fix write access error

Dockerfile

@@ -1,23 +1,26 @@
+# Dockerfile
 FROM python:3.10-slim
 WORKDIR /app
 
-# 1) Install HF tools + your deps
+# 1) Install HF tooling & your deps
 RUN pip install --no-cache-dir huggingface_hub
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-# 2) Snapshot the model into /app/model_cache
+# 2) Create a cache dir and make it world-writable
+RUN mkdir -p /app/model_cache && chmod 777 /app/model_cache
+
+# 3) Download the model into /app/model_cache
 RUN python - <<EOF
 from huggingface_hub import snapshot_download
+# this writes under /app/model_cache/models--numind--NuExtract-1.5-tiny/...
 snapshot_download(
     repo_id="numind/NuExtract-1.5-tiny",
-    cache_dir="/app/model_cache",
-    local_dir="numind/NuExtract-1.5-tiny",
-    local_dir_use_symlinks=False
+    cache_dir="/app/model_cache"
 )
 EOF
 
-# 3) Copy your code
+# 4) Copy your code
 COPY . .
 
 ENV PORT=7860

app.py

@@ -41,17 +41,17 @@ def load_model():
     # model_name, torch_dtype=dtype, trust_remote_code=True
     # )
     model = AutoModelForCausalLM.from_pretrained(
-    model_cache_path,
-    local_files_only=True,
-    torch_dtype=dtype,
-    trust_remote_code=True
+        "/app/model_cache/models--numind--NuExtract-1.5-tiny",  # include the full folder name
+        local_files_only=True,
+        torch_dtype=dtype,
+        trust_remote_code=True
     ).to(device).eval()
     # tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
     tokenizer = AutoTokenizer.from_pretrained(
-    model_cache_path,
-    local_files_only=True,
-    trust_remote_code=True
-)
+        "/app/model_cache/models--numind--NuExtract-1.5-tiny",  # include the full folder name
+        local_files_only=True,
+        trust_remote_code=True
+    )
     print("Model and tokenizer loaded.", flush=True)
 
 def predict_NuExtract(texts, template, batch_size=10, max_length=5096, max_new_tokens=1024):