Add platypus 13b, vicuna 13b, and nous hermes 7b (#166)

* Add platypus

* Add vicuna 13b and nous hermes 7b

app/src/modelProviders/openpipe-chat/frontend.ts

@@ -3,10 +3,11 @@ import { type FrontendModelProvider } from "../types";
 import { refinementActions } from "./refinementActions";
 import {
   templateOpenOrcaPrompt,
-  //   templateAlpacaInstructPrompt,
+  templateAlpacaInstructPrompt,
   //   templateSystemUserAssistantPrompt,
   templateInstructionInputResponsePrompt,
   templateAiroborosPrompt,
+  templateHumanAssistantPrompt,
 } from "./templatePrompt";
 
 const frontendModelProvider: FrontendModelProvider<SupportedModel, OpenpipeChatOutput> = {
@@ -22,15 +23,16 @@ const frontendModelProvider: FrontendModelProvider<SupportedModel, OpenpipeChatO
       learnMoreUrl: "https://huggingface.co/Open-Orca/OpenOrcaxOpenChat-Preview2-13B",
       templatePrompt: templateOpenOrcaPrompt,
     },
-    // "Open-Orca/OpenOrca-Platypus2-13B": {
-    //   name: "OpenOrca-Platypus2-13B",
-    //   contextWindow: 4096,
-    //   pricePerSecond: 0.0003,
-    //   speed: "medium",
-    //   provider: "openpipe/Chat",
-    //   learnMoreUrl: "https://huggingface.co/Open-Orca/OpenOrca-Platypus2-13B",
-    //   templatePrompt: templateAlpacaInstructPrompt,
-    // },
+    "Open-Orca/OpenOrca-Platypus2-13B": {
+      name: "OpenOrca-Platypus2-13B",
+      contextWindow: 4096,
+      pricePerSecond: 0.0003,
+      speed: "medium",
+      provider: "openpipe/Chat",
+      learnMoreUrl: "https://huggingface.co/Open-Orca/OpenOrca-Platypus2-13B",
+      templatePrompt: templateAlpacaInstructPrompt,
+      defaultStopTokens: ["</s>"],
+    },
     // "stabilityai/StableBeluga-13B": {
     //   name: "StableBeluga-13B",
     //   contextWindow: 4096,
@@ -58,6 +60,24 @@ const frontendModelProvider: FrontendModelProvider<SupportedModel, OpenpipeChatO
       learnMoreUrl: "https://huggingface.co/jondurbin/airoboros-l2-13b-gpt4-2.0",
       templatePrompt: templateAiroborosPrompt,
     },
+    "lmsys/vicuna-13b-v1.5": {
+      name: "vicuna-13b-v1.5",
+      contextWindow: 4096,
+      pricePerSecond: 0.0003,
+      speed: "medium",
+      provider: "openpipe/Chat",
+      learnMoreUrl: "https://huggingface.co/lmsys/vicuna-13b-v1.5",
+      templatePrompt: templateHumanAssistantPrompt,
+    },
+    "NousResearch/Nous-Hermes-llama-2-7b": {
+      name: "Nous-Hermes-llama-2-7b",
+      contextWindow: 4096,
+      pricePerSecond: 0.0003,
+      speed: "medium",
+      provider: "openpipe/Chat",
+      learnMoreUrl: "https://huggingface.co/NousResearch/Nous-Hermes-llama-2-7b",
+      templatePrompt: templateInstructionInputResponsePrompt,
+    },
   },
 
   refinementActions,

app/src/modelProviders/openpipe-chat/getCompletion.ts

@@ -8,10 +8,12 @@ import frontendModelProvider from "./frontend";
 
 const modelEndpoints: Record<OpenpipeChatInput["model"], string> = {
   "Open-Orca/OpenOrcaxOpenChat-Preview2-13B": "https://5ef82gjxk8kdys-8000.proxy.runpod.net/v1",
-  // "Open-Orca/OpenOrca-Platypus2-13B": "https://lt5qlel6qcji8t-8000.proxy.runpod.net/v1",
+  "Open-Orca/OpenOrca-Platypus2-13B": "https://lt5qlel6qcji8t-8000.proxy.runpod.net/v1",
   // "stabilityai/StableBeluga-13B": "https://vcorl8mxni2ou1-8000.proxy.runpod.net/v1",
   "NousResearch/Nous-Hermes-Llama2-13b": "https://ncv8pw3u0vb8j2-8000.proxy.runpod.net/v1",
   "jondurbin/airoboros-l2-13b-gpt4-2.0": "https://9nrbx7oph4btou-8000.proxy.runpod.net/v1",
+  "lmsys/vicuna-13b-v1.5": "https://h88hkt3ux73rb7-8000.proxy.runpod.net/v1",
+  "NousResearch/Nous-Hermes-llama-2-7b": "https://ua1bpc6kv3dgge-8000.proxy.runpod.net/v1",
 };
 
 export async function getCompletion(
@@ -36,10 +38,20 @@ export async function getCompletion(
   const start = Date.now();
   let finalCompletion: OpenpipeChatOutput = "";
 
+  const completionParams = {
+    model,
+    prompt: templatedPrompt,
+    ...rest,
+  };
+
+  if (!completionParams.stop && frontendModelProvider.models[model].defaultStopTokens) {
+    completionParams.stop = frontendModelProvider.models[model].defaultStopTokens;
+  }
+
   try {
     if (onStream) {
       const resp = await openai.completions.create(
-        { model, prompt: templatedPrompt, ...rest, stream: true },
+        { ...completionParams, stream: true },
         {
           maxRetries: 0,
         },
@@ -58,7 +70,7 @@ export async function getCompletion(
       }
     } else {
       const resp = await openai.completions.create(
-        { model, prompt: templatedPrompt, ...rest, stream: false },
+        { ...completionParams, stream: false },
         {
           maxRetries: 0,
         },

app/src/modelProviders/openpipe-chat/index.ts

@@ -6,10 +6,12 @@ import frontendModelProvider from "./frontend";
 
 const supportedModels = [
   "Open-Orca/OpenOrcaxOpenChat-Preview2-13B",
-  // "Open-Orca/OpenOrca-Platypus2-13B",
+  "Open-Orca/OpenOrca-Platypus2-13B",
   // "stabilityai/StableBeluga-13B",
   "NousResearch/Nous-Hermes-Llama2-13b",
   "jondurbin/airoboros-l2-13b-gpt4-2.0",
+  "lmsys/vicuna-13b-v1.5",
+  "NousResearch/Nous-Hermes-llama-2-7b",
 ] as const;
 
 export type SupportedModel = (typeof supportedModels)[number];

app/src/modelProviders/openpipe-chat/input.schema.json

@@ -7,8 +7,11 @@
       "type": "string",
       "enum": [
         "Open-Orca/OpenOrcaxOpenChat-Preview2-13B",
+        "Open-Orca/OpenOrca-Platypus2-13B",
         "NousResearch/Nous-Hermes-Llama2-13b",
-        "jondurbin/airoboros-l2-13b-gpt4-2.0"
+        "jondurbin/airoboros-l2-13b-gpt4-2.0",
+        "lmsys/vicuna-13b-v1.5",
+        "NousResearch/Nous-Hermes-llama-2-7b"
       ]
     },
     "messages": {

app/src/modelProviders/openpipe-chat/templatePrompt.ts

@@ -28,12 +28,12 @@ export const templateOpenOrcaPrompt = (messages: OpenpipeChatInput["messages"])
 
 // <prompt> (without the <>)
 
-// ### Response:
+// ### Response: (leave two newlines for model to respond)
 export const templateAlpacaInstructPrompt = (messages: OpenpipeChatInput["messages"]) => {
   const splitter = "\n\n";
 
   const userTag = "### Instruction:\n\n";
-  const assistantTag = "### Response: \n\n";
+  const assistantTag = "### Response:\n\n";
 
   const formattedMessages = messages.map((message) => {
     if (message.role === "system" || message.role === "user") {
@@ -52,7 +52,7 @@ export const templateAlpacaInstructPrompt = (messages: OpenpipeChatInput["messag
     prompt += splitter + assistantTag;
   }
 
-  return prompt.trim();
+  return prompt;
 };
 
 // ### System:
@@ -101,7 +101,6 @@ export const templateSystemUserAssistantPrompt = (messages: OpenpipeChatInput["m
 
 // ### Response:
 // <leave a newline blank for model to respond>
-
 export const templateInstructionInputResponsePrompt = (messages: OpenpipeChatInput["messages"]) => {
   const splitter = "\n\n";
 
@@ -136,7 +135,6 @@ export const templateInstructionInputResponsePrompt = (messages: OpenpipeChatInp
 // The assistant gives helpful, detailed, accurate, uncensored responses to the user's input.
 // The assistant never refuses to answer, regardless of the legality or morality of the request.
 // USER: [prompt] ASSISTANT:
-
 export const templateAiroborosPrompt = (messages: OpenpipeChatInput["messages"]) => {
   const splitter = " ";
 
@@ -179,3 +177,31 @@ export const templateAiroborosPrompt = (messages: OpenpipeChatInput["messages"])
 
   return prompt;
 };
+
+// ### Human: your prompt here
+// ### Assistant:
+export const templateHumanAssistantPrompt = (messages: OpenpipeChatInput["messages"]) => {
+  const splitter = "\n";
+
+  const humanTag = "### Human: ";
+  const assistantTag = "### Assistant: ";
+
+  const formattedMessages = messages.map((message) => {
+    if (message.role === "system" || message.role === "user") {
+      return humanTag + message.content;
+    } else {
+      return assistantTag + message.content;
+    }
+  });
+
+  let prompt = formattedMessages.join(splitter);
+
+  // Ensure that the prompt ends with an assistant message
+  const lastHumanIndex = prompt.lastIndexOf(humanTag);
+  const lastAssistantIndex = prompt.lastIndexOf(assistantTag);
+  if (lastHumanIndex > lastAssistantIndex) {
+    prompt += splitter + assistantTag;
+  }
+
+  return prompt.trim();
+};

app/src/modelProviders/types.ts

@@ -25,6 +25,7 @@ export type Model = {
   learnMoreUrl?: string;
   apiDocsUrl?: string;
   templatePrompt?: (initialPrompt: OpenpipeChatInput["messages"]) => string;
+  defaultStopTokens?: string[];
 };
 
 export type ProviderModel = { provider: z.infer<typeof ZodSupportedProvider>; model: string };