+
diff --git a/public/script.js b/public/script.js
index 22ab14db2..2bbc88c9e 100644
--- a/public/script.js
+++ b/public/script.js
@@ -5655,20 +5655,31 @@ function extractMessageFromData(data) {
return data;
}
- switch (main_api) {
- case 'kobold':
- return data.results[0].text;
- case 'koboldhorde':
- return data.text;
- case 'textgenerationwebui':
- return data.choices?.[0]?.text ?? data.content ?? data.response ?? '';
- case 'novel':
- return data.output;
- case 'openai':
- return data?.choices?.[0]?.message?.content ?? data?.choices?.[0]?.text ?? data?.text ?? data?.message?.content?.[0]?.text ?? data?.message?.tool_plan ?? '';
- default:
- return '';
+ function getTextContext() {
+ switch (main_api) {
+ case 'kobold':
+ return data.results[0].text;
+ case 'koboldhorde':
+ return data.text;
+ case 'textgenerationwebui':
+ return data.choices?.[0]?.text ?? data.content ?? data.response ?? '';
+ case 'novel':
+ return data.output;
+ case 'openai':
+ return data?.choices?.[0]?.message?.content ?? data?.choices?.[0]?.text ?? data?.text ?? data?.message?.content?.[0]?.text ?? data?.message?.tool_plan ?? '';
+ default:
+ return '';
+ }
}
+
+ const content = getTextContext();
+
+ if (main_api === 'openai' && oai_settings.chat_completion_source === chat_completion_sources.DEEPSEEK && oai_settings.show_thoughts) {
+ const thoughts = data?.choices?.[0]?.message?.reasoning_content ?? '';
+ return [thoughts, content].filter(x => x).join('\n\n');
+ }
+
+ return content;
}
/**
diff --git a/public/scripts/openai.js b/public/scripts/openai.js
index 4ec0c7749..55a9858af 100644
--- a/public/scripts/openai.js
+++ b/public/scripts/openai.js
@@ -2030,6 +2030,16 @@ async function sendOpenAIRequest(type, messages, signal) {
// https://api-docs.deepseek.com/api/create-chat-completion
if (isDeepSeek) {
generate_data.top_p = generate_data.top_p || Number.EPSILON;
+
+ if (generate_data.model.endsWith('-reasoner')) {
+ delete generate_data.top_p;
+ delete generate_data.temperature;
+ delete generate_data.frequency_penalty;
+ delete generate_data.presence_penalty;
+ delete generate_data.top_logprobs;
+ delete generate_data.logprobs;
+ delete generate_data.logit_bias;
+ }
}
if ((isOAI || isOpenRouter || isMistral || isCustom || isCohere || isNano) && oai_settings.seed >= 0) {
@@ -2085,6 +2095,7 @@ async function sendOpenAIRequest(type, messages, signal) {
let text = '';
const swipes = [];
const toolCalls = [];
+ const state = {};
while (true) {
const { done, value } = await reader.read();
if (done) return;
@@ -2095,9 +2106,9 @@ async function sendOpenAIRequest(type, messages, signal) {
if (Array.isArray(parsed?.choices) && parsed?.choices?.[0]?.index > 0) {
const swipeIndex = parsed.choices[0].index - 1;
- swipes[swipeIndex] = (swipes[swipeIndex] || '') + getStreamingReply(parsed);
+ swipes[swipeIndex] = (swipes[swipeIndex] || '') + getStreamingReply(parsed, state);
} else {
- text += getStreamingReply(parsed);
+ text += getStreamingReply(parsed, state);
}
ToolManager.parseToolCalls(toolCalls, parsed);
@@ -2129,14 +2140,27 @@ async function sendOpenAIRequest(type, messages, signal) {
}
}
-function getStreamingReply(data) {
+/**
+ * Extracts the reply from the response data from a chat completions-like source
+ * @param {object} data Response data from the chat completions-like source
+ * @param {object} state Additional state to keep track of
+ * @returns {string} The reply extracted from the response data
+ */
+function getStreamingReply(data, state) {
if (oai_settings.chat_completion_source === chat_completion_sources.CLAUDE) {
return data?.delta?.text || '';
} else if (oai_settings.chat_completion_source === chat_completion_sources.MAKERSUITE) {
return data?.candidates?.[0]?.content?.parts?.filter(x => oai_settings.show_thoughts || !x.thought)?.map(x => x.text)?.filter(x => x)?.join('\n\n') || '';
} else if (oai_settings.chat_completion_source === chat_completion_sources.COHERE) {
return data?.delta?.message?.content?.text || data?.delta?.message?.tool_plan || '';
- } else {
+ } else if (oai_settings.chat_completion_source === chat_completion_sources.DEEPSEEK) {
+ const hadThoughts = state.hadThoughts;
+ const thoughts = data.choices?.filter(x => oai_settings.show_thoughts || !x?.delta?.reasoning_content)?.[0]?.delta?.reasoning_content || '';
+ const content = data.choices?.[0]?.delta?.content || '';
+ state.hadThoughts = !!thoughts;
+ const separator = hadThoughts && !thoughts ? '\n\n' : '';
+ return [thoughts, separator, content].filter(x => x).join('\n\n');
+ } else {
return data.choices?.[0]?.delta?.content ?? data.choices?.[0]?.message?.content ?? data.choices?.[0]?.text ?? '';
}
}
@@ -4488,7 +4512,7 @@ async function onModelChange() {
if (oai_settings.chat_completion_source === chat_completion_sources.DEEPSEEK) {
if (oai_settings.max_context_unlocked) {
$('#openai_max_context').attr('max', unlocked_max);
- } else if (oai_settings.deepseek_model == 'deepseek-chat') {
+ } else if (['deepseek-reasoner', 'deepseek-chat'].includes(oai_settings.deepseek_model)) {
$('#openai_max_context').attr('max', max_64k);
} else if (oai_settings.deepseek_model == 'deepseek-coder') {
$('#openai_max_context').attr('max', max_16k);
diff --git a/public/scripts/sse-stream.js b/public/scripts/sse-stream.js
index c1c5f1bd6..3921e7d58 100644
--- a/public/scripts/sse-stream.js
+++ b/public/scripts/sse-stream.js
@@ -220,6 +220,21 @@ async function* parseStreamData(json) {
}
return;
}
+ else if (typeof json.choices[0].delta.reasoning_content === 'string' && json.choices[0].delta.reasoning_content.length > 0) {
+ for (let j = 0; j < json.choices[0].delta.reasoning_content.length; j++) {
+ const str = json.choices[0].delta.reasoning_content[j];
+ const isLastSymbol = j === json.choices[0].delta.reasoning_content.length - 1;
+ const choiceClone = structuredClone(json.choices[0]);
+ choiceClone.delta.reasoning_content = str;
+ choiceClone.delta.content = isLastSymbol ? choiceClone.delta.content : '';
+ const choices = [choiceClone];
+ yield {
+ data: { ...json, choices },
+ chunk: str,
+ };
+ }
+ return;
+ }
else if (typeof json.choices[0].delta.content === 'string' && json.choices[0].delta.content.length > 0) {
for (let j = 0; j < json.choices[0].delta.content.length; j++) {
const str = json.choices[0].delta.content[j];
diff --git a/src/endpoints/backends/chat-completions.js b/src/endpoints/backends/chat-completions.js
index 252e16d95..d67f309da 100644
--- a/src/endpoints/backends/chat-completions.js
+++ b/src/endpoints/backends/chat-completions.js
@@ -61,6 +61,7 @@ const API_DEEPSEEK = 'https://api.deepseek.com/beta';
* @returns
*/
function postProcessPrompt(messages, type, names) {
+ const addAssistantPrefix = x => x.length && (x[x.length - 1].role !== 'assistant' || (x[x.length - 1].prefix = true)) ? x : x;
switch (type) {
case 'merge':
case 'claude':
@@ -70,7 +71,9 @@ function postProcessPrompt(messages, type, names) {
case 'strict':
return mergeMessages(messages, names, true, true);
case 'deepseek':
- return (x => x.length && (x[x.length - 1].role !== 'assistant' || (x[x.length - 1].prefix = true)) ? x : x)(mergeMessages(messages, names, true, false));
+ return addAssistantPrefix(mergeMessages(messages, names, true, false));
+ case 'deepseek-reasoner':
+ return addAssistantPrefix(mergeMessages(messages, names, true, true));
default:
return messages;
}
@@ -965,7 +968,8 @@ router.post('/generate', jsonParser, function (request, response) {
bodyParams['logprobs'] = true;
}
- request.body.messages = postProcessPrompt(request.body.messages, 'deepseek', getPromptNames(request));
+ const postProcessType = String(request.body.model).endsWith('-reasoner') ? 'deepseek-reasoner' : 'deepseek';
+ request.body.messages = postProcessPrompt(request.body.messages, postProcessType, getPromptNames(request));
} else {
console.log('This chat completion source is not supported yet.');
return response.status(400).send({ error: true });