diff --git a/examples/backends/bitblas_int4_demo.py b/examples/backends/bitblas_int4_demo.py
index 24571fd..3593b0a 100644
--- a/examples/backends/bitblas_int4_demo.py
+++ b/examples/backends/bitblas_int4_demo.py
@@ -20,18 +20,18 @@
 
 #Quantize
 #all 4-bit
-quant_config = BaseQuantizeConfig(nbits=4, group_size=64, quant_scale=False, quant_zero=False, axis=1)
+quant_config = BaseQuantizeConfig(nbits=4, group_size=64, axis=1)
 
 #Mixed 4-bit (bitblas) / 2-bit (ATEN)
 # quant_config = {
-#     "self_attn.q_proj": BaseQuantizeConfig(nbits=2, group_size=32, quant_scale=False, quant_zero=False, axis=0),
-#     "self_attn.k_proj": BaseQuantizeConfig(nbits=2, group_size=32, quant_scale=False, quant_zero=False, axis=0),
-#     "self_attn.v_proj": BaseQuantizeConfig(nbits=2, group_size=32, quant_scale=False, quant_zero=False, axis=0),
-#     "self_attn.o_proj": BaseQuantizeConfig(nbits=2, group_size=32, quant_scale=False, quant_zero=False, axis=0),
-
-#     "mlp.gate_proj": BaseQuantizeConfig(nbits=4, group_size=64, quant_scale=False, quant_zero=False, axis=1),
-#     "mlp.up_proj":   BaseQuantizeConfig(nbits=4, group_size=64, quant_scale=False, quant_zero=False, axis=1),
-#     "mlp.down_proj": BaseQuantizeConfig(nbits=4, group_size=64, quant_scale=False, quant_zero=False, axis=1),
+#     "self_attn.q_proj": BaseQuantizeConfig(nbits=2, group_size=32, axis=0),
+#     "self_attn.k_proj": BaseQuantizeConfig(nbits=2, group_size=32, axis=0),
+#     "self_attn.v_proj": BaseQuantizeConfig(nbits=2, group_size=32, axis=0),
+#     "self_attn.o_proj": BaseQuantizeConfig(nbits=2, group_size=32, axis=0),
+
+#     "mlp.gate_proj": BaseQuantizeConfig(nbits=4, group_size=64, axis=1),
+#     "mlp.up_proj":   BaseQuantizeConfig(nbits=4, group_size=64, axis=1),
+#     "mlp.down_proj": BaseQuantizeConfig(nbits=4, group_size=64, axis=1),
 # }
 # HQQLinear.set_backend(HQQBackend.ATEN)
 
diff --git a/examples/backends/marlin_int4_demo.py b/examples/backends/marlin_int4_demo.py
index ce7a49f..39efacd 100755
--- a/examples/backends/marlin_int4_demo.py
+++ b/examples/backends/marlin_int4_demo.py
@@ -19,7 +19,7 @@
 model        = AutoModelForCausalLM.from_pretrained(model_id, cache_dir=cache_path, torch_dtype=compute_dtype, attn_implementation="sdpa")
 
 #Quantize
-quant_config = BaseQuantizeConfig(nbits=4, group_size=None, quant_scale=False, quant_zero=False, axis=1)
+quant_config = BaseQuantizeConfig(nbits=4, group_size=None, axis=1)
 AutoHQQHFModel.setup_model(model)
 AutoHQQHFModel.quantize_model(model, quant_config=quant_config, compute_dtype=compute_dtype, device=device)
 HQQLinear.set_backend(HQQBackend.PYTORCH)
diff --git a/examples/backends/quantize_and_run.py b/examples/backends/quantize_and_run.py
index 7dcb15d..25cfe47 100644
--- a/examples/backends/quantize_and_run.py
+++ b/examples/backends/quantize_and_run.py
@@ -18,7 +18,7 @@
 model         = AutoModelForCausalLM.from_pretrained(model_id, cache_dir=cache_dir, torch_dtype=compute_dtype, attn_implementation="sdpa")
 tokenizer     = AutoTokenizer.from_pretrained(model_id, cache_dir=cache_dir)
 
-quant_config = BaseQuantizeConfig(nbits=4, group_size=64, quant_scale=False, quant_zero=False, axis=1)
+quant_config = BaseQuantizeConfig(nbits=4, group_size=64, axis=1)
 AutoHQQHFModel.quantize_model(model, quant_config=quant_config, device=device, compute_dtype=compute_dtype)
 
 #Use optimized inference kernels
diff --git a/examples/backends/torchao_int4_demo.py b/examples/backends/torchao_int4_demo.py
index 54e343b..62be2e7 100755
--- a/examples/backends/torchao_int4_demo.py
+++ b/examples/backends/torchao_int4_demo.py
@@ -20,7 +20,7 @@
 model        = AutoModelForCausalLM.from_pretrained(model_id, cache_dir=cache_path, torch_dtype=compute_dtype, attn_implementation="sdpa")
 
 #Quantize
-quant_config = BaseQuantizeConfig(nbits=4, group_size=64, quant_scale=False, quant_zero=False, axis=1)
+quant_config = BaseQuantizeConfig(nbits=4, group_size=64, axis=1)
 AutoHQQHFModel.setup_model(model)
 AutoHQQHFModel.quantize_model(model, quant_config=quant_config, compute_dtype=compute_dtype, device=device)
 HQQLinear.set_backend(HQQBackend.PYTORCH)
diff --git a/examples/backends/transformers_demo.py b/examples/backends/transformers_demo.py
new file mode 100644
index 0000000..65c4953
--- /dev/null
+++ b/examples/backends/transformers_demo.py
@@ -0,0 +1,36 @@
+#Works with multi-gpu as well, tested with BitBlas
+
+import torch, gc
+from transformers import AutoModelForCausalLM, AutoTokenizer, HqqConfig
+
+device    = 'auto'
+dtype     = torch.float16
+model_id  = 'meta-llama/Meta-Llama-3-8B-Instruct'
+cache_dir = '.' 
+
+quant_config  = HqqConfig(nbits=4, group_size=64, axis=1)
+
+model = AutoModelForCausalLM.from_pretrained(
+    model_id, 
+    torch_dtype=dtype, 
+    cache_dir=cache_dir,
+    device_map=device, 
+    quantization_config=quant_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, cache_dir=cache_dir)
+
+#Patching
+from hqq.utils.patching import *
+from hqq.core.quantize import *
+HQQLinear.set_backend(HQQBackend.PYTORCH)
+prepare_for_inference(model, backend='bitblas', verbose=True) #Takes a while
+
+#Import custom HF generator
+from hqq.utils.generation_hf import HFGenerator
+
+#Generate
+gen = HFGenerator(model, tokenizer, max_new_tokens=1000, do_sample=True, compile=None) #Quick test - slower inference
+#gen = HFGenerator(model, tokenizer, max_new_tokens=1000, do_sample=True, compile="partial").warmup() #Takes a while - fastest
+
+out = gen.generate("Write an essay about large language models.", print_tokens=True)
diff --git a/examples/llama2_benchmark/quant_llama2_hqq_demo.py b/examples/llama2_benchmark/quant_llama2_hqq_demo.py
index a5db11d..5c7be24 100755
--- a/examples/llama2_benchmark/quant_llama2_hqq_demo.py
+++ b/examples/llama2_benchmark/quant_llama2_hqq_demo.py
@@ -18,11 +18,11 @@
 ######################################################################################
 from hqq.core.quantize import *
 
-#quant_config = BaseQuantizeConfig(nbits=8, group_size=128)
-quant_config = BaseQuantizeConfig(nbits=4, group_size=64)
-#quant_config = BaseQuantizeConfig(nbits=3, group_size=64)
-#quant_config = BaseQuantizeConfig(nbits=2, group_size=16)
-#quant_config = BaseQuantizeConfig(nbits=2, group_size=16, quant_scale=True) #scale is quantized to 8-bit/g=128
+#quant_config = BaseQuantizeConfig(nbits=8, group_size=128, axis=0)
+quant_config = BaseQuantizeConfig(nbits=4, group_size=64, axis=0)
+#quant_config = BaseQuantizeConfig(nbits=3, group_size=64, axis=0)
+#quant_config = BaseQuantizeConfig(nbits=2, group_size=16, axis=0)
+#quant_config = BaseQuantizeConfig(nbits=2, group_size=16, quant_scale=True, axis=0) #scale is quantized to 8-bit/g=128
 
 model.quantize_model(quant_config=quant_config)
 
diff --git a/examples/vllm/llama2_example.py b/examples/vllm/llama2_example.py
deleted file mode 100755
index e5a29a4..0000000
--- a/examples/vllm/llama2_example.py
+++ /dev/null
@@ -1,25 +0,0 @@
-model_id = 'meta-llama/Llama-2-7b-chat-hf'
-
-#Load VLLM un-quantized model 
-from hqq.engine.vllm import HQQLLM
-model = HQQLLM(model=model_id)
-
-#Quantize the model
-from hqq.core.quantize import *
-model.quantize_model(BaseQuantizeConfig(nbits=4, group_size=64))
-
-#Optional: Save the model
-#model.save_quantized(model_id.split('/')[-1] + '_quantized')
-
-#Optional: Set backend
-HQQLinear.set_backend(HQQBackend.PYTORCH_COMPILE) #set backend
-
-#Generation
-from vllm.entrypoints.llm import SamplingParams
-sampling_params = SamplingParams(temperature=0.6, top_p=0.90, max_tokens=1000, repetition_penalty=1.2)
-
-prompt = "How can I build a car?"
-
-output = model.generate([prompt], sampling_params)[0]
-print(output.prompt)
-print(output.outputs[0].text)
\ No newline at end of file