Removing dead variables.

server/text_generation_server/models/bloom.py

@@ -72,7 +72,6 @@ class BLOOMSharded(BLOOM):
         quantize: Optional[str] = None,
     ):
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16

server/text_generation_server/models/flash_llama.py

@@ -29,7 +29,6 @@ tracer = trace.get_tracer(__name__)
 
 class FlashLlama(FlashCausalLM):
     def __init__(self, model_id: str, revision: Optional[str] = None, quantize=False):
-        self.past_pad = None
         if torch.cuda.is_available():
             device = torch.device("cuda")
             dtype = torch.float16
@@ -150,9 +149,7 @@ class FlashLlamaSharded(FlashLlama):
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
     ):
-        self.past_pad = None
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16

server/text_generation_server/models/flash_neox.py

@@ -33,9 +33,7 @@ class FlashNeoXSharded(FlashNeoX):
     def __init__(
         self, model_id: str, revision: Optional[str] = None, quantize: bool = False
     ):
-        self.past_pad = None
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16
@@ -152,4 +150,4 @@ class FlashNeoXSharded(FlashNeoX):
                     else:
                         module._buffers[param_name] = tensor
 
-        model.post_load_weights(quantize)
+        # model.post_load_weights(quantize)

server/text_generation_server/models/flash_santacoder.py

@@ -28,7 +28,6 @@ tracer = trace.get_tracer(__name__)
 
 class FlashSantacoder(FlashCausalLM):
     def __init__(self, model_id: str, revision: Optional[str] = None, quantize=False):
-        self.past_pad = None
         if torch.cuda.is_available():
             device = torch.device("cuda")
             dtype = torch.float16
@@ -173,9 +172,7 @@ class FlashSantacoderSharded(FlashSantacoder):
     def __init__(
         self, model_id: str, revision: Optional[str] = None, quantize: bool = False
     ):
-        self.past_pad = None
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16

server/text_generation_server/models/galactica.py

@@ -199,7 +199,6 @@ class GalacticaSharded(Galactica):
         quantize: Optional[str] = None,
     ):
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16

server/text_generation_server/models/gpt_neox.py

@@ -38,7 +38,6 @@ class GPTNeoxSharded(CausalLM):
         quantize: Optional[str] = None,
     ):
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16

server/text_generation_server/models/opt.py

@@ -51,7 +51,6 @@ class OPTSharded(OPT):
         self, model_id: str, revision: Optional[str] = None, quantize: bool = False
     ):
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16

server/text_generation_server/models/t5.py

@@ -38,7 +38,6 @@ class T5Sharded(Seq2SeqLM):
         quantize: Optional[str] = None,
     ):
         self.process_group, rank, world_size = initialize_torch_distributed()
-        self.master = rank == 0
         if torch.cuda.is_available():
             device = torch.device(f"cuda:{rank}")
             dtype = torch.float16