Correct Gradio compatibility metadata and requirements

README.md

@@ -4,7 +4,7 @@ emoji: 🧪
 colorFrom: blue
 colorTo: indigo
 sdk: gradio
-sdk_version: 6.25.0
+sdk_version: 5.50.0
 app_file: app.py
 short_description: Load and test HF causal LLMs on ZeroGPU
 python_version: "3.10"

requirements.txt

@@ -1,41 +1,4 @@
----
-title: Dynamic LLM ZeroGPU Playground
-emoji: 🧪
-colorFrom: blue
-colorTo: indigo
-sdk: gradio
-sdk_version: 5.50.0
-app_file: app.py
-short_description: Load and test HF causal LLMs on ZeroGPU
-python_version: "3.10"
-startup_duration_timeout: 1h
----
-
-# Dynamic LLM ZeroGPU Playground
-
-A small, general-purpose playground for testing standard Hugging Face causal
-language models by model ID.
-
-## How it works
-
-1. Enter a model ID such as `Qwen/Qwen2.5-0.5B-Instruct`.
-2. Click **Download** to fetch the snapshot into the Space's CPU-side cache.
-3. Click **Load** to put that model on the ZeroGPU worker.
-4. Chat with the model, or use **Unload** before switching models.
-5. Use **Delete from disk** to unload first, then remove all cached revisions for the selected ID.
-
-Only one model is kept active by the runtime. Switching models releases the
-previous model with `del`, `gc.collect()`, and `torch.cuda.empty_cache()` before
-the new model is loaded. Chat templates are used whenever the tokenizer
-provides `apply_chat_template()`.
-
-This MVP intentionally targets standard `transformers` +
-`AutoModelForCausalLM` checkpoints. The loader is isolated in
-`TransformersCausalLMRuntime`, so AWQ, GPTQ, or FP8 backends can be added later.
-
-## Notes
-
-- Model downloads happen on CPU and are never triggered by the chat handler.
-- A model must be downloaded before it can be loaded or used.
-- Large models may exceed ZeroGPU memory or take a long time to load.
-- Remote model code is disabled in this first version for safety and stability.
+transformers==4.57.6
+accelerate>=0.34.0
+safetensors>=0.4.3
+sentencepiece>=0.2.0