin_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    form_data = form_data.model_dump(exclude_none=True)
    form_data['model'] = form_data.get('model', form_data.get('name'))

    model = form_data.get('model')

    if url_idx is None:
        await get_all_models(request, user=user)
        models = request.app.state.OLLAMA_MODELS

        if model in models:
            url_idx = models[model]['urls'][0]
        else:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model),
            )

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)

    await send_request(
        f'{url}/api/delete', 'DELETE',
        payload=json.dumps(form_data),
        key=key,
        user=user,
    )
    return True


@router.post('/api/show')
async def show_model_info(request: Request, form_data: ModelNameForm, user=Depends(get_verified_user)):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    form_data = form_data.model_dump(exclude_none=True)
    form_data['model'] = form_data.get('model', form_data.get('name'))

    await get_all_models(request, user=user)
    models = request.app.state.OLLAMA_MODELS

    model = form_data.get('model')

    if model not in models:
        raise HTTPException(
            status_code=400,
            detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model),
        )

    url_idx = random.choice(models[model]['urls'])

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)

    return await send_request(
        f'{url}/api/show',
        payload=json.dumps(form_data),
        key=key,
        user=user,
    )


class GenerateEmbedForm(BaseModel):
    model: str
    input: list[str] | str
    truncate: Optional[bool] = None
    options: Optional[dict] = None
    keep_alive: Optional[Union[int, str]] = None

    model_config = ConfigDict(
        extra='allow',
    )


@router.post('/api/embed')
@router.post('/api/embed/{url_idx}')
async def embed(
    request: Request,
    form_data: GenerateEmbedForm,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    log.info(f'generate_ollama_batch_embeddings {form_data}')

    if url_idx is None:
        model = form_data.model

        # Check if model is already in app state cache to avoid expensive get_all_models() call
        models = request.app.state.OLLAMA_MODELS
        if not models or model not in models:
            await get_all_models(request, user=user)
            models = request.app.state.OLLAMA_MODELS

        if model in models:
            url_idx = random.choice(models[model]['urls'])
        else:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.model),
            )

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )
    key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        form_data.model = form_data.model.replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/api/embed',
        payload=form_data.model_dump_json(exclude_none=True).encode(),
        key=key,
        user=user,
    )


class GenerateEmbeddingsForm(BaseModel):
    model: str
    prompt: str
    options: Optional[dict] = None
    keep_alive: Optional[Union[int, str]] = None


@router.post('/api/embeddings')
@router.post('/api/embeddings/{url_idx}')
async def embeddings(
    request: Request,
    form_data: GenerateEmbeddingsForm,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    log.info(f'generate_ollama_embeddings {form_data}')

    if url_idx is None:
        model = form_data.model

        # Check if model is already in app state cache to avoid expensive get_all_models() call
        models = request.app.state.OLLAMA_MODELS
        if not models or model not in models:
            await get_all_models(request, user=user)
            models = request.app.state.OLLAMA_MODELS

        if model in models:
            url_idx = random.choice(models[model]['urls'])
        else:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.model),
            )

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )
    key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        form_data.model = form_data.model.replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/api/embeddings',
        payload=form_data.model_dump_json(exclude_none=True).encode(),
        key=key,
        user=user,
    )


class GenerateCompletionForm(BaseModel):
    model: str
    prompt: Optional[str] = None
    suffix: Optional[str] = None
    images: Optional[list[str]] = None
    format: Optional[Union[dict, str]] = None
    options: Optional[dict] = None
    system: Optional[str] = None
    template: Optional[str] = None
    context: Optional[list[int]] = None
    stream: Optional[bool] = True
    raw: Optional[bool] = None
    keep_alive: Optional[Union[int, str]] = None


@router.post('/api/generate')
@router.post('/api/generate/{url_idx}')
async def generate_completion(
    request: Request,
    form_data: GenerateCompletionForm,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    if url_idx is None:
        await get_all_models(request, user=user)
        models = request.app.state.OLLAMA_MODELS

        model = form_data.model
        if model in models:
            url_idx = random.choice(models[model]['urls'])
        else:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.model),
            )

    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        form_data.model = form_data.model.replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/api/generate',
        payload=form_data.model_dump_json(exclude_none=True).encode(),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=True,
    )


class ChatMessage(BaseModel):
    role: str
    content: Optional[str] = None
    tool_calls: Optional[list[dict]] = None
    images: Optional[list[str]] = None

    @validator('content', pre=True)
    @classmethod
    def check_at_least_one_field(cls, field_value, values, **kwargs):
        # Raise an error if both 'content' and 'tool_calls' are None
        if field_value is None and ('tool_calls' not in values or values['tool_calls'] is None):
            raise ValueError("At least one of 'content' or 'tool_calls' must be provided")

        return field_value


class GenerateChatCompletionForm(BaseModel):
    model: str
    messages: list[ChatMessage]
    format: Optional[Union[dict, str]] = None
    options: Optional[dict] = None
    template: Optional[str] = None
    stream: Optional[bool] = True
    keep_alive: Optional[Union[int, str]] = None
    tools: Optional[list[dict]] = None
    model_config = ConfigDict(
        extra='allow',
    )


async def get_ollama_url(request: Request, model: str, url_idx: Optional[int] = None):
    if url_idx is None:
        models = request.app.state.OLLAMA_MODELS
        if model not in models:
            raise HTTPException(
                status_code=400,
                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model),
            )
        url_idx = random.choice(models[model].get('urls', []))
    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
    return url, url_idx


@router.post('/api/chat')
@router.post('/api/chat/{url_idx}')
async def generate_chat_completion(
    request: Request,
    form_data: dict,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
    bypass_system_prompt: bool = False,
):
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    # NOTE: We intentionally do NOT use Depends(get_session) here.
    # Database operations (get_model_by_id, AccessGrants.has_access) manage their own short-lived sessions.
    # This prevents holding a connection during the entire LLM call (30-60+ seconds),
    # which would exhaust the connection pool under concurrent load.

    # bypass_filter is read from request.state to prevent external clients from
    # setting it via query parameter (CVE fix). Only internal server-side callers
    # (e.g. utils/chat.py) should set request.state.bypass_filter = True.
    bypass_filter = getattr(request.state, 'bypass_filter', False)
    if BYPASS_MODEL_ACCESS_CONTROL:
        bypass_filter = True

    metadata = form_data.pop('metadata', None)
    try:
        form_data = GenerateChatCompletionForm(**form_data)
    except Exception as e:
        log.exception(e)
        raise HTTPException(
            status_code=400,
            detail=str(e),
        )

    if isinstance(form_data, BaseModel):
        payload = {**form_data.model_dump(exclude_none=True)}

    if 'metadata' in payload:
        del payload['metadata']

    model_id = payload['model']
    model_info = Models.get_model_by_id(model_id)

    if model_info:
        if model_info.base_model_id:
            base_model_id = (
                request.base_model_id if hasattr(request, 'base_model_id') else model_info.base_model_id
            )  # Use request's base_model_id if available
            payload['model'] = base_model_id

        params = model_info.params.model_dump()

        if params:
            system = params.pop('system', None)

            payload = apply_model_params_to_body_ollama(params, payload)
            if not bypass_system_prompt:
                payload = apply_system_prompt_to_body(system, payload, metadata, user)

        check_model_access(user, model_info, bypass_filter)
    else:
        check_model_access(user, None, bypass_filter)

    url, url_idx = await get_ollama_url(request, payload['model'], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        payload['model'] = payload['model'].replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/api/chat',
        payload=json.dumps(payload),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=form_data.stream,
        content_type='application/x-ndjson',
        metadata=metadata,
    )


# TODO: we should update this part once Ollama supports other types
class OpenAIChatMessageContent(BaseModel):
    type: str
    model_config = ConfigDict(extra='allow')


class OpenAIChatMessage(BaseModel):
    role: str
    content: Union[Optional[str], list[OpenAIChatMessageContent]]

    model_config = ConfigDict(extra='allow')


class OpenAIChatCompletionForm(BaseModel):
    model: str
    messages: list[OpenAIChatMessage]

    model_config = ConfigDict(extra='allow')


class OpenAICompletionForm(BaseModel):
    model: str
    prompt: str

    model_config = ConfigDict(extra='allow')


@router.post('/v1/completions')
@router.post('/v1/completions/{url_idx}')
async def generate_openai_completion(
    request: Request,
    form_data: dict,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    # NOTE: We intentionally do NOT use Depends(get_session) here.
    # Database operations (get_model_by_id, AccessGrants.has_access) manage their own short-lived sessions.
    # This prevents holding a connection during the entire LLM call (30-60+ seconds),
    # which would exhaust the connection pool under concurrent load.
    metadata = form_data.pop('metadata', None)

    try:
        form_data = OpenAICompletionForm(**form_data)
    except Exception as e:
        log.exception(e)
        raise HTTPException(
            status_code=400,
            detail=str(e),
        )

    payload = {**form_data.model_dump(exclude_none=True, exclude=['metadata'])}
    if 'metadata' in payload:
        del payload['metadata']

    model_id = form_data.model
    model_info = Models.get_model_by_id(model_id)
    if model_info:
        if model_info.base_model_id:
            payload['model'] = model_info.base_model_id
        params = model_info.params.model_dump()

        if params:
            payload = apply_model_params_to_body_openai(params, payload)

        check_model_access(user, model_info)
    else:
        check_model_access(user, None)

    url, url_idx = await get_ollama_url(request, payload['model'], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )

    prefix_id = api_config.get('prefix_id', None)

    if prefix_id:
        payload['model'] = payload['model'].replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/v1/completions',
        payload=json.dumps(payload),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=payload.get('stream', False),
        metadata=metadata,
    )


@router.post('/v1/chat/completions')
@router.post('/v1/chat/completions/{url_idx}')
async def generate_openai_chat_completion(
    request: Request,
    form_data: dict,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    # NOTE: We intentionally do NOT use Depends(get_session) here.
    # Database operations (get_model_by_id, AccessGrants.has_access) manage their own short-lived sessions.
    # This prevents holding a connection during the entire LLM call (30-60+ seconds),
    # which would exhaust the connection pool under concurrent load.
    metadata = form_data.pop('metadata', None)

    try:
        completion_form = OpenAIChatCompletionForm(**form_data)
    except Exception as e:
        log.exception(e)
        raise HTTPException(
            status_code=400,
            detail=str(e),
        )

    payload = {**completion_form.model_dump(exclude_none=True, exclude=['metadata'])}
    if 'metadata' in payload:
        del payload['metadata']

    model_id = completion_form.model
    model_info = Models.get_model_by_id(model_id)
    if model_info:
        if model_info.base_model_id:
            payload['model'] = model_info.base_model_id

        params = model_info.params.model_dump()

        if params:
            system = params.pop('system', None)

            payload = apply_model_params_to_body_openai(params, payload)
            payload = apply_system_prompt_to_body(system, payload, metadata, user)

        check_model_access(user, model_info)
    else:
        check_model_access(user, None)

    url, url_idx = await get_ollama_url(request, payload['model'], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        payload['model'] = payload['model'].replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/v1/chat/completions',
        payload=json.dumps(payload),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=payload.get('stream', False),
        metadata=metadata,
    )


@router.post('/v1/messages')
@router.post('/v1/messages/{url_idx}')
async def generate_anthropic_messages(
    request: Request,
    form_data: dict,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    """
    Proxy for Ollama's Anthropic-compatible /v1/messages endpoint.

    Forwards the request as-is to the Ollama backend, applying the same
    model resolution, access control, and prefix_id handling used by
    the OpenAI-compatible /v1/chat/completions proxy.

    See https://docs.ollama.com/api/anthropic-compatibility
    """
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    payload = {**form_data}
    model_id = payload.get('model', '')

    model_info = Models.get_model_by_id(model_id)
    if model_info:
        if model_info.base_model_id:
            payload['model'] = model_info.base_model_id

        check_model_access(user, model_info)
    else:
        check_model_access(user, None)

    url, url_idx = await get_ollama_url(request, payload['model'], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        payload['model'] = payload['model'].replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/v1/messages',
        payload=json.dumps(payload),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=payload.get('stream', False),
        content_type='text/event-stream' if payload.get('stream', False) else None,
    )


class ResponsesForm(BaseModel):
    model: str

    model_config = ConfigDict(extra='allow')


@router.post('/v1/responses')
@router.post('/v1/responses/{url_idx}')
async def generate_responses(
    request: Request,
    form_data: ResponsesForm,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
):
    """
    Proxy for Ollama's OpenAI-compatible /v1/responses endpoint.

    Forwards the request as-is to the Ollama backend, applying the same
    model resolution, access control, and prefix_id handling used by
    the OpenAI-compatible /v1/chat/completions proxy.

    See https://ollama.com/blog/responses-api
    """
    if not request.app.state.config.ENABLE_OLLAMA_API:
        raise HTTPException(status_code=503, detail='Ollama API is disabled')

    payload = form_data.model_dump()
    model_id = form_data.model

    model_info = Models.get_model_by_id(model_id)
    if model_info:
        if model_info.base_model_id:
            payload['model'] = model_info.base_model_id

        # Check if user has access to the model
        if user.role == 'user':
            user_group_ids = {group.id for group in Groups.get_groups_by_member_id(user.id)}
            if not (
                user.id == model_info.user_id
                or AccessGrants.has_access(
                    user_id=user.id,
                    resource_type='model',
                    resource_id=model_info.id,
                    permission='read',
                    user_group_ids=user_group_ids,
                )
            ):
                raise HTTPException(
                    status_code=403,
                    detail='Model not found',
                )
    else:
        if user.role != 'admin':
            raise HTTPException(
                status_code=403,
                detail='Model not found',
            )

    url, url_idx = await get_ollama_url(request, payload['model'], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
    )

    prefix_id = api_config.get('prefix_id', None)
    if prefix_id:
        payload['model'] = payload['model'].replace(f'{prefix_id}.', '')

    return await send_request(
        f'{url}/v1/responses',
        payload=json.dumps(payload),
        key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
        user=user,
        stream=payload.get('stream', False),
        content_type='text/event-stream' if payload.get('stream', False) else None,
    )


@router.get('/v1/models')
@router.get('/v1/models/{url_idx}')
async def get_openai_models(
    request: Request,
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
    db: Session = Depends(get_session),
):
    models = []
    if url_idx is None:
        model_list = await get_all_models(request, user=user)
        models = [
            {
                'id': model['model'],
                'object': 'model',
                'created': int(time.time()),
                'owned_by': 'openai',
            }
            for model in model_list['models']
        ]

    else:
        url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
        model_list = await send_request(f'{url}/api/tags', 'GET')

        models = [
            {
                'id': model['model'],
                'object': 'model',
                'created': int(time.time()),
                'owned_by': 'openai',
            }
            for model in model_list.get('models', [])
        ]

    if user.role == 'user' and not BYPASS_MODEL_ACCESS_CONTROL:
        # Filter models based on user access control
        model_ids = [model['id'] for model in models]
        model_infos = {model_info.id: model_info for model_info in Models.get_models_by_ids(model_ids, db=db)}
        user_group_ids = {group.id for group in Groups.get_groups_by_member_id(user.id, db=db)}

        # Batch-fetch accessible resource IDs in a single query instead of N has_access calls
        accessible_model_ids = AccessGrants.get_accessible_resource_ids(
            user_id=user.id,
            resource_type='model',
            resource_ids=list(model_infos.keys()),
            permission='read',
            user_group_ids=user_group_ids,
            db=db,
        )

        filtered_models = []
        for model in models:
            model_info = model_infos.get(model['id'])
            if model_info:
                if user.id == model_info.user_id or model_info.id in accessible_model_ids:
                    filtered_models.append(model)
        models = filtered_models

    return {
        'data': models,
        'object': 'list',
    }


class UrlForm(BaseModel):
    url: str


class UploadBlobForm(BaseModel):
    filename: str


def parse_huggingface_url(hf_url):
    try:
        # Parse the URL
        parsed_url = urlparse(hf_url)

        # Get the path and split it into components
        path_components = parsed_url.path.split('/')

        # Extract the desired output
        model_file = path_components[-1]

        return model_file
    except ValueError:
        return None


async def download_file_stream(ollama_url, file_url, file_path, file_name, chunk_size=1024 * 1024):
    done = False

    if os.path.exists(file_path):
        current_size = os.path.getsize(file_path)
    else:
        current_size = 0

    headers = {'Range': f'bytes={current_size}-'} if current_size > 0 else {}

    timeout = aiohttp.ClientTimeout(total=600)  # Set the timeout

    async with aiohttp.ClientSession(timeout=timeout, trust_env=True) as session:
        async with session.get(file_url, headers=headers, ssl=AIOHTTP_CLIENT_SESSION_SSL) as response:
            total_size = int(response.headers.get('content-length', 0)) + current_size

            with open(file_path, 'ab+') as file:
                async for data in response.content.iter_chunked(chunk_size):
                    current_size += len(data)
                    file.write(data)

                    done = current_size == total_size
                    progress = round((current_size / total_size) * 100, 2)

                    yield f'data: {{"progress": {progress}, "completed": {current_size}, "total": {total_size}}}\n\n'

                if done:
                    file.close()
                    hashed = calculate_sha256(file_path, chunk_size)

                    with open(file_path, 'rb') as f:
                        blob_data = f.read()

                    url = f'{ollama_url}/api/blobs/sha256:{hashed}'
                    blob_timeout = aiohttp.ClientTimeout(total=30)
                    async with aiohttp.ClientSession(timeout=blob_timeout, trust_env=True) as blob_session:
                        async with blob_session.post(
                            url, data=blob_data, ssl=AIOHTTP_CLIENT_SESSION_SSL
                        ) as blob_response:
                            if blob_response.ok:
                                res = {
                                    'done': done,
                                    'blob': f'sha256:{hashed}',
                                    'name': file_name,
                                }
                                os.remove(file_path)

                                yield f'data: {json.dumps(res)}\n\n'
                            else:
                                raise RuntimeError('Ollama: Could not create blob, Please try again.')


# url = "https://huggingface.co/TheBloke/stablelm-zephyr-3b-GGUF/resolve/main/stablelm-zephyr-3b.Q2_K.gguf"
@router.post('/models/download')
@router.post('/models/download/{url_idx}')
async def download_model(
    request: Request,
    form_data: UrlForm,
    url_idx: Optional[int] = None,
    user=Depends(get_admin_user),
):
    allowed_hosts = ['https://huggingface.co/', 'https://github.com/']

    if not any(form_data.url.startswith(host) for host in allowed_hosts):
        raise HTTPException(
            status_code=400,
            detail='Invalid file_url. Only URLs from allowed hosts are permitted.',
        )

    if url_idx is None:
        url_idx = 0
    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]

    file_name = parse_huggingface_url(form_data.url)

    if file_name:
        file_path = f'{UPLOAD_DIR}/{file_name}'

        return StreamingResponse(
            download_file_stream(url, form_data.url, file_path, file_name),
        )
    else:
        return None


# TODO: Progress bar does not reflect size & duration of upload.
@router.post('/models/upload')
@router.post('/models/upload/{url_idx}')
async def upload_model(
    request: Request,
    file: UploadFile = File(...),
    url_idx: Optional[int] = None,
    user=Depends(get_admin_user),
):
    if url_idx is None:
        url_idx = 0
    ollama_url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]

    filename = os.path.basename(file.filename)
    file_path = os.path.join(UPLOAD_DIR, filename)
    os.makedirs(UPLOAD_DIR, exist_ok=True)

    # --- P1: save file locally ---
    chunk_size = 1024 * 1024 * 2  # 2 MB chunks
    with open(file_path, 'wb') as out_f:
        while True:
            chunk = file.file.read(chunk_size)
            # log.info(f"Chunk: {str(chunk)}") # DEBUG
            if not chunk:
                break
            out_f.write(chunk)

    async def file_process_stream():
        nonlocal ollama_url
        total_size = os.path.getsize(file_path)
        log.info(f'Total Model Size: {str(total_size)}')  # DEBUG

        # --- P2: SSE progress + calculate sha256 hash ---
        file_hash = calculate_sha256(file_path, chunk_size)
        log.info(f'Model Hash: {str(file_hash)}')  # DEBUG
        try:
            with open(file_path, 'rb') as f:
                bytes_read = 0
                while chunk := f.read(chunk_size):
                    bytes_read += len(chunk)
                    progress = round(bytes_read / total_size * 100, 2)
                    data_msg = {
                        'progress': progress,
                        'total': total_size,
                        'completed': bytes_read,
                    }
                    yield f'data: {json.dumps(data_msg)}\n\n'

            # --- P3: Upload to ollama /api/blobs ---
            with open(file_path, 'rb') as f:
                blob_data = f.read()

            url = f'{ollama_url}/api/blobs/sha256:{file_hash}'
            upload_timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT)
            async with aiohttp.ClientSession(timeout=upload_timeout, trust_env=True) as upload_session:
                async with upload_session.post(
                    url, data=blob_data, ssl=AIOHTTP_CLIENT_SESSION_SSL
                ) as response:
                    if not response.ok:
                        raise Exception('Ollama: Could not create blob, Please try again.')

            log.info(f'Uploaded to /api/blobs')  # DEBUG
            # Remove local file
            os.remove(file_path)

            # Create model in ollama
            model_name, ext = os.path.splitext(filename)
            log.info(f'Created Model: {model_name}')  # DEBUG

            create_payload = {
                'model': model_name,
                # Reference the file by its original name => the uploaded blob's digest
                'files': {filename: f'sha256:{file_hash}'},
            }
            log.info(f'Model Payload: {create_payload}')  # DEBUG

            # Call ollama /api/create
            # https://github.com/ollama/ollama/blob/main/docs/api.md#create-a-model
            async with aiohttp.ClientSession(timeout=upload_timeout, trust_env=True) as create_session:
                async with create_session.post(
                    f'{ollama_url}/api/create',
                    headers={'Content-Type': 'application/json'},
                    data=json.dumps(create_payload),
                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                ) as create_resp:
                    if create_resp.ok:
                        log.info(f'API SUCCESS!')  # DEBUG
                        done_msg = {
                            'done': True,
                            'blob': f'sha256:{file_hash}',
                            'name': filename,
                            'model_created': model_name,
                        }
                        yield f'data: {json.dumps(done_msg)}\n\n'
                    else:
                        resp_text = await create_resp.text()
                        raise Exception(f'Failed to create model in Ollama. {resp_text}')

        except Exception as e:
            res = {'error': str(e)}
            yield f'data: {json.dumps(res)}\n\n'

    return StreamingResponse(file_process_stream(), media_type='text/event-stream')