Removed kv_cache from HPU graph output (#19)

server/text_generation_server/models/causal_lm.py

@@ -34,6 +34,8 @@ from text_generation_server.models.types import (
 from text_generation_server.pb import generate_pb2
 from text_generation_server.utils import HeterogeneousNextTokenChooser, StoppingCriteria, Sampling, make_tokenizer_optional, is_tokenizer_transparent
 from loguru import logger
+from functools import wraps
+
 
 tracer = trace.get_tracer(__name__)
 
@@ -114,6 +116,27 @@ def shift_all(srcs, dim, offsets):
     return [shift(src, dim, offset) for src, offset in zip(srcs, offsets)]
 
 
+def remove_kv_cache_from_output(module):
+    orig_fwd = module.forward
+
+    @wraps(orig_fwd)
+    def forward(*args, **kwargs):
+        if kwargs["past_key_values"] is not None:
+            kwargs["return_dict"] = False
+            output = orig_fwd(*args, **kwargs)
+            first_value, second_value, *_ = output
+            if first_value.nelement() < 2:
+                return second_value
+            else:
+                return first_value
+        else:
+            kwargs["return_dict"] = True
+            return orig_fwd(*args, **kwargs)
+
+    module.forward = forward
+    return module
+
+
 def pad_tensors(tensors, paddings, dim, value):
     for i, (tensor, padding) in enumerate(zip(tensors, paddings)):
         if padding > 0:
@@ -510,7 +533,8 @@ class CausalLM(Model):
             # Initialize the model
             ds_inference_kwargs = {"dtype": dtype}
             ds_inference_kwargs["tensor_parallel"] = {"tp_size": world_size}
-            ds_inference_kwargs["enable_cuda_graph"] = self.enable_hpu_graph
+            ds_inference_kwargs["enable_cuda_graph"] = False
+
 
             if load_to_meta:
                 # model loaded to meta is managed differently
@@ -519,6 +543,10 @@ class CausalLM(Model):
                 ds_inference_kwargs["checkpoint"] = checkpoints_json.name
             model = deepspeed.init_inference(model, **ds_inference_kwargs)
             model = model.module
+            model = remove_kv_cache_from_output(model)
+            if self.enable_hpu_graph:
+                model = wrap_in_hpu_graph(model, disable_tensor_cache=True)
+
         else:
             get_repo_root(model_id)
             model = AutoModelForCausalLM.from_pretrained(
@@ -528,6 +556,7 @@ class CausalLM(Model):
             )
             model = model.eval().to(device)
             #wrap in hpu_graph only if self.enable_hpu_graph is set
+            model = remove_kv_cache_from_output(model)
             if self.enable_hpu_graph:
                 model = wrap_in_hpu_graph(model, disable_tensor_cache=True)
 
@@ -624,6 +653,9 @@ class CausalLM(Model):
             kwargs["bypass_hpu_graphs"] = bypass_hpu_graph
 
         kwargs.update(self.kwargs)
+        if past_key_values is not None:
+            return self.model.forward(**kwargs)
+        else:
             outputs = self.model.forward(**kwargs)
             return outputs.logits, outputs.past_key_values
 
@@ -659,6 +691,7 @@ class CausalLM(Model):
         else:
             input_ids = batch.input_ids
 
+        if prefill:
             logits, past = self.forward(
                 input_ids,
                 attention_mask,
@@ -667,6 +700,15 @@ class CausalLM(Model):
                 batch.past_key_values,
                 bypass_hpu_graph = prefill and self.limit_hpu_graph if self.enable_hpu_graph else None
             )
+        else:
+            logits = self.forward(
+                input_ids,
+                attention_mask,
+                batch.position_ids,
+                token_idx,
+                batch.past_key_values,
+                bypass_hpu_graph = prefill and self.limit_hpu_graph if self.enable_hpu_graph else None
+            )
 
         # Results
         generations: List[Generation] = []
@@ -829,7 +871,9 @@ class CausalLM(Model):
         else:
             batch.position_ids += 1
         # Update past key values
+        if prefill:
             batch.past_key_values = past
+
         if self.hb_profer_started == True:
             self.hb_profer.step()
         htorch.core.mark_step()