Fix llama-cpp-python blocking on HF Spaces due to excessive threads

- Limit effective vCPUs to 2 on HF Spaces free tier (detects 16 but only 2 available)
- Update vCPU detection logic in utils.py with SPACE_ID environment check
- Enhance logging in summarization.py to show detected vs effective vCPUs
- Add HF Spaces limitation indicator in startup messages

src/summarization.py

@@ -11,6 +11,8 @@ but llama_cpp directly for LLM calls (better performance).
 import time
 from functools import lru_cache
 from typing import Iterator
+import os
+import multiprocessing
 
 from llama_cpp import Llama
 from langchain.text_splitter import RecursiveCharacterTextSplitter
@@ -19,7 +21,9 @@ from langchain.prompts import PromptTemplate
 from .utils import available_gguf_llms, num_vcpus, s2tw_converter
 
 # Detection of available logical cores
-print(f"Detected vCPUs: {num_vcpus}")
+detected_cpus = multiprocessing.cpu_count()
+is_hf_spaces = os.environ.get('SPACE_ID') is not None
+print(f"Detected vCPUs: {detected_cpus}, Effective vCPUs: {num_vcpus}" + (" (HF Spaces limited)" if is_hf_spaces else ""))
 
 
 @lru_cache(maxsize=1)

src/utils.py

@@ -7,7 +7,13 @@ from typing import Optional
 import multiprocessing
 
 # Detect logical cores (vCPUs available to the container)
-num_vcpus = multiprocessing.cpu_count()
+# On HF Spaces free tier, cpu_count() reports 16 but only 2 are actually available
+detected_cpus = multiprocessing.cpu_count()
+if os.environ.get('SPACE_ID'):
+    # HF Spaces free tier limitation
+    num_vcpus = min(detected_cpus, 2)
+else:
+    num_vcpus = detected_cpus
 
 model_names = {
     "tiny English":"tiny",