replaced_lambda_function

Author: LiyuanLucasLiu

flash_rl/flash_quantization.py

@@ -115,15 +115,20 @@ def flash_quantize_fp8_tensor(weights, profile):
         else:
             yield (name, tensor)
 
+def flash_noquantize(weights, profile):
+    logger.debug("flash_rl quantization is called")
+    for name, tensor in weights:
+        yield (name, tensor)
+
 quant_fn_map = {
     'int8': flash_quantize,
     'int8_fast': flash_quantize,
     'int8_wo_prune': flash_quantize,
     'int8_prune': flash_quantize_with_prune,
-    'fp8': lambda weights, profile: weights,
-    'fp8_vllm': lambda weights, profile: weights,
-    'fp8_fast': lambda weights, profile: weights,
-    'fp8_vllm_fast': lambda weights, profile: weights,
+    'fp8': flash_noquantize,
+    'fp8_vllm': flash_noquantize,
+    'fp8_fast': flash_noquantize,
+    'fp8_vllm_fast': flash_noquantize,
     'fp8_tensor': flash_quantize_fp8_tensor,
     'fp8_channel': flash_quantize_fp8_channel,
 }