From a402753539205675cd49b01b5fb2b55adf08e1c5 Mon Sep 17 00:00:00 2001
From: Sweaterdog <Sweaterdog5475@gmail.com>
Date: Fri, 28 Feb 2025 12:11:30 -0800
Subject: [PATCH] Update glhf.js

Fixed reasoning models in glhf.js
---
 src/models/glhf.js | 71 ++++++++++++++++++++++++++--------------------
 1 file changed, 40 insertions(+), 31 deletions(-)

diff --git a/src/models/glhf.js b/src/models/glhf.js
index 9c27799..9655603 100644
--- a/src/models/glhf.js
+++ b/src/models/glhf.js
@@ -1,18 +1,14 @@
+// glhf-no-logger.js
 import OpenAIApi from 'openai';
 import { getKey } from '../utils/keys.js';
 
-// glhf doesn't supply an SDK for their models, but fully supports OpenAI SDKs
 export class glhf {
     constructor(model_name, url) {
         this.model_name = model_name;
-
-        // Retrieve the API key from keys.json
         const apiKey = getKey('GHLF_API_KEY');
         if (!apiKey) {
             throw new Error('API key not found. Please check keys.json and ensure GHLF_API_KEY is defined.');
         }
-
-        // Configure OpenAIApi with the retrieved API key and base URL
         this.openai = new OpenAIApi({
             apiKey,
             baseURL: url || "https://glhf.chat/api/openai/v1"
@@ -20,43 +16,56 @@ export class glhf {
     }
 
     async sendRequest(turns, systemMessage, stop_seq = '***') {
-        // Construct the message array for the API request
-        let messages = [{ 'role': 'system', 'content': systemMessage }].concat(turns);
-
+        // Construct the message array for the API request.
+        let messages = [{ role: 'system', content: systemMessage }].concat(turns);
         const pack = {
             model: this.model_name || "hf:meta-llama/Llama-3.1-405B-Instruct",
             messages,
             stop: [stop_seq]
         };
 
-        let res = null;
-        try {
-            console.log('Awaiting glhf.chat API response...');
-            // Uncomment the line below if you need to debug the messages
-            // console.log('Messages:', messages);
+        const maxAttempts = 5;
+        let attempt = 0;
+        let finalRes = null;
 
-            let completion = await this.openai.chat.completions.create(pack);
-            if (completion.choices[0].finish_reason === 'length') {
-                throw new Error('Context length exceeded');
-            }
-
-            console.log('Received.');
-            res = completion.choices[0].message.content;
-        } catch (err) {
-            if ((err.message === 'Context length exceeded' || err.code === 'context_length_exceeded') && turns.length > 1) {
-                console.log('Context length exceeded, trying again with shorter context.');
-                return await this.sendRequest(turns.slice(1), systemMessage, stop_seq);
-            } else {
-                console.log(err);
-                res = 'My brain disconnected, try again.';
+        while (attempt < maxAttempts) {
+            attempt++;
+            console.log(`Awaiting glhf.chat API response... (attempt: ${attempt})`);
+            try {
+                let completion = await this.openai.chat.completions.create(pack);
+                if (completion.choices[0].finish_reason === 'length') {
+                    throw new Error('Context length exceeded');
+                }
+                let res = completion.choices[0].message.content;
+                // If there's an open <think> tag without a corresponding </think>, retry.
+                if (res.includes("<think>") && !res.includes("</think>")) {
+                    console.warn("Partial <think> block detected. Re-generating...");
+                    continue;
+                }
+                // If there's a closing </think> tag but no opening <think>, prepend one.
+                if (res.includes("</think>") && !res.includes("<think>")) {
+                    res = "<think>" + res;
+                }
+                finalRes = res.replace(/<\|separator\|>/g, '*no response*');
+                break; // Valid response obtained.
+            } catch (err) {
+                if ((err.message === 'Context length exceeded' || err.code === 'context_length_exceeded') && turns.length > 1) {
+                    console.log('Context length exceeded, trying again with shorter context.');
+                    return await this.sendRequest(turns.slice(1), systemMessage, stop_seq);
+                } else {
+                    console.error(err);
+                    finalRes = 'My brain disconnected, try again.';
+                    break;
+                }
             }
         }
-
-        // Replace special tokens in the response
-        return res.replace(/<\|separator\|>/g, '*no response*');
+        if (finalRes === null) {
+            finalRes = "I thought too hard, sorry, try again";
+        }
+        return finalRes;
     }
 
     async embed(text) {
         throw new Error('Embeddings are not supported by glhf.');
     }
-}
\ No newline at end of file
+}