Fixing exl2 and other quanize tests again.

server/Makefile-exllamav2

@@ -1,7 +1,7 @@
-exllamav2_commit := v0.1.8
+exllamav2_commit := 872386c89eaebe0bde5b245a890f1da9522768b3
 
 build-exllamav2:
-	git clone https://github.com/turboderp/exllamav2.git exllamav2 && \
+	git clone https://github.com/Narsil/exllamav2.git exllamav2 && \
 	cd exllamav2 && git fetch && git checkout $(exllamav2_commit)  && \
 	git submodule update --init --recursive && \
 	pip install -r requirements.txt && \

server/text_generation_server/models/causal_lm.py

@@ -652,6 +652,7 @@ class CausalLM(Model):
             dtype=dtype,
             device=device,
         )
+        self.quantize = quantize
         return self
 
     @property

server/text_generation_server/models/mamba.py

@@ -412,6 +412,7 @@ class Mamba(Model):
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
+        self.quantize = quantize
         self.process_group, _rank, world_size = initialize_torch_distributed()
         if world_size > 1:
             raise RuntimeError("Mamba does not support Tensor Parallelism (TP)")

server/text_generation_server/models/seq2seq_lm.py

@@ -676,6 +676,7 @@ class Seq2SeqLM(Model):
             dtype=dtype,
             device=device,
         )
+        self.quantize = quantize
         return self
 
     @property