Upload folder using huggingface_hub

README.md

@@ -1,13 +1,14 @@
 ---
 title: Kimi K3 Chat
-emoji: 🏃
-colorFrom: blue
-colorTo: indigo
+emoji: 🌙
+colorFrom: indigo
+colorTo: blue
 sdk: gradio
 sdk_version: 6.20.0
-python_version: '3.13'
 app_file: app.py
-pinned: false
+short_description: Chat with Moonshot AI's Kimi K3 via Inference Providers
+python_version: "3.12"
+hf_oauth: true
+hf_oauth_scopes:
+  - inference-api
 ---
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,100 @@
+import base64
+import mimetypes
+
+import gradio as gr
+from huggingface_hub import InferenceClient
+
+MODEL = "moonshotai/Kimi-K3"
+PROVIDER = "together"
+
+SYSTEM_PROMPT = (
+    "You are Kimi K3, Moonshot AI's open-weight native multimodal agentic model. "
+    "Be helpful, concise, and accurate."
+)
+
+
+def to_data_url(path: str) -> str:
+    mime = mimetypes.guess_type(path)[0] or "image/png"
+    with open(path, "rb") as f:
+        b64 = base64.b64encode(f.read()).decode()
+    return f"data:{mime};base64,{b64}"
+
+
+def respond(message, history, token: gr.OAuthToken | None):
+    """Chat with Kimi K3 (Moonshot AI) via HF Inference Providers. Supports text and images."""
+    if token is None:
+        yield "Please sign in with your Hugging Face account (sidebar) to chat — inference is billed to your own account."
+        return
+
+    client = InferenceClient(api_key=token.token, provider=PROVIDER, timeout=600)
+
+    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
+    for msg in history:
+        if msg["role"] == "assistant" and isinstance(msg.get("content"), str):
+            # strip rendered reasoning before sending back
+            content = msg["content"]
+            if "</think>" in content:
+                content = content.split("</think>", 1)[1]
+            messages.append({"role": "assistant", "content": content.strip()})
+        elif msg["role"] == "user":
+            messages.append({"role": "user", "content": msg["content"]})
+
+    content = []
+    for path in message.get("files", []):
+        content.append({"type": "image_url", "image_url": {"url": to_data_url(path)}})
+    if message.get("text"):
+        content.append({"type": "text", "text": message["text"]})
+    messages.append({"role": "user", "content": content if len(content) > 1 else (message.get("text") or "")})
+
+    stream = client.chat.completions.create(
+        model=MODEL,
+        messages=messages,
+        max_tokens=8192,
+        stream=True,
+    )
+
+    reasoning, answer, in_think = "", "", False
+    for chunk in stream:
+        if not chunk.choices:
+            continue
+        delta = chunk.choices[0].delta
+        r = getattr(delta, "reasoning_content", None)
+        if r:
+            reasoning += r
+        if delta.content:
+            answer += delta.content
+        out = ""
+        if reasoning:
+            out += f"<think>{reasoning}</think>"
+        out += answer
+        yield out
+
+
+with gr.Blocks(title="Kimi K3 Chat", fill_height=True) as demo:
+    with gr.Sidebar():
+        gr.LoginButton("Sign in with Hugging Face")
+        gr.Markdown(
+            "**Kimi K3** — Moonshot AI's open 3T-class multimodal agentic model "
+            "(2.8T params, 104B active MoE, 1M context).\n\n"
+            "Served via [HF Inference Providers](https://huggingface.co/moonshotai/Kimi-K3) "
+            "(Together AI). Sign-in required; usage is billed to your own HF account.\n\n"
+            "🖼️ Attach images to try multimodal chat."
+        )
+    gr.ChatInterface(
+        fn=respond,
+        multimodal=True,
+        chatbot=gr.Chatbot(reasoning_tags=[("<think>", "</think>")], scale=1),
+        textbox=gr.MultimodalTextbox(
+            placeholder="Message Kimi K3 — attach an image for multimodal chat…",
+            file_types=["image"],
+            sources=["upload", "clipboard"],
+        ),
+        examples=[
+            {"text": "Explain mixture-of-experts routing like I'm five."},
+            {"text": "Write a Python function that streams tokens from an SSE endpoint."},
+            {"text": "What are the tradeoffs of MXFP4 quantization for a 3T-param model?"},
+        ],
+        cache_examples=False,
+    )
+
+demo.launch(mcp_server=True)