Upload folder using huggingface_hub

README.md

@@ -1,5 +1,5 @@
 ---
-title: GSE Labs Chat
+title: Beexly Studio Chat
 emoji: 🧠
 colorFrom: indigo
 colorTo: purple
@@ -7,34 +7,41 @@ sdk: gradio
 sdk_version: 5.9.1
 app_file: app.py
 pinned: false
-license: apache-2.0
+license: other
+license_name: qwen
+license_link: https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE
 short_description: Chat with an open-weights model on free ZeroGPU
 tags:
   - zero-gpu
 ---
 
-# GSE Labs Chat (ZeroGPU)
+# Beexly Studio Chat (ZeroGPU)
 
-A personal Gradio chat UI running **Qwen2.5-72B-Instruct** (4-bit) on
-Hugging Face's free **ZeroGPU** hardware (time-sliced Nvidia GPU, allocated
-per request). Personal-use tool, not part of the GSE product surface.
+Personal chat tool. **Not connected to any Galaxy Sports Edge product, data,
+or prediction engine — this is Garrett's own tooling, unrelated to GSE.**
 
-No API key, no per-token billing — inference runs on the Space's shared free
-GPU quota instead of a paid hosted API. Responses stream token-by-token, GPU
-time requested per reply scales with how many tokens you ask for
-(`_duration()` in `app.py`), attention uses `sdpa`, sampling uses
-`top_p`/`repetition_penalty` for quality, and chat history is capped at the
-last 12 turns to keep long sessions from exceeding context/GPU limits.
+- **Model**: `Qwen/Qwen2.5-72B-Instruct`, loaded 4-bit (nf4, bitsandbytes)
+- **License**: `qwen` (Alibaba's custom Qwen license, confirmed from the
+  model card metadata — not Apache-2.0/MIT; read the full terms on the
+  [model card](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct) before any
+  commercial or redistribution use, this Space is personal-use only)
+- **Measured limits** (live-tested, see the deploy report): packed footprint
+  44.5GB, ~13.4 tokens/sec including overhead, max 1024 output tokens per
+  reply (calibrated so a full-length reply stays under ZeroGPU's 120s grant
+  with margin — a 1024-token reply measured 84.0s), cold start ~4-5 min
+  (full weight re-download; no persistent storage on the free tier)
+
+## Features
+
+- System-prompt picker: General / Code / Writing / Structured JSON output
+- Streaming responses
+- Copy button on every reply (hover a message)
+- Chat history capped at the last 12 turns to bound context growth
 
 ## Swapping the model
 
 Change `MODEL_ID` in `app.py` to any other instruct-tuned model on the Hub.
 Verified working sizes so far (4-bit, packed footprint): 32B → ~20.7GB,
-72B → 44.5GB, both confirmed loading and generating cleanly. If a bigger
-model OOMs or times out, the previous known-good `MODEL_ID` is
+72B → 44.5GB, both loading and generating cleanly. If a bigger model OOMs
+or times out, the previous known-good `MODEL_ID` is
 `Qwen/Qwen2.5-32B-Instruct`.
-
-Measured 72B throughput: ~13.4 tokens/sec including overhead (800 tokens in
-59.8s, live test). The "Max new tokens" slider is capped at 1024 so a
-full-length request stays inside the 120s ZeroGPU duration ceiling with
-margin — raise it only alongside `_duration()`'s cap, and re-test.

app.py

@@ -13,6 +13,24 @@ from transformers import (
 MODEL_ID = "Qwen/Qwen2.5-72B-Instruct"
 MAX_HISTORY_TURNS = 12
 
+SYSTEM_PROMPTS = {
+    "General": "You are a helpful assistant.",
+    "Code": (
+        "You are an expert software engineer. Answer with correct, working code. "
+        "Prefer minimal, direct solutions. Explain tricky parts briefly, in comments "
+        "only where the reasoning isn't obvious from the code itself."
+    ),
+    "Writing": (
+        "You are a skilled editor and writer. Write clearly and concisely. "
+        "Match the requested tone and length exactly. Avoid filler and cliches."
+    ),
+    "Structured JSON output": (
+        "You output ONLY valid JSON matching what the user asks for — no prose, "
+        "no markdown code fences, no explanation before or after. If the request "
+        "is ambiguous, make a reasonable schema choice and stay consistent."
+    ),
+}
+
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
@@ -26,15 +44,16 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 
 
-def _duration(message, history, system_prompt, max_new_tokens, temperature):
+def _duration(message, history, system_prompt_name, max_new_tokens, temperature):
     return min(120, 40 + int(max_new_tokens) // 10)
 
 
 @spaces.GPU(duration=_duration)
-def generate(message, history, system_prompt, max_new_tokens, temperature):
+def generate(message, history, system_prompt_name, max_new_tokens, temperature):
     history = history[-(MAX_HISTORY_TURNS * 2):]
+    system_prompt = SYSTEM_PROMPTS.get(system_prompt_name, SYSTEM_PROMPTS["General"])
 
-    messages = [{"role": "system", "content": system_prompt}] if system_prompt else []
+    messages = [{"role": "system", "content": system_prompt}]
     for turn in history:
         messages.append({"role": turn["role"], "content": turn["content"]})
     messages.append({"role": "user", "content": message})
@@ -69,16 +88,24 @@ def generate(message, history, system_prompt, max_new_tokens, temperature):
     thread.join()
 
 
+chatbot = gr.Chatbot(type="messages", show_copy_button=True, label="Chat")
+
 demo = gr.ChatInterface(
     fn=generate,
     type="messages",
+    chatbot=chatbot,
     additional_inputs=[
-        gr.Textbox(label="System prompt", value="You are a helpful assistant."),
+        gr.Dropdown(
+            list(SYSTEM_PROMPTS.keys()), value="General", label="System prompt"
+        ),
         gr.Slider(64, 1024, value=512, step=64, label="Max new tokens"),
         gr.Slider(0.0, 1.5, value=0.7, step=0.1, label="Temperature"),
     ],
-    title="GSE Labs Chat (ZeroGPU)",
-    description=f"Personal free-GPU chat — `{MODEL_ID}` (4-bit), streaming, on ZeroGPU.",
+    title="Beexly Studio Chat (ZeroGPU)",
+    description=(
+        f"Personal free-GPU chat — `{MODEL_ID}` (4-bit), streaming, on ZeroGPU. "
+        "Hover any reply for a copy button. Not part of any Galaxy Sports Edge product."
+    ),
 )
 
 if __name__ == "__main__":