Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+examples/cat_tabby.jpg filter=lfs diff=lfs merge=lfs -text
+examples/city_skyline_night.jpg filter=lfs diff=lfs merge=lfs -text
+examples/sushi_platter.jpg filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,30 @@
 ---
-title: Grug 9b Demo
-emoji: 👀
-colorFrom: pink
-colorTo: gray
+title: Grug-9B Reasoning VLM
+emoji: 🧠
+colorFrom: red
+colorTo: yellow
 sdk: gradio
 sdk_version: 6.20.0
-python_version: '3.12'
 app_file: app.py
-pinned: false
+short_description: 9B reasoning VLM with vision and shortened thinking
+python_version: "3.12"
+startup_duration_timeout: 1h
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Grug-9B Reasoning VLM
+
+A demo for [ProCreations/grug-9b](https://huggingface.co/ProCreations/grug-9b) — a 9B-parameter vision-language reasoning model fine-tuned from Ornith-1.0-9B (Qwen3.5 architecture) to produce shorter internal reasoning while maintaining coding and agent capabilities.
+
+## Features
+
+- **Vision + Language**: Upload an image and ask questions about it.
+- **Reasoning**: The model thinks inside `` tags before producing its answer.
+- **Chat**: Multi-turn conversation with the model.
+- **Code**: Ask the model to write functions, explain concepts, etc.
+
+## Usage
+
+1. Optionally upload an image.
+2. Type your message in the text box.
+3. Press Send or hit Enter.
+4. Adjust advanced settings (max tokens, temperature, thinking mode) as needed.

app.py

@@ -0,0 +1,219 @@
+import os
+
+os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
+
+import spaces  # MUST come before torch / transformers
+import torch
+import gradio as gr
+import numpy as np
+from PIL import Image
+from threading import Thread
+from transformers import (
+    AutoProcessor,
+    Qwen3_5ForConditionalGeneration,
+    TextIteratorStreamer,
+)
+
+MODEL_ID = "ProCreations/grug-9b"
+
+print(f"Loading model {MODEL_ID} ...")
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = Qwen3_5ForConditionalGeneration.from_pretrained(
+    MODEL_ID,
+    torch_dtype=torch.bfloat16,
+).to("cuda").eval()
+print("Model loaded.")
+
+
+@spaces.GPU(duration=120)
+def chat_with_image(
+    image: Image.Image,
+    message: str,
+    history: list,
+    max_new_tokens: int = 1024,
+    temperature: float = 0.7,
+    enable_thinking: bool = True,
+):
+    """Chat with the grug-9b reasoning VLM about an image.
+
+    Args:
+        image: An optional image to discuss with the model.
+        message: The user's text message / question.
+        history: Prior conversation turns (list of [user, assistant] pairs).
+        max_new_tokens: Maximum number of tokens to generate.
+        temperature: Sampling temperature; lower = more deterministic.
+        enable_thinking: Whether the model produces reasoning inside  tags.
+    """
+    if not message or not message.strip():
+        yield history + [(message, "")], "Please enter a message."
+        return
+
+    # Build the conversation
+    messages = []
+    if image is not None:
+        messages.append({
+            "role": "user",
+            "content": [
+                {"type": "image", "image": image},
+                {"type": "text", "text": message},
+            ],
+        })
+    else:
+        messages.append({"role": "user", "content": message})
+
+    # Add conversation history
+    conv_messages = []
+    for user_msg, asst_msg in history:
+        conv_messages.append({"role": "user", "content": user_msg})
+        if asst_msg:
+            conv_messages.append({"role": "assistant", "content": asst_msg})
+    conv_messages.extend(messages)
+
+    # Prepare inputs
+    if image is not None:
+        text = processor.apply_chat_template(
+            conv_messages,
+            tokenize=False,
+            add_generation_prompt=True,
+            enable_thinking=enable_thinking,
+        )
+        inputs = processor(
+            text=[text],
+            images=[image],
+            return_tensors="pt",
+            padding=True,
+        ).to("cuda")
+    else:
+        text = processor.apply_chat_template(
+            conv_messages,
+            tokenize=False,
+            add_generation_prompt=True,
+            enable_thinking=enable_thinking,
+        )
+        inputs = processor(
+            text=[text],
+            return_tensors="pt",
+            padding=True,
+        ).to("cuda")
+
+    # Stream the response
+    streamer = TextIteratorStreamer(
+        processor.tokenizer,
+        skip_prompt=True,
+        skip_special_tokens=True,
+        timeout=120,
+    )
+
+    generation_kwargs = dict(
+        **inputs,
+        streamer=streamer,
+        max_new_tokens=max_new_tokens,
+        temperature=temperature,
+        use_cache=True,
+    )
+
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    thread.start()
+
+    full_response = ""
+    new_history = history + [(message, "")]
+    for token in streamer:
+        full_response += token
+        new_history[-1] = (message, full_response)
+        yield new_history, ""
+
+    thread.join()
+    yield new_history, ""
+
+
+CSS = """
+#col-container { max-width: 1100px; margin: 0 auto; }
+.dark .gradio-container { color: var(--body-text-color); }
+"""
+
+with gr.Blocks(theme=gr.themes.Citrus(), css=CSS) as demo:
+    gr.Markdown(
+        """
+        # 🧠 Grug-9B Reasoning VLM
+        A 9B-parameter vision-language reasoning model fine-tuned from Ornith-1.0-9B (Qwen3.5 architecture)
+        to "think small" — shorter internal reasoning while maintaining coding and agent capabilities.
+
+        Upload an image (optional) and ask a question. The model reasons inside `` tags before answering.
+        """
+    )
+
+    with gr.Column(elem_id="col-container"):
+        with gr.Row():
+            image_input = gr.Image(
+                label="Image (optional)",
+                type="pil",
+                height=350,
+            )
+            with gr.Column(scale=2):
+                chatbot = gr.Chatbot(
+                    label="Chat",
+                    height=450,
+                    show_label=True,
+                )
+                msg_input = gr.Textbox(
+                    label="Message",
+                    placeholder="Ask something about the image, or just chat...",
+                    show_label=False,
+                    lines=2,
+                )
+                with gr.Row():
+                    send_btn = gr.Button("Send", variant="primary", scale=2)
+                    clear_btn = gr.Button("Clear", scale=1)
+
+        with gr.Accordion("Advanced settings", open=False):
+            with gr.Row():
+                max_tokens = gr.Slider(
+                    label="Max new tokens",
+                    minimum=128,
+                    maximum=4096,
+                    value=1024,
+                    step=128,
+                )
+                temperature = gr.Slider(
+                    label="Temperature",
+                    minimum=0.1,
+                    maximum=2.0,
+                    value=0.7,
+                    step=0.1,
+                )
+                thinking_toggle = gr.Checkbox(
+                    label="Enable thinking (reasoning mode)",
+                    value=True,
+                )
+
+        gr.Examples(
+            examples=[
+                [None, "Write a Python function that checks if a number is prime.", 512, 0.7, True],
+                [None, "Explain the difference between TCP and UDP in networking.", 1024, 0.7, True],
+                ["examples/cat_tabby.jpg", "What breed is this cat? Describe what you see in detail.", 1024, 0.7, True],
+                ["examples/city_skyline_night.jpg", "What city might this be? Describe the architectural style.", 1024, 0.7, True],
+                ["examples/sushi_platter.jpg", "What kinds of sushi are on this platter?", 1024, 0.7, True],
+            ],
+            inputs=[image_input, msg_input, max_tokens, temperature, thinking_toggle],
+            outputs=[chatbot, msg_input],
+            fn=chat_with_image,
+            cache_examples=True,
+            cache_mode="lazy",
+        )
+
+    def clear_chat():
+        return [], ""
+
+    clear_btn.click(clear_chat, outputs=[chatbot, msg_input])
+    send_btn.click(
+        chat_with_image,
+        inputs=[image_input, msg_input, chatbot, max_tokens, temperature, thinking_toggle],
+        outputs=[chatbot, msg_input],
+    )
+    msg_input.submit(
+        chat_with_image,
+        inputs=[image_input, msg_input, chatbot, max_tokens, temperature, thinking_toggle],
+        outputs=[chatbot, msg_input],
+    )
+
+demo.launch(mcp_server=True)

requirements.txt

@@ -0,0 +1,8 @@
+transformers==5.13.0
+accelerate
+sentencepiece
+qwen-vl-utils
+einops
+pillow
+numpy
+torchvision