Upload 3 files

README.md

@@ -4,9 +4,7 @@ emoji: 🤖
 colorFrom: blue
 colorTo: purple
 sdk: gradio
-sdk_version: 4.44.0
 app_file: app.py
 pinned: false
 license: apache-2.0
-hardware: zero-gpu
 ---

app.py

@@ -1,37 +1,30 @@
-import spaces
 import torch
 import gradio as gr
 from huggingface_hub import hf_hub_download
 from safetensors.torch import load_file
 
 MODEL_REPO = "ClokAI/ci-base"
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
-@spaces.GPU
-def load_model():
-    from clokai import CiModel, CiConfig
-    from transformers import AutoTokenizer
-    
-    config = CiConfig()
-    model = CiModel(config)
-    
-    weights_path = hf_hub_download(repo_id=MODEL_REPO, filename="model.safetensors")
-    sd = load_file(weights_path)
-    model.load_state_dict(sd, strict=False)
-    model.eval()
-    
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO)
-    return model, tokenizer
+print(f"Loading model on {DEVICE}...")
+
+from clokai import CiModel, CiConfig
+from transformers import AutoTokenizer
 
-print("Loading model...")
-model, tokenizer = load_model()
+config = CiConfig()
+model = CiModel(config).to(DEVICE)
+
+weights_path = hf_hub_download(repo_id=MODEL_REPO, filename="model.safetensors")
+sd = load_file(weights_path)
+model.load_state_dict(sd, strict=False)
+model.eval()
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO)
 print("Model ready!")
 
-@spaces.GPU
 def chat(message, history, temperature, max_tokens):
-    device = next(model.parameters()).device
-    
     formatted = f"<s> User: {message} Assistant:"
-    inputs = tokenizer(formatted, return_tensors="pt").to(device)
+    inputs = tokenizer(formatted, return_tensors="pt").to(DEVICE)
     
     generated = []
     input_ids = inputs["input_ids"]

requirements.txt

@@ -2,6 +2,5 @@ torch>=2.0.0
 transformers>=4.35.0
 safetensors>=0.4.0
 huggingface_hub>=0.20.0
-gradio>=4.44.0
+gradio>=4.0.0
 clokai>=1.1.1
-spaces>=0.30.0