lemonade-sdk 8.1.11__py3-none-any.whl → 8.2.2__py3-none-any.whl

{lemonade_sdk-8.1.11.dist-info → lemonade_sdk-8.2.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lemonade-sdk
-Version: 8.1.11
+Version: 8.2.2
 Summary: Lemonade SDK: Your LLM Aide for Validation and Deployment
 Author-email: lemonade@amd.com
 Requires-Python: >=3.10, <3.14
@@ -22,19 +22,20 @@ Requires-Dist: pytz
 Requires-Dist: zstandard
 Requires-Dist: fastapi
 Requires-Dist: uvicorn[standard]
-Requires-Dist: openai<1.97.1,>=1.81.0
+Requires-Dist: openai<3.0.0,>=2.0.0
 Requires-Dist: transformers<=4.53.2
 Requires-Dist: jinja2
 Requires-Dist: tabulate
 Requires-Dist: sentencepiece
 Requires-Dist: huggingface-hub[hf_xet]==0.33.0
 Requires-Dist: python-dotenv
+Requires-Dist: python-multipart
 Requires-Dist: rumps>=0.4.0; sys_platform == "darwin"
 Provides-Extra: oga-ryzenai
-Requires-Dist: onnxruntime-genai-directml-ryzenai==0.7.0.2.1; extra == "oga-ryzenai"
+Requires-Dist: onnxruntime-genai-directml-ryzenai==0.9.2.1; 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-genai==0.9.2; extra == "oga-cpu"
 Requires-Dist: onnxruntime>=1.22.0; extra == "oga-cpu"
 Provides-Extra: dev
 Requires-Dist: torch>=2.6.0; extra == "dev"
@@ -264,6 +265,7 @@ This project is:
   - [OnnxRuntime GenAI](https://github.com/microsoft/onnxruntime-genai)
   - [Hugging Face Hub](https://github.com/huggingface/huggingface_hub)
   - [OpenAI API](https://github.com/openai/openai-python)
+  - [IRON/MLIR-AIE](https://github.com/Xilinx/mlir-aie)
   - and more...
 - Accelerated by mentorship from the OCV Catalyst program.
 - Licensed under the [Apache 2.0 License](https://github.com/lemonade-sdk/lemonade/blob/main/LICENSE).

{lemonade_sdk-8.1.11.dist-info → lemonade_sdk-8.2.2.dist-info}/RECORD

@@ -1,10 +1,10 @@
 lemonade/__init__.py,sha256=W1Qk7r0rnQqFhPNHp6BIBT_q-OH3s-8Q_POoVfAmKW0,117
 lemonade/api.py,sha256=Oc4yBA3LZg8FrTsbuDq1p9-XE74pqNnIEUhXyKa7qg8,5786
-lemonade/cache.py,sha256=5iZbk273TiTMqK_vdzPOPYTo6VsWW2gNByOISA9zi1w,3002
+lemonade/cache.py,sha256=fUjtHYkRdHTULjNMrDNOFDGW_QMVUg54ZE1NukBP2oM,3314
 lemonade/cli.py,sha256=qU5bW7RQAUKNSpvrhVyzn68NMxyi-336Ke_JU4bsv1Q,5708
 lemonade/sequence.py,sha256=KSH7BPsiyDKsOsg_ziQKEGsDwMmuO_YbgPRBxkZd0pw,13267
 lemonade/state.py,sha256=sdSezla7Cd7KYL90xY3p9kcNV4ndSyN6UvNLOr3vBMA,5261
-lemonade/version.py,sha256=et6OH4dSRF6oHhfToZjlMBObl75tflM6mBJXEWJArps,23
+lemonade/version.py,sha256=qHoUgbIt6pjWzUYt9wXlO8qcrd9dEl09FyPsRTkJshI,22
 lemonade/common/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 lemonade/common/build.py,sha256=zTb0m1-kuUx6zw5QHp2SNnVuN6jOTMQ2FCdj9iH374U,6140
 lemonade/common/cli_helpers.py,sha256=hjBfXrTtFl8gmCFlL-ksviXR0mOcdPtTWVNKoEp3PG4,4993
@@ -13,8 +13,8 @@ lemonade/common/filesystem.py,sha256=QV3cHhKNu-7W2rr8wZ4JQfD2rP_5T2Js7jiDQBYWHVQ
 lemonade/common/inference_engines.py,sha256=3bUGQe9wtfTiwt8kvI_ry077uyc9lid2G1fJX95kN1A,12969
 lemonade/common/network.py,sha256=qXpUjDYQEYM_gH3JwTtU-pu_yCKcaa1IeohJRPy91-A,2903
 lemonade/common/printing.py,sha256=GFFzrXIineIOMa9yu0lo5sL4j6A5BBg_T9aUCdP-juw,3229
-lemonade/common/status.py,sha256=xSOZN508cdRtrs1HVyr9zmASYg69EsZBLSs0lroLoCM,16519
-lemonade/common/system_info.py,sha256=Msa0pCSj3ZN3nerjY8wdqjjJLg6GPhbWf2htSNcFIHc,49607
+lemonade/common/status.py,sha256=dxAahChPGVmfT91DJW949Xjgm9r5E-Y9KOLPEw7BMh8,16562
+lemonade/common/system_info.py,sha256=PWpEtOz6CqvnUYL8_3lX61_GNbrbp0O2x4ZIwRh6RJg,48780
 lemonade/common/test_helpers.py,sha256=Gwk-pa_6xYAo2oro-2EJNfuouAfw8k_brCbcMC-E-r0,758
 lemonade/profilers/__init__.py,sha256=JKVonvJ4XZ9_6sKXPWsiMLQCNyzQOxhQw5BEHR1qOfU,31
 lemonade/profilers/agt_power.py,sha256=t_37VEg8LPapjSKSjJln-jFznZtTIf5UpzlAXcVGOrc,16771
@@ -22,9 +22,9 @@ lemonade/profilers/hwinfo_power.py,sha256=UQr-EHq7B4T-IvzmErCRK0-QxcFnho4ftCaWy5
 lemonade/profilers/memory_tracker.py,sha256=1iuKt0FmNVYLDnOc-oZM8dX9TUksvoxO0m2EoYWjhYQ,9367
 lemonade/profilers/profiler.py,sha256=Y5FSbc386bMlTVbqCuya9pYrso5aTthxahR1V_ZKQ9E,1902
 lemonade/tools/__init__.py,sha256=_6xRc-FHxmujoLjLjWtpYrWYEXtCSneSy-5ya01kyPk,53
-lemonade/tools/accuracy.py,sha256=9HCmczDngkBUuUrt49d2CkRo4J0qyWoFYs5cj20bGkg,11714
-lemonade/tools/adapter.py,sha256=Ex63Y1SPCOSV4M_QtzEn3YVd39d3yew0lpmEFgp8aH4,3169
-lemonade/tools/bench.py,sha256=aN5LMA_EH6-ZhAH3Gf26JYL7s0eKpUd3j-bReRhzvEY,10016
+lemonade/tools/accuracy.py,sha256=X2OLzx3VDhpOZO9aHKH3vrjeeKI1lTYXZFeZF9Qpslg,15188
+lemonade/tools/adapter.py,sha256=fFRSnLiQk20WYQmYFD0JK1t-fDLKEWhD6u9qqB_BECk,3496
+lemonade/tools/bench.py,sha256=jyL85yqBPzyT2pr9OWzihvyoN0v3JQg_BHtIHy9d2zg,10881
 lemonade/tools/humaneval.py,sha256=JbxuoOzvR4iyxZv4R6MI7a3gUt5ef_Jj6Ie-9VP2wzY,9531
 lemonade/tools/management_tools.py,sha256=HQBcr7LYuMqVRYQtvnkNpfutBTA7lblszyoAjjVGu1Y,10201
 lemonade/tools/mmlu.py,sha256=c2QaIMDzjqxCvgHlMXmy_dP1sAFkwkDxL7RO2nogI6s,11071
@@ -32,51 +32,52 @@ lemonade/tools/perplexity.py,sha256=eiaTZ3yhqF2pfwOffVbKKJLwjSri7Im2pC-tBJr7LLU,
 lemonade/tools/prompt.py,sha256=PyLksp1k8jsZsU7XBRK61k1DUHhbdLa20h-AP8Noh3w,9011
 lemonade/tools/tool.py,sha256=UsxVYukfm_iM3BfeGYPZxQlTK5UfDfDOl3RIyLr8A1Y,13256
 lemonade/tools/flm/__init__.py,sha256=NQ4CEzJZGS_VvxPMlfrK4Dcx48bQSoUR4iG8e7yZjas,46
-lemonade/tools/flm/utils.py,sha256=hHjSiRlkw239n03CyZeRQomxtmsJptM7m5M3CNnPlqo,8126
-lemonade/tools/huggingface/bench.py,sha256=-mTfldCtquL4mspq8ykVwDc9Mut5Ecv_jHJnSb0CYGE,6734
+lemonade/tools/flm/utils.py,sha256=5KS0IND_8E3lgKNdIFeJqMcdiCsPpYCjgL1FzH8YvGk,9933
+lemonade/tools/huggingface/bench.py,sha256=OMkIm_6o4znu8xZo4MzvHULGu0WPQl529kpAngJut-g,6907
 lemonade/tools/huggingface/load.py,sha256=KsSGOBBD-tNEIfYC8mCWV_jpnkjHMhN3juVmC1Ln4uQ,7745
 lemonade/tools/huggingface/utils.py,sha256=j1S-IgjDsznUIVwkHSqqChmFyqIx9f3WcEelzohWwvU,13955
-lemonade/tools/llamacpp/bench.py,sha256=1fkE02ecg-jRk92i5dTAXz6re14WH8bd-Z9l-m3lbDA,4844
-lemonade/tools/llamacpp/load.py,sha256=DFCvQN548Ch9H8U_rHOiYviinzw6vixb5-V7xLj7XE4,6499
-lemonade/tools/llamacpp/utils.py,sha256=JpI9McEYbrZXQXb0Wo7EoQ8-0LLmmZuwbgGSuuYTiyQ,35221
+lemonade/tools/llamacpp/bench.py,sha256=JfsKboJCDdkZ5ZJ9Xl-ty5cCBj6pAlUrXZiEQ7Tt60c,9689
+lemonade/tools/llamacpp/load.py,sha256=YvDHqXe4tKGuT5aG_wP4xH3Px0RmG92u73kVOTHSEgc,7671
+lemonade/tools/llamacpp/utils.py,sha256=kHL2xaCnQ7hLY15FoQJLYGf29KddxoEyczXmN1TWfQY,46182
 lemonade/tools/oga/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lemonade/tools/oga/bench.py,sha256=PJXv4UchcS2YPwijNzef8DY4DSAKYxIYY1ycHuH3T34,5005
-lemonade/tools/oga/load.py,sha256=x-A-nhoni-WyDpVCLcWRAMfs5ouac9MJzxT-rsnLPw8,34226
+lemonade/tools/oga/bench.py,sha256=V-qc8mbFHvOv13QTrgFOXSmUAhzzGTK8S7UpVCl-KIw,4329
+lemonade/tools/oga/load.py,sha256=yxeRPUUrDOgrjv9d_R8MGLInunvI7HId_Q2cN8K_8h4,31113
+lemonade/tools/oga/migration.py,sha256=zo34chEbtkWk4GaDwgW326o09eF2wAsoOiTzqQ6DMLU,13505
 lemonade/tools/oga/utils.py,sha256=F8UVLKlfYcLa2SUqlehar8-jaX2Aw4u58DjHNNvLdOA,17675
 lemonade/tools/report/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 lemonade/tools/report/llm_report.py,sha256=bVHhwCINA-Ok2EdSwAsLubsc83N3KWOVuwTguw7jDcE,6676
-lemonade/tools/report/table.py,sha256=Kv_Epd8a6KIrdzSC2EgIl6uTKw7E5eMq10Tg16O0WxM,27996
+lemonade/tools/report/table.py,sha256=7hsJNr9x9yxvpTTVp64FpTAfxyUDvXHcDvGk5L84j1g,30187
 lemonade/tools/server/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lemonade/tools/server/flm.py,sha256=lErpSYLIB6vyVavDd5c-XOz-85m8yPPlyHhiceKWf6c,4119
-lemonade/tools/server/llamacpp.py,sha256=dSab9hR_CcctmU8HeQYQ7U0XCbAQNd_QGZUt6q7cxHA,9952
-lemonade/tools/server/serve.py,sha256=ONdONSN0Noh2c65Z1m9zfQuHIDgb1c42Do0ao2MG534,68771
+lemonade/tools/server/flm.py,sha256=NVpjFzcmX6YzW6wCWyMEvQEe_QpRK-Jlcaajxy3kWuo,4025
+lemonade/tools/server/llamacpp.py,sha256=0scaDxJXne-PfUGh-r25UyqDSIOJdFkD1Js1WPFPzv0,11676
+lemonade/tools/server/serve.py,sha256=WYsKZTtPFp6JAp5K-N734wTsD3sltiTYrwV6oCs7ZWk,83487
 lemonade/tools/server/tool_calls.py,sha256=xrAlQwKG-nv2xLlf8f9CDSaUbyMn8ZtHkds9iZLG9K8,5230
 lemonade/tools/server/tray.py,sha256=EFnSc2Ra4owiHVz6ykoMhxi2fYqZAK1g21AynAYBiyk,24426
 lemonade/tools/server/webapp.py,sha256=GGSVIzN19C2ZaadOEPBg_D7Lt0PuF339NuWwjMPfZu8,1225
 lemonade/tools/server/wrapped_server.py,sha256=uh7ifrRX1Hx0IuRwZRCGPyQOukitE7kKQipCCz0bSGA,19844
 lemonade/tools/server/static/favicon.ico,sha256=hMmP9qGJNeZ0mFS86JIqPbZstXMZn0Z76_HfHQpREAU,126745
-lemonade/tools/server/static/logs.html,sha256=BSpdRJ8XJLStpD7XijXLbeeDaVW-FrJbI29zHNBxvYM,1321
-lemonade/tools/server/static/styles.css,sha256=GYJgRtlZSgz3pShDeuatu-J9TpVwGgOjOcIWwNnWYck,49100
-lemonade/tools/server/static/webapp.html,sha256=QQRMMMf8fbtJfGZYqBaRVd0-bENmdPfkVo8sdc4092Q,19151
-lemonade/tools/server/static/js/chat.js,sha256=jxyMyu4MfvI2YmsMbJQ8ZwDNBnLzu2nbjm-qLfgWSNI,42182
-lemonade/tools/server/static/js/model-settings.js,sha256=JXHeG7xVrRU181Hj7CZflERAi1Z6t-qwYFR4aH5nf5I,5820
-lemonade/tools/server/static/js/models.js,sha256=es3LwrU49UtoC59e-AusUpdSXQnIRNsZvw8f05keAM0,37856
-lemonade/tools/server/static/js/shared.js,sha256=NVu7lXotLnVVh1x_eXqxX1bLzYyW_eMQttOZ9f0RqUQ,17591
+lemonade/tools/server/static/logs.html,sha256=b9BtKWQwXJgr8xc2lWLDTx1zwYgM0Nsv5ZHRT_EdBM8,6157
+lemonade/tools/server/static/styles.css,sha256=XOi3-Aw_W0WOBEGBnyYZcniwjeHx7dhRBU2Lt4ePN00,52815
+lemonade/tools/server/static/webapp.html,sha256=QgUcUo7coLNDpy2HuF7IHu1t6wySSTNwcZQ2-Q8WqlU,21506
+lemonade/tools/server/static/js/chat.js,sha256=xil7PrnGohiNGtlqBcZ3upBr9hw5Y0P6KYqjbHpUYh0,39812
+lemonade/tools/server/static/js/model-settings.js,sha256=MZZkflb7cRaRqSagyknR_Pgii6pbDSlVKiAbgYqrGsQ,6737
+lemonade/tools/server/static/js/models.js,sha256=OirBQ7dEflazOaPwr4XtCjqcuaW_SUZY6vMkAAhctFg,53178
+lemonade/tools/server/static/js/shared.js,sha256=QPekwJQ9BLD2c0D6-CkoB0VjsehVy05TF8vTQmWGf-c,19312
 lemonade/tools/server/utils/macos_tray.py,sha256=xwHW44ZN5hDVlJcwIpHHfqn4VRXWxXHuDACaT-ZqdO8,7095
 lemonade/tools/server/utils/port.py,sha256=J7-g-Aqygb50jNoHLhhRfBZVM-uhGlcB5-oYBAehvgw,2263
 lemonade/tools/server/utils/thread.py,sha256=Z-PDzGcpgfN2qxTmtlROWqrUN0B2fXdPrqo_J10fR_w,2772
 lemonade/tools/server/utils/windows_tray.py,sha256=2z5aTmUPlkT-QfkcfwHsyA6dv6nSNBT0gXUErarhac8,13170
 lemonade_install/__init__.py,sha256=26zohKg2jgr_5y7tObduWMYQg8zCTWMZHL8lfi2zZVQ,40
-lemonade_install/install.py,sha256=p3pYqhUnLQ9JJMcbjlSYDYqN-amnU_535O9Oj1yPbyM,27608
-lemonade_sdk-8.1.11.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-lemonade_sdk-8.1.11.dist-info/licenses/NOTICE.md,sha256=RSca9LE5e6pvdWA_LXAUCcACIHPmINKqkRX-AVRqBGo,3499
-lemonade_server/cli.py,sha256=mzlIqLeGkU96KHuib8lmZn0snU7XjQ9lspOy83dVplo,23401
-lemonade_server/model_manager.py,sha256=6W6_nQea6hLD82Il2o_EgQ7oNaLfICXPKNjZiY9Y1Xk,26331
-lemonade_server/pydantic_models.py,sha256=5U3PZ__UqcWQh-dNXVBc-vyJc6-In2vngZXP9VmiScM,3954
-lemonade_server/server_models.json,sha256=_GxymNW7gBkJcTjzncYUvPUJ8kc-I3qqWwHqPztzYcA,14644
+lemonade_install/install.py,sha256=3NtQ4JayFM9VW7IOHommaV4s9RgdightyfhhxQ190O4,7753
+lemonade_sdk-8.2.2.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+lemonade_sdk-8.2.2.dist-info/licenses/NOTICE.md,sha256=RSca9LE5e6pvdWA_LXAUCcACIHPmINKqkRX-AVRqBGo,3499
+lemonade_server/cli.py,sha256=BItlVpuykHwTnCZJs-7v1qWRPBCnEx-aeP43a7S9uFk,23918
+lemonade_server/model_manager.py,sha256=dboE-7wF1Y9iLGUcBkaA-lpF8XJKSIN_y0mUYzgCm2w,32657
+lemonade_server/pydantic_models.py,sha256=QpsQrBjplv0IlwwUEt259o3eGv5YUuhq_sSiX2xBAyM,4074
+lemonade_server/server_models.json,sha256=cZAxlDkxQ5neb57jQ8j2fvIeuAfyk4Ga1Yele58OeTY,15062
 lemonade_server/settings.py,sha256=JOlZmirUXO9rA6BCODVFwyXrrHtYoH_LiKYm49lGm_c,1260
-lemonade_sdk-8.1.11.dist-info/METADATA,sha256=uLq8pbAD_uriJbpPT25i5gk-Ukq_RbWzzEckgtceFDo,15334
-lemonade_sdk-8.1.11.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lemonade_sdk-8.1.11.dist-info/entry_points.txt,sha256=7sRvpNhi1E7amnM7RZo57e8yFF9iA5uuRaIeJ1Xre6w,193
-lemonade_sdk-8.1.11.dist-info/top_level.txt,sha256=10ap5GNiPhalO4V50LRoxA1FqRT9g3Xkia6BITu880k,42
-lemonade_sdk-8.1.11.dist-info/RECORD,,
+lemonade_sdk-8.2.2.dist-info/METADATA,sha256=e_usINpNwof9VkWrKlpWNqyLH3UD0j6D6zX_Bimynww,15417
+lemonade_sdk-8.2.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lemonade_sdk-8.2.2.dist-info/entry_points.txt,sha256=7sRvpNhi1E7amnM7RZo57e8yFF9iA5uuRaIeJ1Xre6w,193
+lemonade_sdk-8.2.2.dist-info/top_level.txt,sha256=10ap5GNiPhalO4V50LRoxA1FqRT9g3Xkia6BITu880k,42
+lemonade_sdk-8.2.2.dist-info/RECORD,,

lemonade_server/cli.py

@@ -326,6 +326,16 @@ def run(
             ctx_size=ctx_size,
         )
     else:
+        # macOS: Check for port conflicts when server is already running
+        if platform.system() == "Darwin":
+            requested_port = port if port is not None else DEFAULT_PORT
+            if running_port != requested_port:
+                print(
+                    f"Lemonade Server is already running on port {running_port}\n"
+                    f"You requested port {requested_port}. Please stop the existing server first "
+                )
+                sys.exit(ExitCodes.SERVER_ALREADY_RUNNING)
+
         port = running_port
 
     # Pull model

lemonade_server/model_manager.py

@@ -8,8 +8,16 @@ from importlib.metadata import distributions
 from lemonade_server.pydantic_models import PullConfig
 from lemonade_server.pydantic_models import PullConfig
 from lemonade.cache import DEFAULT_CACHE_DIR
-from lemonade.tools.llamacpp.utils import parse_checkpoint, download_gguf
+from lemonade.tools.llamacpp.utils import (
+    parse_checkpoint,
+    download_gguf,
+    resolve_local_gguf_model,
+)
 from lemonade.common.network import custom_snapshot_download
+from lemonade.tools.oga.migration import (
+    detect_incompatible_ryzenai_models,
+    delete_incompatible_models,
+)
 
 USER_MODELS_FILE = os.path.join(DEFAULT_CACHE_DIR, "user_models.json")
 
@@ -61,10 +69,17 @@ class ModelManager:
 
         # Add the model name as a key in each entry, to make it easier
         # to access later
-
+        # Also convert labels to boolean fields for LoadConfig compatibility
         for key, value in models.items():
             value["model_name"] = key
 
+            # Convert labels to boolean fields for backwards compatibility with LoadConfig
+            labels = value.get("labels", [])
+            if "reasoning" in labels and "reasoning" not in value:
+                value["reasoning"] = True
+            if "vision" in labels and "vision" not in value:
+                value["vision"] = True
+
         return models
 
     @property
@@ -88,6 +103,8 @@ class ModelManager:
         Returns a dictionary of locally available models.
         For GGUF models with variants, checks if the specific variant files exist.
         """
+        from huggingface_hub.constants import HF_HUB_CACHE
+
         downloaded_models = {}
         downloaded_checkpoints = self.downloaded_hf_checkpoints
 
@@ -106,6 +123,19 @@ class ModelManager:
                 checkpoint = model_info["checkpoint"]
                 base_checkpoint, variant = parse_checkpoint(checkpoint)
 
+                # Special handling for locally uploaded user models (not internet-downloaded)
+                if (
+                    model.startswith("user.")
+                    and model_info.get("source") == "local_upload"
+                ):
+                    # Locally uploaded model: checkpoint is in cache directory format (models--xxx)
+                    local_model_path = os.path.join(HF_HUB_CACHE, base_checkpoint)
+                    if os.path.exists(local_model_path):
+                        downloaded_models[model] = model_info
+                    continue
+
+                # For all other models (server models and internet-downloaded user models),
+                # use the standard verification logic with variant checks
                 if base_checkpoint in downloaded_checkpoints:
                     # For GGUF models with variants, verify the specific variant files exist
                     if variant and model_info.get("recipe") == "llamacpp":
@@ -147,6 +177,55 @@ class ModelManager:
                         downloaded_models[model] = model_info
         return downloaded_models
 
+    def register_local_model(
+        self,
+        model_name: str,
+        checkpoint: str,
+        recipe: str,
+        reasoning: bool = False,
+        vision: bool = False,
+        mmproj: str = "",
+        snapshot_path: str = "",
+    ):
+
+        model_name_clean = model_name[5:]
+
+        # Prepare model info
+        labels = ["custom"]
+        if reasoning:
+            labels.append("reasoning")
+        if vision:
+            labels.append("vision")
+
+        new_user_model = {
+            "checkpoint": checkpoint,
+            "recipe": recipe,
+            "suggested": True,
+            "labels": labels,
+            "source": "local_upload",
+        }
+        if mmproj:
+            new_user_model["mmproj"] = mmproj
+
+        # Load existing user models
+        user_models = {}
+        if os.path.exists(USER_MODELS_FILE):
+            with open(USER_MODELS_FILE, "r", encoding="utf-8") as file:
+                user_models = json.load(file)
+
+        # Check for conflicts
+        if model_name_clean in user_models:
+            raise ValueError(
+                f"{model_name_clean} is already registered."
+                f"Please use a different model name or delete the existing model."
+            )
+
+        # Save to user_models.json
+        user_models[model_name_clean] = new_user_model
+        os.makedirs(os.path.dirname(USER_MODELS_FILE), exist_ok=True)
+        with open(USER_MODELS_FILE, "w", encoding="utf-8") as file:
+            json.dump(user_models, file)
+
     @property
     def downloaded_models_enabled(self) -> dict:
         """
@@ -310,7 +389,7 @@ class ModelManager:
                 # Check if FLM is available, and install it if not
                 if not is_flm_available():
                     print(
-                        "FLM is not installed or not at the minimum required version. Installing FLM..."
+                        "FLM is not installed or not at the latest version. Installing FLM..."
                     )
                     install_flm()
 
@@ -328,11 +407,23 @@ class ModelManager:
                         f"Please manually install FLM using 'lemonade-install --flm'."
                     ) from e
             elif "gguf" in checkpoint_to_download.lower():
-                download_gguf(
-                    gguf_model_config.checkpoint,
-                    gguf_model_config.mmproj,
-                    do_not_upgrade=do_not_upgrade,
+                # Parse checkpoint to check local cache first
+                base_checkpoint, variant = parse_checkpoint(
+                    gguf_model_config.checkpoint
+                )
+                local_result = resolve_local_gguf_model(
+                    base_checkpoint, variant, gguf_model_config.mmproj
                 )
+
+                # Only download if not found locally
+                if not local_result:
+                    download_gguf(
+                        gguf_model_config.checkpoint,
+                        gguf_model_config.mmproj,
+                        do_not_upgrade=do_not_upgrade,
+                    )
+                else:
+                    print(f"Model already exists locally, skipping download")
             else:
                 custom_snapshot_download(
                     checkpoint_to_download, do_not_upgrade=do_not_upgrade
@@ -342,6 +433,11 @@ class ModelManager:
             # We do this registration after the download so that we don't register
             # any incorrectly configured models where the download would fail
             if new_registration_model_config:
+                # For models downloaded from the internet (HuggingFace),
+                # keep the original checkpoint format (e.g., "amd/Llama-3.2-1B-Instruct-...")
+                # Do NOT convert to cache directory format - that's only for locally uploaded models
+                new_user_model["checkpoint"] = checkpoint
+
                 if os.path.exists(USER_MODELS_FILE):
                     with open(USER_MODELS_FILE, "r", encoding="utf-8") as file:
                         user_models: dict = json.load(file)
@@ -365,7 +461,7 @@ class ModelManager:
 
         installed_packages = {dist.metadata["Name"].lower() for dist in distributions()}
 
-        hybrid_installed = (
+        ryzenai_installed = (
             "onnxruntime-vitisai" in installed_packages
             and "onnxruntime-genai-directml-ryzenai" in installed_packages
         )
@@ -422,9 +518,9 @@ class ModelManager:
         for model, value in models.items():
             recipe = value.get("recipe")
 
-            # Filter OGA hybrid models based on package availability
-            if recipe == "oga-hybrid":
-                if not hybrid_installed:
+            # Filter Ryzen AI  models based on package availability
+            if recipe == "oga-hybrid" or recipe == "oga-npu":
+                if not ryzenai_installed:
                     continue
 
             if recipe == "flm":
@@ -444,6 +540,8 @@ class ModelManager:
         Deletes the specified model from local storage.
         For GGUF models with variants, only deletes the specific variant files.
         """
+        from huggingface_hub.constants import HF_HUB_CACHE
+
         if model_name not in self.supported_models:
             raise ValueError(
                 f"Model {model_name} is not supported. Please choose from the following: "
@@ -464,6 +562,37 @@ class ModelManager:
             except subprocess.CalledProcessError as e:
                 raise ValueError(f"Failed to delete FLM model {model_name}: {e}") from e
 
+        if checkpoint.startswith("models--"):
+            # This is already in cache directory format (local model)
+            # Extract just the base directory name (models--{name}) from checkpoint
+            # which might contain full file path like models--name\files\model.gguf
+            checkpoint_parts = checkpoint.replace("\\", "/").split("/")
+            base_checkpoint = checkpoint_parts[0]  # Just the models--{name} part
+            model_cache_dir = os.path.join(HF_HUB_CACHE, base_checkpoint)
+
+            if os.path.exists(model_cache_dir):
+                shutil.rmtree(model_cache_dir)
+                print(
+                    f"Successfully deleted local model {model_name} from {model_cache_dir}"
+                )
+            else:
+                print(
+                    f"Model {model_name} directory not found at {model_cache_dir} - may have been manually deleted"
+                )
+
+            # Clean up user models registry
+            if model_name.startswith("user.") and os.path.exists(USER_MODELS_FILE):
+                with open(USER_MODELS_FILE, "r", encoding="utf-8") as file:
+                    user_models = json.load(file)
+
+                base_model_name = model_name[5:]  # Remove "user." prefix
+                if base_model_name in user_models:
+                    del user_models[base_model_name]
+                    with open(USER_MODELS_FILE, "w", encoding="utf-8") as file:
+                        json.dump(user_models, file)
+                    print(f"Removed {model_name} from user models registry")
+
+            return
         # Parse checkpoint to get base and variant
         base_checkpoint, variant = parse_checkpoint(checkpoint)
 
@@ -592,6 +721,38 @@ class ModelManager:
                     json.dump(user_models, file)
                 print(f"Removed {model_name} from user models registry")
 
+    def get_incompatible_ryzenai_models(self):
+        """
+        Get information about incompatible RyzenAI models in the cache.
+
+        Returns:
+            dict with 'models' list and 'total_size' info
+        """
+        # Get HF_HOME from environment
+        hf_home = os.environ.get("HF_HOME", None)
+
+        incompatible_models, total_size = detect_incompatible_ryzenai_models(
+            DEFAULT_CACHE_DIR, hf_home
+        )
+
+        return {
+            "models": incompatible_models,
+            "total_size": total_size,
+            "count": len(incompatible_models),
+        }
+
+    def cleanup_incompatible_models(self, model_paths: list):
+        """
+        Delete incompatible RyzenAI models from the cache.
+
+        Args:
+            model_paths: List of model paths to delete
+
+        Returns:
+            dict with deletion results
+        """
+        return delete_incompatible_models(model_paths)
+
 
 # This file was originally licensed under Apache 2.0. It has been modified.
 # Modifications Copyright (c) 2025 AMD

lemonade_server/pydantic_models.py

@@ -70,6 +70,7 @@ class CompletionRequest(BaseModel):
     top_k: int | None = None
     top_p: float | None = None
     max_tokens: int | None = None
+    enable_thinking: bool | None = True
 
 
 class ChatCompletionRequest(BaseModel):
@@ -93,6 +94,7 @@ class ChatCompletionRequest(BaseModel):
     max_tokens: int | None = None
     max_completion_tokens: int | None = None
     response_format: dict | None = None
+    enable_thinking: bool | None = True
 
 
 class EmbeddingsRequest(BaseModel):
@@ -132,6 +134,7 @@ class ResponsesRequest(BaseModel):
     top_k: int | None = None
     top_p: float | None = None
     stream: bool = False
+    enable_thinking: bool | None = True
 
 
 class PullConfig(LoadConfig):