ad correct MODEL PATH

Dockerfile

@@ -2,28 +2,28 @@
 FROM python:3.10-slim
 WORKDIR /app
 
-# 1) Install HF tooling & 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) Create a cache dir and make it world-writable
-RUN mkdir -p /app/model_cache && chmod 777 /app/model_cache
+# 2) Create a world-writable cache directory
+RUN mkdir -p /app/model && chmod 777 /app/model
 
-# 3) Download the model into /app/model_cache
+# 3) Download the entire model into /app/model (build time)
 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"
+    cache_dir="/app",                # places under /app/models--numind--NuExtract-1.5-tiny/…
+    local_dir="model",               # so final: /app/model/models--…/snapshots/<hash>/
+    local_dir_use_symlinks=False     # ensure actual files, no symlinks needed
 )
 EOF
+# 3b) Make sure perms survive into runtime
+RUN chmod -R 755 /app/model
 
-# 3b) Fix permissions on everything we just downloaded
-RUN chmod -R 777 /app/model_cache
-
-# 4) Copy your code
+# 4) Copy your FastAPI code
 COPY . .
 
 ENV PORT=7860

app.py

@@ -15,17 +15,23 @@ from huggingface_hub import snapshot_download
 load_dotenv()
 app = FastAPI()
 
-
+# // FOR RUNNING IN SPACES
 model_name = "numind/NuExtract-1.5-tiny"
-MODEL_CACHE_DIR = "/app/model_cache"
-model_cache_path = snapshot_download(
-    repo_id="numind/NuExtract-1.5-tiny",
-    cache_dir=MODEL_CACHE_DIR
-)
+# Path inside your container
+MODEL_PATH = "/app/model/models--numind--NuExtract-1.5-tiny/snapshots/<commit_hash>"
+# If you used local_dir="model", snapshot_download will still create models--… subfolder.
+# You can also symlink or copy it to /app/model directly in Dockerfile.
+
+# // FOR RUNNING LOCALLY
+# MODEL_CACHE_DIR = "/app/model_cache"
+# model_cache_path = snapshot_download(
+#     repo_id="numind/NuExtract-1.5-tiny",
+#     cache_dir=MODEL_CACHE_DIR
+# )
 
-print(">>> MODEL CACHE PATH:", model_cache_path, os.listdir(model_cache_path))
+print(">>> MODEL CACHE PATH:", MODEL_PATH, os.listdir(MODEL_PATH))
 
-device     = "gpu" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
+device = "gpu" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
 dtype = torch.float16 if device in ("mps", "gpu") else torch.float32
 
 @app.on_event("startup")
@@ -44,14 +50,14 @@ def load_model():
     # model_name, torch_dtype=dtype, trust_remote_code=True
     # )
     model = AutoModelForCausalLM.from_pretrained(
-        model_cache_path,
+        MODEL_PATH,
         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,
+        MODEL_PATH,
         local_files_only=True,
         trust_remote_code=True
     )