Fixes for runpod test

install.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+
+cd /
+git clone https://git.nielsgeens.be/llm/SIA.git
+cd /SIA
+pip3 install -r requirements.txt
+
+curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash
+nvm install node
+cd /SIA/web
+npm install
+npm install -D tailwindcss
+npm run build
+mv /root/SIA/web/dist/ /root/SIA/static
+
+apt update
+apt install -y vim tmux
+vim .env
+
+cd /root/SIA
+python3 -m sia
+
+#The SIA source is located in /SIA. Not all features are implemented yet. Look at the readme and code to find what is missing. Make sure to unit test your work.

sia/local_llm_engine.py

@@ -72,7 +72,7 @@ class LocalLlmEngine(LlmEngine):
             text_inputs=prompt,
             do_sample=True,
             temperature=self._temperature,
-            max_new_tokens=self._token_limit,
+            max_new_tokens=self.token_limit(),
             streamer=streamer
         )
         thread = Thread(target=self._pipeline, kwargs=pipeline_kwargs)
@@ -106,4 +106,7 @@ class LocalLlmEngine(LlmEngine):
         Returns:
             int: Maximum number of tokens the model can process
         """
-        return self._pipeline.model.config.max_position_embeddings
+        if self._token_limit is not None:
+            return self._token_limit
+        else:
+            return self._pipeline.model.config.max_position_embeddings

sia/repeat_entry.py

@@ -99,19 +99,22 @@ class RepeatEntry(Entry):
         element = ET.Element("repeat", {
             "id": self.id,
         })
+        if self._timeout:
+            element.set("timeout", str(self._timeout))
         element.text = self.script
-        if self._exit_code is not None:
-            element.set("exit_code", str(self._exit_code))
         if self._timed_out:
             element.set("timed_out", "true")
-        else:
+        elif self._exit_code is not None:
-            if len(self._stdout) > self._limit:
+            element.set("exit_code", str(self._exit_code))
+            if self._limit:
+                element.set("limit", str(self._limit))
+            if len(self._stdout) > (self._limit or self.default_limit):
                 element.set("stdout_truncated", "true")
             stdout_elem = ET.SubElement(element, "stdout")
-            stdout_elem.text = self._stdout[:self._limit]
+            stdout_elem.text = self._stdout[:(self._limit or self.default_limit)]
-            if len(self._stderr) > self._limit:
+            if len(self._stderr) > (self._limit or self.default_limit):
                 element.set("stderr_truncated", "true")
             stderr_elem = ET.SubElement(element, "stderr")
-            stderr_elem.text = self._stderr[:self._limit]
+            stderr_elem.text = self._stderr[:(self._limit or self.default_limit)]
         
         return element

sia/response_parser.py

@@ -62,8 +62,13 @@ class ResponseParser:
         entry_id = str(uuid.uuid4())
         timestamp = datetime.now()
         
+        parser = ET.XMLPullParser(events=("start", "end"))
+        parser.feed(xml)
+        root = None
         try:
-            root = ET.fromstring(xml.strip())
+            for event, root in parser.read_events():
+                if event == "start":
+                    break
         except ET.ParseError as e:
             return ParseErrorEntry(entry_id, timestamp, xml, f"Invalid XML: {str(e)}")
         

sia/single_entry.py

@@ -43,7 +43,7 @@ class SingleEntry(Entry):
         super().__init__(id, timestamp)
         self._script = script
         self._timeout = timeout or self.default_timeout
-        self._limit = limit or self.default_limit
+        self._limit = limit
         self._stdout = ""
         self._stderr = ""
         self._exit_code: Optional[int] = None
@@ -102,17 +102,21 @@ class SingleEntry(Entry):
         element = ET.Element("single", {
             "id": self.id,
         })
+        if self._timeout:
+            element.set("timeout", str(self._timeout))
         element.text = self.script
         if self._timed_out:
             element.set("timed_out", "true")
         elif self._executed:
             element.set("exit_code", str(self._exit_code))
-            if len(self._stdout) > self._limit:
+            if self._limit:
+                element.set("limit", str(self._limit))
+            if len(self._stdout) > (self._limit or self.default_limit):
                 element.set("stdout_truncated", "true")
             stdout_elem = ET.SubElement(element, "stdout")
-            stdout_elem.text = self._stdout[:self._limit]
+            stdout_elem.text = self._stdout[:(self._limit or self.default_limit)]
-            if len(self._stderr) > self._limit:
+            if len(self._stderr) > (self._limit or self.default_limit):
                 element.set("stderr_truncated", "true")
             stderr_elem = ET.SubElement(element, "stderr")
-            stderr_elem.text = self._stderr[:self._limit]
+            stderr_elem.text = self._stderr[:(self._limit or self.default_limit)]
         return element

web/src/components/App.jsx

@@ -30,8 +30,7 @@ const App = () => {
     response_timeout: 10.0
   });
 
-  const { wsState, sendMessage, addMessageHandler } = useWebSocket('ws://localhost:8080/ws');
+  const { wsState, sendMessage, addMessageHandler } = useWebSocket(`${window.location.protocol === 'https:' ? 'wss:' : 'ws:'}//${window.location.host}/ws`);
-
   useEffect(() => {
     addMessageHandler((message) => {
       switch(message.type) {

web/tailwind.config.js

@@ -1,5 +1,7 @@
+import tailwindcssAnimate from "tailwindcss-animate";
+
 /** @type {import('tailwindcss').Config} */
-module.exports = {
+export default {
   darkMode: ["class"],
   content: ["./src/**/*.{js,jsx}"],
   theme: {
@@ -63,5 +65,5 @@ module.exports = {
       },
     },
   },
-  plugins: [require("tailwindcss-animate")],
+  plugins: [tailwindcssAnimate],
 }