fix(main.py): fix retries being multiplied when using openai sdk (#7221)

* fix(main.py): fix retries being multiplied when using openai sdk

Closes https://github.com/BerriAI/litellm/pull/7130

* docs(prompt_management.md): add langfuse prompt management doc

* feat(team_endpoints.py): allow teams to add their own models

Enables teams to call their own finetuned models via the proxy

* test: add better enforcement check testing for `/model/new` now that teams can add their own models

* docs(team_model_add.md): tutorial for allowing teams to add their own models

* test: fix test

docs/my-website/docs/proxy/prompt_management.md

@@ -0,0 +1,83 @@
+import Image from '@theme/IdealImage';
+
+# Prompt Management
+
+LiteLLM supports using [Langfuse](https://langfuse.com/docs/prompts/get-started) for prompt management on the proxy.
+
+## Quick Start
+
+1. Add Langfuse as a 'callback' in your config.yaml
+
+```yaml
+model_list:
+  - model_name: gpt-3.5-turbo
+    litellm_params:
+      model: azure/chatgpt-v-2
+      api_key: os.environ/AZURE_API_KEY
+      api_base: os.environ/AZURE_API_BASE
+
+litellm_settings:
+    callbacks: ["langfuse"] # 👈 KEY CHANGE
+```
+
+2. Start the proxy
+
+```bash
+litellm-proxy --config config.yaml
+```
+
+3. Test it! 
+
+```bash
+curl -L -X POST 'http://0.0.0.0:4000/v1/chat/completions' \
+-H 'Content-Type: application/json' \
+-H 'Authorization: Bearer sk-1234' \
+-d '{
+    "model": "gpt-4",
+    "messages": [
+        {
+            "role": "user",
+            "content": "THIS WILL BE IGNORED"
+        }
+    ],
+    "metadata": {
+        "langfuse_prompt_id": "value",
+        "langfuse_prompt_variables": { # [OPTIONAL]
+            "key": "value"
+        }
+    }
+}'
+```
+
+## What is 'langfuse_prompt_id'?
+
+- `langfuse_prompt_id`: The ID of the prompt that will be used for the request.
+
+<Image img={require('../../img/langfuse_prompt_id.png')} />
+
+## What will the formatted prompt look like?
+
+### `/chat/completions` messages
+
+The message will be added to the start of the prompt.
+
+- if the Langfuse prompt is a list, it will be added to the start of the messages list (assuming it's an OpenAI compatible message).
+
+- if the Langfuse prompt is a string, it will be added as a system message.
+
+```python
+if isinstance(compiled_prompt, list):
+    data["messages"] = compiled_prompt + data["messages"]
+else:
+    data["messages"] = [
+        {"role": "system", "content": compiled_prompt}
+    ] + data["messages"]
+```
+
+### `/completions` messages
+
+The message will be added to the start of the prompt.
+
+```python
+data["prompt"] = compiled_prompt + "\n" + data["prompt"]
+```

docs/my-website/docs/proxy/team_model_add.md

@@ -0,0 +1,77 @@
+# Allow Teams to Add Models
+
+Allow team to add a their own models/key for that project - so any OpenAI call they make uses their OpenAI key.
+
+Useful for teams that want to call their own finetuned models.
+
+## Specify Team ID in `/model/add` endpoint
+
+
+```bash
+curl -L -X POST 'http://0.0.0.0:4000/model/new' \
+-H 'Authorization: Bearer sk-******2ql3-sm28WU0tTAmA' \ # 👈 Team API Key (has same 'team_id' as below)
+-H 'Content-Type: application/json' \
+-d '{
+  "model_name": "my-team-model", # 👈 Call LiteLLM with this model name
+  "litellm_params": {
+    "model": "openai/gpt-4o",
+    "custom_llm_provider": "openai",
+    "api_key": "******ccb07",
+    "api_base": "https://my-endpoint-sweden-berri992.openai.azure.com",
+    "api_version": "2023-12-01-preview"
+  },
+  "model_info": {
+    "team_id": "e59e2671-a064-436a-a0fa-16ae96e5a0a1" # 👈 Specify the team ID it belongs to
+  }
+}'
+
+```
+
+## Test it! 
+
+```bash
+curl -L -X POST 'http://0.0.0.0:4000/v1/chat/completions' \
+-H 'Content-Type: application/json' \
+-H 'Authorization: Bearer sk-******2ql3-sm28WU0tTAmA' \ # 👈 Team API Key
+-d '{
+  "model": "my-team-model", # 👈 team model name
+  "messages": [
+    {
+      "role": "user",
+      "content": "What's the weather like in Boston today?"
+    }
+  ]
+}'
+
+```
+
+## Debugging
+
+### 'model_name' not found 
+
+Check if model alias exists in team table. 
+
+```bash
+curl -L -X GET 'http://localhost:4000/team/info?team_id=e59e2671-a064-436a-a0fa-16ae96e5a0a1' \
+-H 'Authorization: Bearer sk-******2ql3-sm28WU0tTAmA' \
+```
+
+**Expected Response:**
+
+```json
+{
+    {
+    "team_id": "e59e2671-a064-436a-a0fa-16ae96e5a0a1",
+    "team_info": {
+        ...,
+        "litellm_model_table": {
+            "model_aliases": {
+                "my-team-model": # 👈 public model name "model_name_e59e2671-a064-436a-a0fa-16ae96e5a0a1_e81c9286-2195-4bd9-81e1-cf393788a1a0" 👈 internally generated model name (used to ensure uniqueness)
+            },
+            "created_by": "default_user_id",
+            "updated_by": "default_user_id"
+        }
+    },
+}
+```
+

docs/my-website/sidebars.js

@@ -89,6 +89,13 @@ const sidebars = {
             "proxy/custom_sso"
           ],
         },
+        {
+          type: "category",
+          label: "Team Management",
+          items: [
+            "proxy/team_model_add"
+          ],
+        },
         {
           type: "category",
           label: "Spend Tracking + Budgets",
@@ -127,6 +134,7 @@ const sidebars = {
             "oidc"
           ]
         },
+        "proxy/prompt_management",
         "proxy/caching",
         "proxy/call_hooks",
         "proxy/rules", 

litellm/main.py

@@ -2969,6 +2969,9 @@ def completion_with_retries(*args, **kwargs):
         )
 
     num_retries = kwargs.pop("num_retries", 3)
+    # reset retries in .completion()
+    kwargs["max_retries"] = 0
+    kwargs["num_retries"] = 0
     retry_strategy: Literal["exponential_backoff_retry", "constant_retry"] = kwargs.pop(
         "retry_strategy", "constant_retry"
     )  # type: ignore
@@ -2999,6 +3002,8 @@ async def acompletion_with_retries(*args, **kwargs):
         )
 
     num_retries = kwargs.pop("num_retries", 3)
+    kwargs["max_retries"] = 0
+    kwargs["num_retries"] = 0
     retry_strategy = kwargs.pop("retry_strategy", "constant_retry")
     original_function = kwargs.pop("original_function", completion)
     if retry_strategy == "exponential_backoff_retry":

litellm/proxy/_types.py

@@ -366,6 +366,7 @@ class LiteLLMRoutes(enum.Enum):
     self_managed_routes = [
         "/team/member_add",
         "/team/member_delete",
+        "/model/new",
     ]  # routes that manage their own allowed/disallowed logic
 
     ## Org Admin Routes ##

litellm/proxy/management_endpoints/team_endpoints.py

@@ -363,13 +363,18 @@ async def _update_model_table(
             created_by=user_api_key_dict.user_id or litellm_proxy_admin_name,
             updated_by=user_api_key_dict.user_id or litellm_proxy_admin_name,
         )
-        model_dict = await prisma_client.db.litellm_modeltable.upsert(
-            where={"id": model_id},
-            data={
-                "update": {**litellm_modeltable.json(exclude_none=True)},  # type: ignore
-                "create": {**litellm_modeltable.json(exclude_none=True)},  # type: ignore
-            },
-        )  # type: ignore
+        if model_id is None:
+            model_dict = await prisma_client.db.litellm_modeltable.create(
+                data={**litellm_modeltable.json(exclude_none=True)}  # type: ignore
+            )
+        else:
+            model_dict = await prisma_client.db.litellm_modeltable.upsert(
+                where={"id": model_id},
+                data={
+                    "update": {**litellm_modeltable.json(exclude_none=True)},  # type: ignore
+                    "create": {**litellm_modeltable.json(exclude_none=True)},  # type: ignore
+                },
+            )  # type: ignore
 
         _model_id = model_dict.id
 

litellm/proxy/proxy_server.py

@@ -199,6 +199,7 @@ from litellm.proxy.management_endpoints.team_callback_endpoints import (
     router as team_callback_router,
 )
 from litellm.proxy.management_endpoints.team_endpoints import router as team_router
+from litellm.proxy.management_endpoints.team_endpoints import update_team
 from litellm.proxy.management_endpoints.ui_sso import router as ui_sso_router
 from litellm.proxy.management_helpers.audit_logs import create_audit_log_for_update
 from litellm.proxy.openai_files_endpoints.files_endpoints import is_known_model
@@ -6202,6 +6203,94 @@ async def delete_budget(
 #### MODEL MANAGEMENT ####
 
 
+async def _add_model_to_db(
+    model_params: Deployment,
+    user_api_key_dict: UserAPIKeyAuth,
+    prisma_client: PrismaClient,
+):
+    # encrypt litellm params #
+    _litellm_params_dict = model_params.litellm_params.dict(exclude_none=True)
+    _orignal_litellm_model_name = model_params.litellm_params.model
+    for k, v in _litellm_params_dict.items():
+        encrypted_value = encrypt_value_helper(value=v)
+        model_params.litellm_params[k] = encrypted_value
+    _data: dict = {
+        "model_id": model_params.model_info.id,
+        "model_name": model_params.model_name,
+        "litellm_params": model_params.litellm_params.model_dump_json(exclude_none=True),  # type: ignore
+        "model_info": model_params.model_info.model_dump_json(  # type: ignore
+            exclude_none=True
+        ),
+        "created_by": user_api_key_dict.user_id or litellm_proxy_admin_name,
+        "updated_by": user_api_key_dict.user_id or litellm_proxy_admin_name,
+    }
+    if model_params.model_info.id is not None:
+        _data["model_id"] = model_params.model_info.id
+    model_response = await prisma_client.db.litellm_proxymodeltable.create(
+        data=_data  # type: ignore
+    )
+    return model_response
+
+
+async def _add_team_model_to_db(
+    model_params: Deployment,
+    user_api_key_dict: UserAPIKeyAuth,
+    prisma_client: PrismaClient,
+):
+    """
+    If 'team_id' is provided,
+
+    - generate a unique 'model_name' for the model (e.g. 'model_name_{team_id}_{uuid})
+    - store the model in the db with the unique 'model_name'
+    - store a team model alias mapping {"model_name": "model_name_{team_id}_{uuid}"}
+    """
+    _team_id = model_params.model_info.team_id
+    original_model_name = model_params.model_name
+    if _team_id is None:
+        return None
+
+    unique_model_name = f"model_name_{_team_id}_{uuid.uuid4()}"
+
+    model_params.model_name = unique_model_name
+
+    ## CREATE MODEL IN DB ##
+    model_response = await _add_model_to_db(
+        model_params=model_params,
+        user_api_key_dict=user_api_key_dict,
+        prisma_client=prisma_client,
+    )
+
+    ## CREATE MODEL ALIAS IN DB ##
+    await update_team(
+        data=UpdateTeamRequest(
+            team_id=_team_id,
+            model_aliases={original_model_name: unique_model_name},
+        ),
+        user_api_key_dict=user_api_key_dict,
+        http_request=Request(scope={"type": "http"}),
+    )
+
+    return model_response
+
+
+def check_if_team_id_matches_key(
+    team_id: Optional[str], user_api_key_dict: UserAPIKeyAuth
+) -> bool:
+    can_make_call = True
+    if (
+        user_api_key_dict.user_role
+        and user_api_key_dict.user_role == LitellmUserRoles.PROXY_ADMIN
+    ):
+        return True
+    if team_id is None:
+        if user_api_key_dict.user_role != LitellmUserRoles.PROXY_ADMIN:
+            can_make_call = False
+    else:
+        if user_api_key_dict.team_id != team_id:
+            can_make_call = False
+    return can_make_call
+
+
 #### [BETA] - This is a beta endpoint, format might change based on user feedback. - https://github.com/BerriAI/litellm/issues/964
 @router.post(
     "/model/new",
@@ -6217,8 +6306,6 @@ async def add_new_model(
     try:
         import base64
 
-        global prisma_client
-
         if prisma_client is None:
             raise HTTPException(
                 status_code=500,
@@ -6227,6 +6314,14 @@ async def add_new_model(
                 },
             )
 
+        if not check_if_team_id_matches_key(
+            team_id=model_params.model_info.team_id, user_api_key_dict=user_api_key_dict
+        ):
+            raise HTTPException(
+                status_code=403,
+                detail={"error": "Team ID does not match the API key's team ID"},
+            )
+
         model_response = None
         # update DB
         if store_model_in_db is True:
@@ -6234,43 +6329,35 @@ async def add_new_model(
             - store model_list in db
             - store keys separately
             """
-            # encrypt litellm params #
-            _litellm_params_dict = model_params.litellm_params.dict(exclude_none=True)
-            _orignal_litellm_model_name = model_params.litellm_params.model
-            for k, v in _litellm_params_dict.items():
-                encrypted_value = encrypt_value_helper(value=v)
-                model_params.litellm_params[k] = encrypted_value
-            _data: dict = {
-                "model_id": model_params.model_info.id,
-                "model_name": model_params.model_name,
-                "litellm_params": model_params.litellm_params.model_dump_json(exclude_none=True),  # type: ignore
-                "model_info": model_params.model_info.model_dump_json(  # type: ignore
-                    exclude_none=True
-                ),
-                "created_by": user_api_key_dict.user_id or litellm_proxy_admin_name,
-                "updated_by": user_api_key_dict.user_id or litellm_proxy_admin_name,
-            }
-            if model_params.model_info.id is not None:
-                _data["model_id"] = model_params.model_info.id
-            model_response = await prisma_client.db.litellm_proxymodeltable.create(
-                data=_data  # type: ignore
-            )
 
-            await proxy_config.add_deployment(
-                prisma_client=prisma_client, proxy_logging_obj=proxy_logging_obj
-            )
             try:
+                _original_litellm_model_name = model_params.model_name
+                if model_params.model_info.team_id is None:
+                    model_response = await _add_model_to_db(
+                        model_params=model_params,
+                        user_api_key_dict=user_api_key_dict,
+                        prisma_client=prisma_client,
+                    )
+                else:
+                    model_response = await _add_team_model_to_db(
+                        model_params=model_params,
+                        user_api_key_dict=user_api_key_dict,
+                        prisma_client=prisma_client,
+                    )
+                await proxy_config.add_deployment(
+                    prisma_client=prisma_client, proxy_logging_obj=proxy_logging_obj
+                )
                 # don't let failed slack alert block the /model/new response
                 _alerting = general_settings.get("alerting", []) or []
                 if "slack" in _alerting:
                     # send notification - new model added
                     await proxy_logging_obj.slack_alerting_instance.model_added_alert(
                         model_name=model_params.model_name,
-                        litellm_model_name=_orignal_litellm_model_name,
+                        litellm_model_name=_original_litellm_model_name,
                         passed_model_info=model_params.model_info,
                     )
-            except Exception:
-                pass
+            except Exception as e:
+                verbose_proxy_logger.exception(f"Exception in add_new_model: {e}")
 
         else:
             raise HTTPException(

litellm/proxy/utils.py

@@ -1485,7 +1485,8 @@ class PrismaClient:
             elif table_name == "team":
                 if query_type == "find_unique":
                     response = await self.db.litellm_teamtable.find_unique(
-                        where={"team_id": team_id}  # type: ignore
+                        where={"team_id": team_id},  # type: ignore
+                        include={"litellm_model_table": True},  # type: ignore
                     )
                 elif query_type == "find_all" and reset_at is not None:
                     response = await self.db.litellm_teamtable.find_many(

litellm/types/router.py

@@ -103,6 +103,7 @@ class ModelInfo(BaseModel):
         None  # specify if the base model is azure/gpt-3.5-turbo etc for accurate cost tracking
     )
     tier: Optional[Literal["free", "paid"]] = None
+    team_id: Optional[str] = None  # the team id that this model belongs to
 
     def __init__(self, id: Optional[Union[str, int]] = None, **params):
         if id is None:

package-lock.json

@@ -5,6 +5,7 @@
   "packages": {
     "": {
       "dependencies": {
+        "prism-react-renderer": "^2.4.1",
         "prisma": "^5.17.0",
         "react-copy-to-clipboard": "^5.1.0"
       },
@@ -52,6 +53,11 @@
         "@prisma/debug": "5.17.0"
       }
     },
+    "node_modules/@types/prismjs": {
+      "version": "1.26.5",
+      "resolved": "https://registry.npmjs.org/@types/prismjs/-/prismjs-1.26.5.tgz",
+      "integrity": "sha512-AUZTa7hQ2KY5L7AmtSiqxlhWxb4ina0yd8hNbl4TWuqnv/pFP0nDMb3YrfSBf4hJVGLh2YEIBfKaBW/9UEl6IQ=="
+    },
     "node_modules/@types/prop-types": {
       "version": "15.7.12",
       "resolved": "https://registry.npmjs.org/@types/prop-types/-/prop-types-15.7.12.tgz",
@@ -77,6 +83,14 @@
         "@types/react": "*"
       }
     },
+    "node_modules/clsx": {
+      "version": "2.1.1",
+      "resolved": "https://registry.npmjs.org/clsx/-/clsx-2.1.1.tgz",
+      "integrity": "sha512-eYm0QWBtUrBWZWG0d386OGAw16Z995PiOVo2B7bjWSbHedGl5e0ZWaq65kOGgUSNesEIDkB9ISbTg/JK9dhCZA==",
+      "engines": {
+        "node": ">=6"
+      }
+    },
     "node_modules/copy-to-clipboard": {
       "version": "3.3.3",
       "resolved": "https://registry.npmjs.org/copy-to-clipboard/-/copy-to-clipboard-3.3.3.tgz",
@@ -115,6 +129,18 @@
         "node": ">=0.10.0"
       }
     },
+    "node_modules/prism-react-renderer": {
+      "version": "2.4.1",
+      "resolved": "https://registry.npmjs.org/prism-react-renderer/-/prism-react-renderer-2.4.1.tgz",
+      "integrity": "sha512-ey8Ls/+Di31eqzUxC46h8MksNuGx/n0AAC8uKpwFau4RPDYLuE3EXTp8N8G2vX2N7UC/+IXeNUnlWBGGcAG+Ig==",
+      "dependencies": {
+        "@types/prismjs": "^1.26.0",
+        "clsx": "^2.0.0"
+      },
+      "peerDependencies": {
+        "react": ">=16.0.0"
+      }
+    },
     "node_modules/prisma": {
       "version": "5.17.0",
       "resolved": "https://registry.npmjs.org/prisma/-/prisma-5.17.0.tgz",

package.json

@@ -1,5 +1,6 @@
 {
   "dependencies": {
+    "prism-react-renderer": "^2.4.1",
     "prisma": "^5.17.0",
     "react-copy-to-clipboard": "^5.1.0"
   },

tests/local_testing/test_add_update_models.py

@@ -107,6 +107,49 @@ async def test_add_new_model(prisma_client):
     assert _new_model_in_db is not None
 
 
+@pytest.mark.parametrize(
+    "team_id, key_team_id, user_role, expected_result",
+    [
+        ("1234", "1234", LitellmUserRoles.PROXY_ADMIN.value, True),
+        (
+            "1234",
+            "1235",
+            LitellmUserRoles.PROXY_ADMIN.value,
+            True,
+        ),  # proxy admin can add models for any team
+        (None, "1234", LitellmUserRoles.PROXY_ADMIN.value, True),
+        (None, None, LitellmUserRoles.PROXY_ADMIN.value, True),
+        (
+            "1234",
+            "1234",
+            LitellmUserRoles.INTERNAL_USER.value,
+            True,
+        ),  # internal users can add models for their team
+        ("1234", "1235", LitellmUserRoles.INTERNAL_USER.value, False),
+        (None, "1234", LitellmUserRoles.INTERNAL_USER.value, False),
+        (
+            None,
+            None,
+            LitellmUserRoles.INTERNAL_USER.value,
+            False,
+        ),  # internal users cannot add models by default
+    ],
+)
+def test_can_add_model(team_id, key_team_id, user_role, expected_result):
+    from litellm.proxy.proxy_server import check_if_team_id_matches_key
+
+    args = {
+        "team_id": team_id,
+        "user_api_key_dict": UserAPIKeyAuth(
+            user_role=user_role,
+            api_key="sk-1234",
+            team_id=key_team_id,
+        ),
+    }
+
+    assert check_if_team_id_matches_key(**args) is expected_result
+
+
 @pytest.mark.asyncio
 @pytest.mark.skip(reason="new feature, tests passing locally")
 async def test_add_update_model(prisma_client):

tests/local_testing/test_completion_cost.py

@@ -1252,6 +1252,7 @@ def test_completion_cost_anthropic_prompt_caching():
     assert cost_1 > cost_2
 
 
+@pytest.mark.flaky(retries=6, delay=2)
 @pytest.mark.parametrize(
     "model",
     [

tests/local_testing/test_completion_with_retries.py

@@ -11,7 +11,7 @@ sys.path.insert(
 import pytest
 import openai
 import litellm
-from litellm import completion_with_retries, completion
+from litellm import completion_with_retries, completion, acompletion_with_retries
 from litellm import (
     AuthenticationError,
     BadRequestError,
@@ -113,3 +113,36 @@ async def test_completion_with_retry_policy_no_error(sync_mode):
             await completion(**data)
     except Exception as e:
         print(e)
+
+
+@pytest.mark.parametrize("sync_mode", [True, False])
+@pytest.mark.asyncio
+async def test_completion_with_retries(sync_mode):
+    """
+    If completion_with_retries is called with num_retries=3, and max_retries=0, then litellm.completion should receive num_retries , max_retries=0
+    """
+    from unittest.mock import patch, MagicMock, AsyncMock
+
+    if sync_mode:
+        target_function = "completion"
+    else:
+        target_function = "acompletion"
+
+    with patch.object(litellm, target_function) as mock_completion:
+        if sync_mode:
+            completion_with_retries(
+                model="gpt-3.5-turbo",
+                messages=[{"gm": "vibe", "role": "user"}],
+                num_retries=3,
+                original_function=mock_completion,
+            )
+        else:
+            await acompletion_with_retries(
+                model="gpt-3.5-turbo",
+                messages=[{"gm": "vibe", "role": "user"}],
+                num_retries=3,
+                original_function=mock_completion,
+            )
+        mock_completion.assert_called_once()
+        assert mock_completion.call_args.kwargs["num_retries"] == 0
+        assert mock_completion.call_args.kwargs["max_retries"] == 0