lemonade-sdk 7.0.0__py3-none-any.whl → 7.0.2__py3-none-any.whl

lemonade_server/model_manager.py

@@ -50,7 +50,7 @@ class ModelManager:
         downloaded_models = {}
         for model in self.supported_models:
             if (
-                self.supported_models[model]["checkpoint"]
+                self.supported_models[model]["checkpoint"].split(":")[0]
                 in self.downloaded_hf_checkpoints
             ):
                 downloaded_models[model] = self.supported_models[model]
@@ -62,22 +62,17 @@ class ModelManager:
         Returns a dictionary of locally available models that are enabled by
         the current installation.
         """
-        hybrid_installed = (
-            "onnxruntime-vitisai" in pkg_resources.working_set.by_key
-            and "onnxruntime-genai-directml-ryzenai" in pkg_resources.working_set.by_key
-        )
+        return self.filter_models_by_backend(self.downloaded_models)
 
-        downloaded_models_enabled = {}
-        for model, value in self.downloaded_models.items():
-            if value["recipe"] == "oga-hybrid" and hybrid_installed:
-                downloaded_models_enabled[model] = value
-            else:
-                # All other models are CPU models right now
-                # This logic will get more sophisticated when we
-                # start to support more backends
-                downloaded_models_enabled[model] = value
-
-        return downloaded_models_enabled
+    def download_gguf(self, checkpoint) -> str:
+        # The colon after the checkpoint name indicates which
+        # specific GGUF to download
+        repo_id = checkpoint.split(":")[0]
+        pattern_to_match = f'*{checkpoint.split(":")[1]}.gguf'
+        return huggingface_hub.snapshot_download(
+            repo_id=repo_id,
+            allow_patterns=[pattern_to_match],
+        )
 
     def download_models(self, models: list[str]):
         """
@@ -91,7 +86,29 @@ class ModelManager:
                 )
             checkpoint = self.supported_models[model]["checkpoint"]
             print(f"Downloading {model} ({checkpoint})")
-            huggingface_hub.snapshot_download(repo_id=checkpoint)
+
+            if "gguf" in checkpoint.lower():
+                self.download_gguf(checkpoint)
+            else:
+                huggingface_hub.snapshot_download(repo_id=checkpoint)
+
+    def filter_models_by_backend(self, models: dict) -> dict:
+        """
+        Returns a filtered dict of models that are enabled by the
+        current environment.
+        """
+        hybrid_installed = (
+            "onnxruntime-vitisai" in pkg_resources.working_set.by_key
+            and "onnxruntime-genai-directml-ryzenai" in pkg_resources.working_set.by_key
+        )
+        filtered = {}
+        for model, value in models.items():
+            if value.get("recipe") == "oga-hybrid":
+                if hybrid_installed:
+                    filtered[model] = value
+            else:
+                filtered[model] = value
+        return filtered
 
 
 # This file was originally licensed under Apache 2.0. It has been modified.

lemonade_server/server_models.json

@@ -9,13 +9,13 @@
         "checkpoint": "amd/Llama-3.2-1B-Instruct-awq-uint4-float16-cpu-onnx",
         "recipe": "oga-cpu",
         "reasoning": false,
-        "suggested": true
+        "suggested": false
     },
     "Llama-3.2-3B-Instruct-CPU": {
         "checkpoint": "amd/Llama-3.2-3B-Instruct-awq-uint4-float16-cpu-onnx",
         "recipe": "oga-cpu",
         "reasoning": false,
-        "suggested": true
+        "suggested": false
     },
     "Phi-3-Mini-Instruct-CPU": {
         "checkpoint": "amd/Phi-3-mini-4k-instruct_int4_float16_onnx_cpu",
@@ -103,6 +103,13 @@
         "max_prompt_length": 2000,
         "suggested": true
     },
+    "Llama-xLAM-2-8b-fc-r-Hybrid": {
+        "checkpoint": "amd/Llama-xLAM-2-8b-fc-r-awq-g128-int4-asym-bfp16-onnx-hybrid",
+        "recipe": "oga-hybrid",
+        "reasoning": false,
+        "max_prompt_length": 2000,
+        "suggested": true
+    },
     "Llama-3.2-1B-Instruct-DirectML": {
         "checkpoint": "amd/Llama-3.2-1B-Instruct-dml-int4-awq-block-128-directml",
         "recipe": "oga-igpu",
@@ -138,5 +145,47 @@
         "recipe": "oga-igpu",
         "reasoning": false,
         "suggested": false
+    },
+    "Qwen3-0.6B-GGUF": {
+        "checkpoint": "unsloth/Qwen3-0.6B-GGUF:Q4_0",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
+    },
+    "Qwen3-1.7B-GGUF": {
+        "checkpoint": "unsloth/Qwen3-1.7B-GGUF:Q4_0",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
+    },
+    "Qwen3-4B-GGUF": {
+        "checkpoint": "unsloth/Qwen3-4B-GGUF:Q4_0",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
+    },
+    "Qwen3-8B-GGUF": {
+        "checkpoint": "unsloth/Qwen3-8B-GGUF:Q4_1",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
+    },
+    "DeepSeek-Qwen3-8B-GGUF": {
+        "checkpoint": "unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_1",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
+    },
+    "Qwen3-14B-GGUF": {
+        "checkpoint": "unsloth/Qwen3-14B-GGUF:Q4_0",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
+    },
+    "Qwen3-30B-A3B-GGUF": {
+        "checkpoint": "unsloth/Qwen3-30B-A3B-GGUF:Q4_0",
+        "recipe": "llamacpp",
+        "reasoning": true,
+        "suggested": true
     }
-}
+}