Initial

app.py

@@ -1,44 +1,75 @@
 # Standard library imports
 import os
+import threading
 
-# Related third-party imports
+# Third-party imports
 import gradio as gr
-from huggingface_hub import InferenceClient
+from peft import PeftModel
+from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
 
-client = InferenceClient("bunyaminergen/Qwen2.5-Coder-1.5B-Instruct-Reasoning", token=os.getenv("HF_ACCESS_TOKEN"))
+HF_TOKEN = os.getenv("HF_TOKEN")
+
+tokenizer = AutoTokenizer.from_pretrained(
+    "bunyaminergen/Qwen2.5-Coder-1.5B-Instruct-Reasoning",
+    use_auth_token=HF_TOKEN,
+    trust_remote_code=True
+)
+
+base_model = AutoModelForCausalLM.from_pretrained(
+    "Qwen/Qwen2.5-Coder-1.5B-Instruct",
+    device_map="auto",
+    torch_dtype="auto",
+    use_auth_token=HF_TOKEN
+)
+model = PeftModel.from_pretrained(
+    base_model,
+    "bunyaminergen/Qwen2.5-Coder-1.5B-Instruct-Reasoning",
+    use_auth_token=HF_TOKEN
+)
+model.eval()
 
 
 def respond(
-        message,
+        message: str,
         history: list[tuple[str, str]],
-        system_message,
-        max_tokens,
-        temperature,
-        top_p,
+        system_message: str,
+        max_tokens: int,
+        temperature: float,
+        top_p: float,
 ):
     messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
+    for u, a in history:
+        if u:
+            messages.append({"role": "user", "content": u})
+        if a:
+            messages.append({"role": "assistant", "content": a})
     messages.append({"role": "user", "content": message})
 
-    response = ""
+    prompt = tokenizer.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
 
-    for message in client.chat_completion(
-            messages,
-            max_tokens=max_tokens,
-            stream=True,
-            temperature=temperature,
-            top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    streamer = TextIteratorStreamer(
+        tokenizer,
+        timeout=10.0,
+        skip_prompt=True,
+        skip_special_tokens=True
+    )
+    generation_kwargs = {
+        **inputs,
+        "max_new_tokens": max_tokens,
+        "temperature": temperature,
+        "top_p": top_p,
+        "streamer": streamer,
+    }
+    thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
+    thread.start()
 
-        response += token
-        yield response
+    output = ""
+    for chunk in streamer:
+        output += chunk
+        yield output
 
 
 demo = gr.ChatInterface(
@@ -47,13 +78,7 @@ demo = gr.ChatInterface(
         gr.Textbox(value="You are a helpful coding assistant.", label="System message"),
         gr.Slider(minimum=512, maximum=8192, value=2048, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
     ],
 )