[fix] fix param for qkv linear, gpt2fused linear; fix requirments;

Author: duanjunwen

colossalai/shardformer/layer/__init__.py

@@ -21,7 +21,7 @@
     "LinearWithGradAccum",
     "Linear1D_Col",
     "Linear1D_Row",
-    "GPT2FusedLinearConv1D_Col",
+    "GPT2FusedLinearConv",
     "GPT2FusedLinearConv1D_Row",
     "GPT2FusedLinearConv1D_Col",
     "DropoutForParallelInput",

colossalai/shardformer/layer/qkv_fused_linear.py

@@ -832,7 +832,6 @@ def __init__(
         bias_: Optional[Parameter] = None,
         weight_initializer: Callable = init.kaiming_uniform_(a=math.sqrt(5)),
         bias_initializer: Callable = init.xavier_uniform_(a=1, scale=1),
-        fp8_communication: bool = False,
         use_zbv: bool = False,
     ):
         super().__init__()
@@ -846,7 +845,6 @@ def __init__(
         self.device = device
         self.split_sizes = split_sizes
         self.process_group = process_group
-        self.fp8_communication = fp8_communication
         self.use_zbv = use_zbv
 
         assert (
@@ -1246,7 +1244,6 @@ def __init__(
         bias_: Optional[Parameter] = None,
         weight_initializer: Callable = init.kaiming_uniform_(a=math.sqrt(5)),
         bias_initializer: Callable = init.xavier_uniform_(a=1, scale=1),
-        fp8_communication: bool = False,
         use_zbv: bool = False,
     ):
         super().__init__()
@@ -1257,7 +1254,6 @@ def __init__(
         self.seq_parallel_dim = seq_parallel_dim
         self.skip_bias_add = skip_bias_add
         self.device = device
-        self.fp8_communication = fp8_communication
         self.use_zbv = use_zbv
 
         if skip_bias_add and not bias:

requirements/requirements.txt

@@ -24,5 +24,3 @@ fastapi
 uvicorn==0.29.0
 galore_torch
 diffusers==0.29.0
-pyramid<=1.10.7
-zope<=5.5.2