Better compatibility for OGA

onnx · Jan 20, 2025 · ac1a740 · ac1a740
1 parent b43c0e9
commit ac1a740
Show file tree

Hide file tree

Showing 2 changed files with 18 additions and 5 deletions.
diff --git a/setup.py b/setup.py
@@ -67,7 +67,7 @@
             "uvicorn[standard]",
         ],
         "llm-oga-cpu": [
-            "onnxruntime-genai==0.5.2",
+            "onnxruntime-genai>=0.5.2",
             "torch>=2.0.0,<2.4",
             "transformers<4.45.0",
             "turnkeyml[llm]",

diff --git a/src/lemonade/tools/ort_genai/oga.py b/src/lemonade/tools/ort_genai/oga.py
@@ -15,6 +15,7 @@
 import shutil
 from fnmatch import fnmatch
 from queue import Queue
+from packaging.version import Version
 from huggingface_hub import snapshot_download
 import onnxruntime_genai as og
 import onnxruntime_genai.models.builder as model_builder
@@ -120,12 +121,19 @@ def generate(
     ):
         params = og.GeneratorParams(self.model)
 
+        # There is a breaking API change in OGA 0.6.0
+        # Determine whether we should use the old or new APIs
+        use_oga_pre_6_api = Version(og.__version__) < Version("0.6.0")
+        use_oga_post_6_api = not use_oga_pre_6_api
+
         if pad_token_id:
             params.pad_token_id = pad_token_id
 
         max_length = len(input_ids) + max_new_tokens
 
-        params.input_ids = input_ids
+        if use_oga_pre_6_api:
+            params.input_ids = input_ids
+
         if self.config and "search" in self.config:
             search_config = self.config["search"]
             params.set_search_options(
@@ -159,10 +167,13 @@ def generate(
         params.try_graph_capture_with_max_batch_size(1)
 
         generator = og.Generator(self.model, params)
+        if use_oga_post_6_api:
+            generator.append_tokens(input_ids)
 
         if streamer is None:
             prompt_start_time = time.perf_counter()
-            generator.compute_logits()
+            if use_oga_pre_6_api:
+                generator.compute_logits()
             generator.generate_next_token()
             prompt_end_time = time.perf_counter()
 
@@ -173,7 +184,8 @@ def generate(
                 token_gen_times = []
                 while not generator.is_done():
                     token_gen_start_time = time.perf_counter()
-                    generator.compute_logits()
+                    if use_oga_pre_6_api:
+                        generator.compute_logits()
                     generator.generate_next_token()
                     token_gen_end_time = time.perf_counter()
 
@@ -194,7 +206,8 @@ def generate(
             stop_early = False
 
             while not generator.is_done() and not stop_early:
-                generator.compute_logits()
+                if use_oga_pre_6_api:
+                    generator.compute_logits()
                 generator.generate_next_token()
 
                 new_token = generator.get_next_tokens()[0]