lemonade-sdk 8.0.6__tar.gz → 8.1.0__tar.gz

{lemonade_sdk-8.0.6/src/lemonade_sdk.egg-info → lemonade_sdk-8.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lemonade-sdk
-Version: 8.0.6
+Version: 8.1.0
 Summary: Lemonade SDK: Your LLM Aide for Validation and Deployment
 Author-email: lemonade@amd.com
 Requires-Python: >=3.10, <3.13
@@ -22,16 +22,15 @@ Requires-Dist: pytz
 Requires-Dist: zstandard
 Requires-Dist: fastapi
 Requires-Dist: uvicorn[standard]
-Requires-Dist: openai>=1.81.0
+Requires-Dist: openai<1.97.1,>=1.81.0
 Requires-Dist: transformers<=4.53.2
 Requires-Dist: jinja2
 Requires-Dist: tabulate
 Requires-Dist: sentencepiece
 Requires-Dist: huggingface-hub==0.33.0
-Provides-Extra: oga-hybrid
-Requires-Dist: onnx==1.16.1; extra == "oga-hybrid"
-Requires-Dist: numpy==1.26.4; extra == "oga-hybrid"
-Requires-Dist: protobuf>=6.30.1; extra == "oga-hybrid"
+Provides-Extra: oga-ryzenai
+Requires-Dist: onnxruntime-genai-directml-ryzenai==0.7.0.2; extra == "oga-ryzenai"
+Requires-Dist: protobuf>=6.30.1; extra == "oga-ryzenai"
 Provides-Extra: oga-cpu
 Requires-Dist: onnxruntime-genai==0.8.2; extra == "oga-cpu"
 Requires-Dist: onnxruntime>=1.22.0; extra == "oga-cpu"
@@ -43,14 +42,32 @@ Requires-Dist: pandas>=1.5.3; extra == "dev"
 Requires-Dist: matplotlib; extra == "dev"
 Requires-Dist: human-eval-windows==1.0.4; extra == "dev"
 Requires-Dist: lm-eval[api]; extra == "dev"
+Provides-Extra: oga-hybrid
+Requires-Dist: lemonade-sdk[oga-ryzenai]; extra == "oga-hybrid"
+Provides-Extra: oga-unified
+Requires-Dist: lemonade-sdk[oga-ryzenai]; extra == "oga-unified"
 Provides-Extra: oga-hybrid-minimal
-Requires-Dist: lemonade-sdk[oga-hybrid]; extra == "oga-hybrid-minimal"
+Requires-Dist: lemonade-sdk[oga-ryzenai]; extra == "oga-hybrid-minimal"
 Provides-Extra: oga-cpu-minimal
 Requires-Dist: lemonade-sdk[oga-cpu]; extra == "oga-cpu-minimal"
+Provides-Extra: oga-npu-minimal
+Requires-Dist: lemonade-sdk[oga-ryzenai]; extra == "oga-npu-minimal"
 Provides-Extra: llm
 Requires-Dist: lemonade-sdk[dev]; extra == "llm"
 Provides-Extra: llm-oga-cpu
 Requires-Dist: lemonade-sdk[dev,oga-cpu]; extra == "llm-oga-cpu"
+Provides-Extra: llm-oga-npu
+Requires-Dist: onnx==1.16.0; extra == "llm-oga-npu"
+Requires-Dist: onnxruntime==1.18.0; extra == "llm-oga-npu"
+Requires-Dist: numpy==1.26.4; extra == "llm-oga-npu"
+Requires-Dist: protobuf>=6.30.1; extra == "llm-oga-npu"
+Requires-Dist: lemonade-sdk[dev]; extra == "llm-oga-npu"
+Provides-Extra: llm-oga-hybrid
+Requires-Dist: onnx==1.16.1; extra == "llm-oga-hybrid"
+Requires-Dist: numpy==1.26.4; extra == "llm-oga-hybrid"
+Requires-Dist: protobuf>=6.30.1; extra == "llm-oga-hybrid"
+Provides-Extra: llm-oga-unified
+Requires-Dist: lemonade-sdk[dev,llm-oga-hybrid]; extra == "llm-oga-unified"
 Provides-Extra: llm-oga-igpu
 Requires-Dist: onnxruntime-genai-directml==0.6.0; extra == "llm-oga-igpu"
 Requires-Dist: onnxruntime-directml<1.22.0,>=1.19.0; extra == "llm-oga-igpu"
@@ -61,16 +78,6 @@ Requires-Dist: onnxruntime-genai-cuda==0.8.2; extra == "llm-oga-cuda"
 Requires-Dist: onnxruntime-gpu>=1.22.0; extra == "llm-oga-cuda"
 Requires-Dist: transformers<=4.51.3; extra == "llm-oga-cuda"
 Requires-Dist: lemonade-sdk[dev]; extra == "llm-oga-cuda"
-Provides-Extra: llm-oga-npu
-Requires-Dist: onnx==1.16.0; extra == "llm-oga-npu"
-Requires-Dist: onnxruntime==1.18.0; extra == "llm-oga-npu"
-Requires-Dist: numpy==1.26.4; extra == "llm-oga-npu"
-Requires-Dist: protobuf>=6.30.1; extra == "llm-oga-npu"
-Requires-Dist: lemonade-sdk[dev]; extra == "llm-oga-npu"
-Provides-Extra: llm-oga-hybrid
-Requires-Dist: lemonade-sdk[dev,oga-hybrid]; extra == "llm-oga-hybrid"
-Provides-Extra: llm-oga-unified
-Requires-Dist: lemonade-sdk[llm-oga-hybrid]; extra == "llm-oga-unified"
 Dynamic: author-email
 Dynamic: description
 Dynamic: description-content-type
@@ -174,7 +181,7 @@ lemonade-server list
 
 ## Model Library
 
-Lemonade supports both GGUF and ONNX models as detailed in the [Supported Configuration](#supported-configurations) section. A list of all built-in models is available [here](https://lemonade-server.ai/docs/server/models/).
+Lemonade supports both GGUF and ONNX models as detailed in the [Supported Configuration](#supported-configurations) section. A list of all built-in models is available [here](https://lemonade-server.ai/docs/server/server_models/).
 
 You can also import custom GGUF and ONNX models from Hugging Face by using our [Model Manager](http://localhost:8000/#model-management) (requires server to be running).
 <p align="center">
@@ -263,7 +270,7 @@ completion = client.chat.completions.create(
 print(completion.choices[0].message.content)
 ```
 
-For more detailed integration instructions, see the [Integration Guide](./server_integration.md).
+For more detailed integration instructions, see the [Integration Guide](./docs/server/server_integration.md).
 
 ## Beyond an LLM Server
 
@@ -272,6 +279,10 @@ The [Lemonade SDK](./docs/README.md) also include the following components:
 - 🐍 **[Lemonade API](./docs/lemonade_api.md)**: High-level Python API to directly integrate Lemonade LLMs into Python applications.
 - 🖥️ **[Lemonade CLI](./docs/dev_cli/README.md)**: The `lemonade` CLI lets you mix-and-match LLMs (ONNX, GGUF, SafeTensors) with prompting templates, accuracy testing, performance benchmarking, and memory profiling to characterize your models on your hardware.
 
+## FAQ
+
+To read our frequently asked questions, see our [FAQ Guide](./docs/faq.md)
+
 ## Contributing
 
 We are actively seeking collaborators from across the industry. If you would like to contribute to this project, please check out our [contribution guide](./docs/contribute.md).

{lemonade_sdk-8.0.6 → lemonade_sdk-8.1.0}/README.md

@@ -92,7 +92,7 @@ lemonade-server list
 
 ## Model Library
 
-Lemonade supports both GGUF and ONNX models as detailed in the [Supported Configuration](#supported-configurations) section. A list of all built-in models is available [here](https://lemonade-server.ai/docs/server/models/).
+Lemonade supports both GGUF and ONNX models as detailed in the [Supported Configuration](#supported-configurations) section. A list of all built-in models is available [here](https://lemonade-server.ai/docs/server/server_models/).
 
 You can also import custom GGUF and ONNX models from Hugging Face by using our [Model Manager](http://localhost:8000/#model-management) (requires server to be running).
 <p align="center">
@@ -181,7 +181,7 @@ completion = client.chat.completions.create(
 print(completion.choices[0].message.content)
 ```
 
-For more detailed integration instructions, see the [Integration Guide](./server_integration.md).
+For more detailed integration instructions, see the [Integration Guide](./docs/server/server_integration.md).
 
 ## Beyond an LLM Server
 
@@ -190,6 +190,10 @@ The [Lemonade SDK](./docs/README.md) also include the following components:
 - 🐍 **[Lemonade API](./docs/lemonade_api.md)**: High-level Python API to directly integrate Lemonade LLMs into Python applications.
 - 🖥️ **[Lemonade CLI](./docs/dev_cli/README.md)**: The `lemonade` CLI lets you mix-and-match LLMs (ONNX, GGUF, SafeTensors) with prompting templates, accuracy testing, performance benchmarking, and memory profiling to characterize your models on your hardware.
 
+## FAQ
+
+To read our frequently asked questions, see our [FAQ Guide](./docs/faq.md)
+
 ## Contributing
 
 We are actively seeking collaborators from across the industry. If you would like to contribute to this project, please check out our [contribution guide](./docs/contribute.md).

{lemonade_sdk-8.0.6 → lemonade_sdk-8.1.0}/setup.py

@@ -44,7 +44,7 @@ setup(
         "zstandard",
         "fastapi",
         "uvicorn[standard]",
-        "openai>=1.81.0",
+        "openai>=1.81.0,<1.97.1",
         "transformers<=4.53.2",
         "jinja2",
         "tabulate",
@@ -54,11 +54,9 @@ setup(
     extras_require={
         # The non-dev extras are meant to deploy specific backends into end-user
         # applications, without including developer-focused tools
-        "oga-hybrid": [
-            # Note: `lemonade-install --ryzenai hybrid` is necessary
-            # to complete installation
-            "onnx==1.16.1",
-            "numpy==1.26.4",
+        # Primary NPU extra using unified PyPI package
+        "oga-ryzenai": [
+            "onnxruntime-genai-directml-ryzenai==0.7.0.2",
             "protobuf>=6.30.1",
         ],
         "oga-cpu": [
@@ -81,11 +79,31 @@ setup(
             "lm-eval[api]",
         ],
         # Keep backwards compatibility for old extras names
-        "oga-hybrid-minimal": ["lemonade-sdk[oga-hybrid]"],
+        "oga-hybrid": ["lemonade-sdk[oga-ryzenai]"],
+        "oga-unified": ["lemonade-sdk[oga-ryzenai]"],
+        "oga-hybrid-minimal": ["lemonade-sdk[oga-ryzenai]"],
         "oga-cpu-minimal": ["lemonade-sdk[oga-cpu]"],
+        "oga-npu-minimal": ["lemonade-sdk[oga-ryzenai]"],
         "llm": ["lemonade-sdk[dev]"],
         "llm-oga-cpu": ["lemonade-sdk[dev,oga-cpu]"],
         # The following extras are deprecated and/or not commonly used
+        "llm-oga-npu": [
+            "onnx==1.16.0",
+            # NPU requires specific onnxruntime version for Ryzen AI compatibility
+            # This may conflict with other OGA extras that require >=1.22.0
+            "onnxruntime==1.18.0",
+            "numpy==1.26.4",
+            "protobuf>=6.30.1",
+            "lemonade-sdk[dev]",
+        ],
+        "llm-oga-hybrid": [
+            # Note: `lemonade-install --ryzenai hybrid` is necessary
+            # to complete installation for RAI 1.4.0.
+            "onnx==1.16.1",
+            "numpy==1.26.4",
+            "protobuf>=6.30.1",
+        ],
+        "llm-oga-unified": ["lemonade-sdk[dev, llm-oga-hybrid]"],
         "llm-oga-igpu": [
             "onnxruntime-genai-directml==0.6.0",
             "onnxruntime-directml>=1.19.0,<1.22.0",
@@ -98,17 +116,6 @@ setup(
             "transformers<=4.51.3",
             "lemonade-sdk[dev]",
         ],
-        "llm-oga-npu": [
-            "onnx==1.16.0",
-            # NPU requires specific onnxruntime version for Ryzen AI compatibility
-            # This may conflict with other OGA extras that require >=1.22.0
-            "onnxruntime==1.18.0",
-            "numpy==1.26.4",
-            "protobuf>=6.30.1",
-            "lemonade-sdk[dev]",
-        ],
-        "llm-oga-hybrid": ["lemonade-sdk[dev,oga-hybrid]"],
-        "llm-oga-unified": ["lemonade-sdk[llm-oga-hybrid]"],
     },
     classifiers=[],
     entry_points={

{lemonade_sdk-8.0.6 → lemonade_sdk-8.1.0}/src/lemonade/common/network.py

@@ -1,7 +1,7 @@
 import os
 from typing import Optional
 import socket
-from huggingface_hub import model_info
+from huggingface_hub import model_info, snapshot_download
 
 
 def is_offline():
@@ -48,3 +48,20 @@ def get_base_model(checkpoint: str) -> Optional[str]:
     except Exception:  # pylint: disable=broad-except
         pass
     return None
+
+
+def custom_snapshot_download(repo_id, **kwargs):
+    """
+    Custom snapshot download with retry logic for Windows symlink privilege errors.
+    """
+    for attempt in range(2):
+        try:
+            return snapshot_download(repo_id=repo_id, **kwargs)
+        except OSError as e:
+            if (
+                hasattr(e, "winerror")
+                and e.winerror == 1314  # pylint: disable=no-member
+                and attempt < 1
+            ):
+                continue
+            raise

{lemonade_sdk-8.0.6 → lemonade_sdk-8.1.0}/src/lemonade/tools/llamacpp/bench.py

@@ -68,7 +68,9 @@ class LlamaCppBench(Bench):
                 # and error handling
                 model.time_to_first_token = None
                 model.tokens_per_second = None
-                raw_output, stderr = model.generate(prompt, return_raw=True)
+                raw_output, stderr = model.generate(
+                    prompt, max_new_tokens=output_tokens, return_raw=True
+                )
 
                 if model.time_to_first_token is None or model.tokens_per_second is None:
                     error_msg = (

{lemonade_sdk-8.0.6 → lemonade_sdk-8.1.0}/src/lemonade/tools/llamacpp/utils.py

@@ -215,10 +215,10 @@ def get_local_checkpoint_path(base_checkpoint, variant):
     full_model_path = None
     model_to_use = None
     try:
-        from huggingface_hub import snapshot_download
+        from lemonade.common.network import custom_snapshot_download
 
-        snapshot_path = snapshot_download(
-            repo_id=base_checkpoint,
+        snapshot_path = custom_snapshot_download(
+            base_checkpoint,
             local_files_only=True,
         )
 
@@ -405,10 +405,10 @@ def download_gguf(config_checkpoint, config_mmproj=None) -> dict:
     core_files, sharded_files = identify_gguf_models(checkpoint, variant, config_mmproj)
 
     # Download the files
-    from huggingface_hub import snapshot_download
+    from lemonade.common.network import custom_snapshot_download
 
-    snapshot_folder = snapshot_download(
-        repo_id=checkpoint,
+    snapshot_folder = custom_snapshot_download(
+        checkpoint,
         allow_patterns=list(core_files.values()) + sharded_files,
     )
 
@@ -573,7 +573,7 @@ class LlamaCppAdapter(ModelAdapter):
                 #
                 if "llama_perf_context_print:        eval time =" in line:
                     parts = line.split("=")[1].split()
-                    self.response_tokens = int(parts[3])
+                    self.response_tokens = int(parts[3]) + 1  # include first token
                     response_time_ms = float(parts[0])
                     self.tokens_per_second = (
                         1000 * self.response_tokens / response_time_ms