🐛 Critical Bug: Unhandled JSONDecodeError Crashes Entire Model Output Parsing

File: src/opengradient/client/_conversions.py

What's broken

In both convert_to_model_output() (line 159) and convert_array_to_model_output() (line 206), JSON tensor values from the blockchain are parsed with bare json.loads() calls — with no error handling:

# convert_to_model_output — line 159
output_dict[name] = np.array(json.loads(value))  # ← unguarded

# convert_array_to_model_output — line 206  
json_data[name] = np.array(json.loads(value))     # ← unguarded

If any single JSON tensor in the response contains malformed data (encoding corruption, contract bug, partial write), json.loads() raises an unhandled json.JSONDecodeError that:

1. Crashes the entire output parsing — all other valid tensors are lost
2. Provides zero diagnostic context (which tensor failed, what the value was)
3. Surfaces as a raw Python traceback rather than a meaningful SDK error

Fix

Wrap each json.loads() call in a try/except and log a detailed warning instead of crashing:

try:
    output_dict[name] = np.array(json.loads(value))
except (json.JSONDecodeError, TypeError, ValueError) as e:
    logging.warning(f"Failed to parse JSON tensor '{name}': {e} (raw value: {value!r})")

Applied to both conversion functions.

Impact

Files Changed

• src/opengradient/client/_conversions.py