Merge pull request #254 from lucyknada/main

add draft_gpu_split option for spec decoding

backends/exllamav2/model.py

@@ -89,7 +89,8 @@ class ExllamaV2Container:
     generation_config: Optional[GenerationConfig] = None
 
     # GPU split vars
-    gpu_split: Optional[list] = None
+    gpu_split: List[float] = []
+    draft_gpu_split: List[float] = []
     gpu_split_auto: bool = True
     autosplit_reserve: List[float] = [96 * 1024**2]
     use_tp: bool = False
@@ -180,6 +181,7 @@ async def create(cls, model_directory: pathlib.Path, quiet=False, **kwargs):
             )
             draft_model_path = draft_model_path / draft_model_name
 
+            self.draft_gpu_split = draft_args.get("draft_gpu_split")
             self.draft_model_dir = draft_model_path
             self.draft_config.model_dir = str(draft_model_path.resolve())
             self.draft_config.prepare()
@@ -232,6 +234,15 @@ async def create(cls, model_directory: pathlib.Path, quiet=False, **kwargs):
                     for value in autosplit_reserve_megabytes
                 ]
 
+            # Change the GPU device list only if gpu_split's list is too small
+            # This allows for an uneven list specification
+            if self.draft_gpu_split and len(self.draft_gpu_split) > len(self.gpu_split):
+                gpu_device_list = [
+                    device_idx
+                    for device_idx, memory in enumerate(self.draft_gpu_split)
+                    if memory > 0
+                ]
+
         # Hardcode max output length to 16
         self.config.max_output_len = 16
 
@@ -375,6 +386,7 @@ async def create(cls, model_directory: pathlib.Path, quiet=False, **kwargs):
             # Set draft cache mode
             self.draft_cache_mode = unwrap(draft_args.get("draft_cache_mode"), "FP16")
 
+            # Edit the draft config size
             if chunk_size:
                 self.draft_config.max_input_len = chunk_size
                 self.draft_config.max_attention_size = chunk_size**2
@@ -619,21 +631,41 @@ def progress(loaded_modules: int, total_modules: int)
 
             # Draft uses the autosplit loader, so create a cache that reflects this
             draft_cache_class = self.get_cache_class(self.draft_cache_mode)
-            self.draft_cache = self.create_cache(
-                cache_class=draft_cache_class,
-                autosplit=True,
-                use_tp=False,
-                model=self.draft_model,
-            )
 
-            for value in self.draft_model.load_autosplit_gen(
-                self.draft_cache,
-                reserve_vram=autosplit_reserve,
-                last_id_only=True,
-                callback_gen=progress_callback,
-            ):
-                if value:
-                    yield value
+            if self.draft_gpu_split:
+                logger.info("Loading with a manual GPU split (or a one GPU setup)")
+
+                for value in self.draft_model.load_gen(
+                    self.draft_gpu_split,
+                    callback_gen=progress_callback,
+                ):
+                    if value:
+                        yield value
+
+                self.draft_cache = self.create_cache(
+                    cache_class=draft_cache_class,
+                    autosplit=False,
+                    use_tp=False,
+                    model=self.draft_model,
+                )
+            else:
+                logger.info("Loading with autosplit")
+
+                self.draft_cache = self.create_cache(
+                    cache_class=draft_cache_class,
+                    autosplit=True,
+                    use_tp=False,
+                    model=self.draft_model,
+                )
+
+                for value in self.draft_model.load_autosplit_gen(
+                    self.draft_cache,
+                    reserve_vram=autosplit_reserve,
+                    last_id_only=True,
+                    callback_gen=progress_callback,
+                ):
+                    if value:
+                        yield value
 
             # Test VRAM allocation with a full-length forward pass
             input_ids = torch.zeros((1, self.config.max_input_len), dtype=torch.long)

common/config_models.py

@@ -351,6 +351,13 @@ class DraftModelConfig(BaseConfigModel):
             f"Possible values: {str(CACHE_SIZES)[15:-1]}."
         ),
     )
+    draft_gpu_split: List[float] = Field(
+        default_factory=list,
+        description=(
+            "An integer array of GBs of VRAM to split between GPUs (default: []).\n"
+            "If this isn't filled in, the draft model is autosplit."
+        ),
+    )
 
 
 class LoraInstanceModel(BaseConfigModel):

config_sample.yml

@@ -20,7 +20,7 @@ network:
   # Turn on this option if you are ONLY connecting from localhost.
   disable_auth: false
 
-  # Disable fetching external content in response to requests, such as images from URLs.
+  # Disable fetching external content in response to requests,such as images from URLs.
   disable_fetch_requests: false
 
   # Send tracebacks over the API (default: False).
@@ -166,6 +166,10 @@ draft_model:
   # Possible values: 'FP16', 'Q8', 'Q6', 'Q4'.
   draft_cache_mode: FP16
 
+  # An integer array of GBs of VRAM to split between GPUs (default: []).
+  # If this isn't filled in, the draft model is autosplit.
+  draft_gpu_split: []
+
 # Options for Loras
 lora:
   # Directory to look for LoRAs (default: loras).