fix startup when torch missing

README.md

@@ -1,12 +1,27 @@
 ---
-title: Kimi Coder Demo
-emoji: 🦀
-colorFrom: red
-colorTo: purple
+title: kimi-coder-135m
+emoji: 🤖
+colorFrom: blue
+colorTo: cyan
 sdk: gradio
-sdk_version: 6.11.0
+sdk_version: "4.44.0"
+python_version: "3.10"
 app_file: app.py
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# kimi-coder-135m Demo
+
+**SmolLM2-135M** fine-tuned on ~15k coding samples distilled from [KIMI-K2.5-700000x](https://huggingface.co/datasets/ianncity/KIMI-K2.5-700000x).
+
+## Model
+
+→ [yava-code/kimi-coder-135m](https://huggingface.co/yava-code/kimi-coder-135m)
+
+## Training details
+
+- Base: `HuggingFaceTB/SmolLM2-135M-Instruct`
+- Data: 15k coding Q&A pairs (Python, JS, C++, etc.)
+- CoT stripped from answers — model learns direct responses
+- 1 epoch, lr=2e-4, cosine schedule, bf16
+- Hardware: Colab T4 (~2h)

app.py

@@ -1,19 +1,34 @@
 import gradio as gr
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
 from threading import Thread
 
-MODEL = "yava-code/kimi-coder-135m"  
+err = None
+try:
+    import torch
+    from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
+except ModuleNotFoundError as e:
+    err = e
 
-tok = AutoTokenizer.from_pretrained(MODEL)
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL,
-    torch_dtype=torch.bfloat16,
-    device_map="auto",
-)
-model.eval()
+MODEL = "yava-code/kimi-coder-135m"
+tok = None
+model = None
+
+if err is None:
+    tok = AutoTokenizer.from_pretrained(MODEL)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL,
+        torch_dtype=torch.bfloat16,
+        device_map="auto",
+    )
+    model.eval()
 
 def respond(msg, history, max_new, temp):
+    if err is not None:
+        yield (
+            f"missing dependency: `{err.name}`\n\n"
+            "add it to requirements.txt and rebuild the space."
+        )
+        return
+
     chat = []
     for u, a in history:
         chat += [{"role": "user", "content": u}, {"role": "assistant", "content": a}]
@@ -50,14 +65,22 @@ SmolLM2-135M fine-tuned on 15k coding samples distilled from KIMI-K2.5.
 Model: [yava-code/kimi-coder-135m](https://huggingface.co/yava-code/kimi-coder-135m)
         """
     )
-    chatbot = gr.ChatInterface(
-        respond,
-        additional_inputs=[
-            gr.Slider(64, 1024, value=512, label="Max new tokens"),
-            gr.Slider(0, 1, value=0.3, step=0.05, label="Temperature"),
-        ],
-        examples=EXAMPLES,
-        title="",
-    )
+    if err is not None:
+        gr.Markdown(
+            f"### startup warning\n"
+            f"missing dependency: `{err.name}`\n\n"
+            f"current requirements include `torch`, so this usually means the build failed.\n"
+            f"try restarting/rebuilding the space."
+        )
+    else:
+        chatbot = gr.ChatInterface(
+            respond,
+            additional_inputs=[
+                gr.Slider(64, 1024, value=512, label="Max new tokens"),
+                gr.Slider(0, 1, value=0.3, step=0.05, label="Temperature"),
+            ],
+            examples=EXAMPLES,
+            title="",
+        )
 
 demo.launch()

requirements.txt

@@ -0,0 +1,4 @@
+transformers>=4.40
+accelerate
+gradio
+torch