biondizzle
02b8ea536f
Server running on B200 port 8000 with full NVFP4→vLLM bridge. All critical bugs fixed: DeepGEMM scale format, compressor shapes, block scale values.
42 lines
1.4 KiB
Python
42 lines
1.4 KiB
Python
#!/usr/bin/python3
|
|
"""Fix the broken docstrings in deepseek_v4.py"""
|
|
|
|
filepath = "/root/nvidia-meeting/deepseek-v4-quant/patches/deepseek_v4.py"
|
|
|
|
with open(filepath, 'r') as f:
|
|
content = f.read()
|
|
|
|
# Replace the problematic triple-quoted docstrings with comments
|
|
content = content.replace(
|
|
' def _convert_nvfp4_attention_to_fp8(self):\n """Convert NVFP4 attention weights to FP8 format."""',
|
|
' def _convert_nvfp4_attention_to_fp8(self): # Convert NVFP4 attention weights to FP8'
|
|
)
|
|
|
|
content = content.replace(
|
|
' def _convert_nvfp4_module_to_fp8(self, mod, e2m1_lut, fp8_max):\n """Convert a single NVFP4 Linear module to FP8 format."""',
|
|
' def _convert_nvfp4_module_to_fp8(self, mod, e2m1_lut, fp8_max): # Convert single NVFP4 module to FP8'
|
|
)
|
|
|
|
# Also remove any stray docstring lines
|
|
lines = content.split('\n')
|
|
new_lines = []
|
|
skip_next = False
|
|
for i, line in enumerate(lines):
|
|
stripped = line.strip()
|
|
if stripped == '"""' or stripped == "'''":
|
|
continue # Skip standalone triple-quote lines
|
|
if '"""Convert NVFP4' in line or '"""Convert a single' in line:
|
|
continue # Skip the one-line docstrings that might remain
|
|
new_lines.append(line)
|
|
|
|
with open(filepath, 'w') as f:
|
|
f.write('\n'.join(new_lines))
|
|
|
|
# Verify syntax
|
|
import ast
|
|
try:
|
|
ast.parse('\n'.join(new_lines))
|
|
print("Syntax OK")
|
|
except SyntaxError as e:
|
|
print(f"Syntax error: {e}")
|