fix merge

Signed-off-by: Kyle Sayers <[email protected]>

Author: kylesayrs

src/compressed_tensors/quantization/lifecycle/initialize.py

@@ -27,7 +27,6 @@
     FP8_E4M3_DATA,
     ActivationOrdering,
     DynamicType,
-    KVCacheScaleType,
     QuantizationArgs,
     QuantizationMetadata,
     QuantizationScheme,
@@ -42,15 +41,11 @@
     disable_hf_hook,
     get_execution_device,
     get_head_dim,
-<<<<<<< HEAD
     get_num_attn_heads,
     get_num_kv_heads,
-=======
->>>>>>> 05ec17e (WIP)
     register_offload_parameter,
 )
 from torch.nn import Module, Parameter
-from transformers import PretrainedConfig
 
 
 __all__ = [
@@ -294,7 +289,6 @@ def initialize_attn_qparams(
     kv_cache: Optional[QuantizedKVCache] = getattr(module, KV_CACHE_ATTR, None)
 
     if impl is None and kv_cache is None:
-<<<<<<< HEAD
         raise ValueError("Attention module has quantization scheme but no attached")
 
     _validate_attention_scheme(scheme)
@@ -310,27 +304,12 @@ def initialize_attn_qparams(
     kv_observed_shape = (num_kv_heads, None, head_dim)
     observed_dtype = next(module.parameters()).dtype
 
-=======
-        raise ValueError("Attention module has quantization scheme but no attached ")
-
-    config: PretrainedConfig = getattr(impl, "config", None) or getattr(
-        kv_cache, "config", None
-    )
-    head_dim = get_head_dim(config)
-    observed_shape = (head_dim,)  # (batch_size, num_attention_heads, slen, head_dim)
-    observed_dtype = next(module.parameters()).dtype
-
->>>>>>> 05ec17e (WIP)
     if impl is not None:
         initialize_qparams(
             module,
             "q",
             scheme.input_activations,
-<<<<<<< HEAD
             observed_shape=q_observed_shape,
-=======
-            observed_shape=observed_shape,
->>>>>>> 05ec17e (WIP)
             observed_dtype=observed_dtype,
             force_zero_point=force_zero_point,
         )
@@ -340,19 +319,14 @@ def initialize_attn_qparams(
             module,
             "k",
             scheme.input_activations,
-<<<<<<< HEAD
             observed_shape=kv_observed_shape,
-=======
-            observed_shape=observed_shape,
->>>>>>> 05ec17e (WIP)
             observed_dtype=observed_dtype,
             force_zero_point=force_zero_point,
         )
         initialize_qparams(
             module,
             "v",
             scheme.input_activations,
-<<<<<<< HEAD
             observed_shape=kv_observed_shape,
             observed_dtype=observed_dtype,
             force_zero_point=force_zero_point,
@@ -373,9 +347,3 @@ def _validate_attention_scheme(scheme: QuantizationScheme):
 
     if scheme.output_activations is not None:
         raise ValueError("Cannot apply output quantization to attention")
-=======
-            observed_shape=observed_shape,
-            observed_dtype=observed_dtype,
-            force_zero_point=force_zero_point,
-        )
->>>>>>> 05ec17e (WIP)