fix(amazon_deepseek_transformation.py): remove </think> from stream o… (#8717)

* fix(amazon_deepseek_transformation.py): remove </think> from stream output - cleanup user facing stream

* fix(key_managenet_endpoints.py): return `/key/list` sorted by created_at

makes it easier to see created key

* style: cleanup team table

* feat(key_edit_view.tsx): support setting model specific tpm/rpm limits on keys

litellm/llms/bedrock/chat/invoke_transformations/amazon_deepseek_transformation.py

@@ -2,6 +2,7 @@ from typing import Any, List, Optional, cast
 
 from httpx import Response
 
+from litellm import verbose_logger
 from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
     _parse_content_for_reasoning,
 )
@@ -93,7 +94,12 @@ class AmazonDeepseekR1ResponseIterator(BaseModelResponseIterator):
         """
         try:
             typed_chunk = AmazonDeepSeekR1StreamingResponse(**chunk)  # type: ignore
-            if "</think>" in typed_chunk["generation"]:
+            generated_content = typed_chunk["generation"]
+            if generated_content == "</think>" and not self.has_finished_thinking:
+                verbose_logger.debug(
+                    "Deepseek r1: </think> received, setting has_finished_thinking to True"
+                )
+                generated_content = ""
                 self.has_finished_thinking = True
 
             prompt_token_count = typed_chunk.get("prompt_token_count") or 0
@@ -110,12 +116,12 @@ class AmazonDeepseekR1ResponseIterator(BaseModelResponseIterator):
                         finish_reason=typed_chunk["stop_reason"],
                         delta=Delta(
                             content=(
-                                typed_chunk["generation"]
+                                generated_content
                                 if self.has_finished_thinking
                                 else None
                             ),
                             reasoning_content=(
-                                typed_chunk["generation"]
+                                generated_content
                                 if not self.has_finished_thinking
                                 else None
                             ),
@@ -124,5 +130,6 @@ class AmazonDeepseekR1ResponseIterator(BaseModelResponseIterator):
                 ],
                 usage=usage,
             )
+
         except Exception as e:
             raise e

litellm/proxy/management_endpoints/key_management_endpoints.py

@@ -1962,6 +1962,10 @@ async def _list_key_helper(
         where=where,  # type: ignore
         skip=skip,  # type: ignore
         take=size,  # type: ignore
+        order=[
+            {"created_at": "desc"},
+            {"token": "desc"},  # fallback sort
+        ],
     )
 
     verbose_proxy_logger.debug(f"Fetched {len(keys)} keys")

ui/litellm-dashboard/src/components/key_edit_view.tsx

@@ -133,11 +133,23 @@ export function KeyEditView({
       </Form.Item>
 
       <Form.Item label="TPM Limit" name="tpm_limit">
-        <InputNumber style={{ width: "100%" }} />
+        <InputNumber style={{ width: "100%" }} min={0}/>
       </Form.Item>
 
       <Form.Item label="RPM Limit" name="rpm_limit">
-        <InputNumber style={{ width: "100%" }} />
+        <InputNumber style={{ width: "100%" }} min={0}/>
+      </Form.Item>
+
+      <Form.Item label="Max Parallel Requests" name="max_parallel_requests">  
+        <InputNumber style={{ width: "100%" }} min={0}/>
+      </Form.Item>
+
+      <Form.Item label="Model TPM Limit" name="model_tpm_limit">
+        <Input.TextArea rows={4}  placeholder='{"gpt-4": 100, "claude-v1": 200}'/>
+      </Form.Item>
+
+      <Form.Item label="Model RPM Limit" name="model_rpm_limit">
+        <Input.TextArea rows={4}  placeholder='{"gpt-4": 100, "claude-v1": 200}'/>
       </Form.Item>
 
       <Form.Item label="Guardrails" name="guardrails">

ui/litellm-dashboard/src/components/key_info_view.tsx

@@ -332,6 +332,9 @@ export default function KeyInfoView({ keyId, onClose, keyData, accessToken, user
                     <Text className="font-medium">Rate Limits</Text>
                     <Text>TPM: {keyData.tpm_limit !== null ? keyData.tpm_limit : "Unlimited"}</Text>
                     <Text>RPM: {keyData.rpm_limit !== null ? keyData.rpm_limit : "Unlimited"}</Text>
+                    <Text>Max Parallel Requests: {keyData.max_parallel_requests !== null ? keyData.max_parallel_requests : "Unlimited"}</Text>
+                    <Text>Model TPM Limits: {keyData.metadata?.model_tpm_limit ? JSON.stringify(keyData.metadata.model_tpm_limit) : "Unlimited"}</Text>
+                    <Text>Model RPM Limits: {keyData.metadata?.model_rpm_limit ? JSON.stringify(keyData.metadata.model_rpm_limit) : "Unlimited"}</Text>
                   </div>
 
                   <div>

ui/litellm-dashboard/src/components/networking.tsx

@@ -2476,6 +2476,25 @@ export const keyUpdateCall = async (
   try {
     console.log("Form Values in keyUpdateCall:", formValues); // Log the form values before making the API call
 
+    if (formValues.model_tpm_limit) {
+      console.log("formValues.model_tpm_limit:", formValues.model_tpm_limit);
+      // if there's an exception JSON.parse, show it in the message
+      try {
+        formValues.model_tpm_limit = JSON.parse(formValues.model_tpm_limit);
+      } catch (error) {
+        throw new Error("Failed to parse model_tpm_limit: " + error);
+      }
+    }
+
+    if (formValues.model_rpm_limit) {
+      console.log("formValues.model_rpm_limit:", formValues.model_rpm_limit);
+      // if there's an exception JSON.parse, show it in the message
+      try {
+        formValues.model_rpm_limit = JSON.parse(formValues.model_rpm_limit);
+      } catch (error) {
+        throw new Error("Failed to parse model_rpm_limit: " + error);
+      }
+    }
     const url = proxyBaseUrl ? `${proxyBaseUrl}/key/update` : `/key/update`;
     const response = await fetch(url, {
       method: "POST",

ui/litellm-dashboard/src/components/teams.tsx

@@ -389,7 +389,6 @@ const Teams: React.FC<TeamProps> = ({
                         >
                           {team["team_alias"]}
                         </TableCell>
-                        <TableRow>
                         <TableCell>
                           <div className="overflow-hidden">
                             <Tooltip title={team.team_id}>
@@ -408,8 +407,6 @@ const Teams: React.FC<TeamProps> = ({
                             </Tooltip>
                           </div>
                         </TableCell>
-                      </TableRow>
-
                         <TableCell
                           style={{
                             maxWidth: "4px",