azureml-registry-tools 0.1.0a27__py3-none-any.whl → 0.1.0a29__py3-none-any.whl

azureml/registry/data/description.md.template

@@ -1,2 +1,65 @@
-<!-- REMOVE THESE HEADERS AFTER UPDATING -->
-<!-- `description.md` is required. -->
+# Description
+
+Include 1-2 sentences as your core value prop of the model.
+
+## Azure Direct Models
+For Microsoft PMs: Include 1-2 sentences as your core value prop of Azure Direct Models.
+
+# Technical specs
+Short paragraph describing the model and model family.
+
+### Training cut-off date
+Date the training was finished.
+
+### Input formats
+Specify the preferred input format for interacting with the model, especially if it was trained on fine-tuned on structured prompts. Provide response output schema to illustrated expected format.
+
+### Supported language
+Supported input human languages.
+
+## Supported Azure regions
+List of supported Azure regions.
+
+## Sample JSON response
+Input and output.
+
+## Model architecture
+10-20 word description of the model architecture.
+
+# Long context
+Indicate whether the model supports extended context lengths and describe the types of tasks this enables.
+
+## Optimizing model performance
+Describes the methods and best practices used to improve a model's efficiency, accuracy, and cost-effectiveness in real-world use.
+
+## Additional assets
+List of additional assets (e.g. training data, technical reports data processing code, model training code, model inference code, model evaluation code), if any, that are made available with a link, description of how each can be accessed and what licenses, if any, relate to their use.
+
+# Key capabilities
+
+## About this model
+A blurb to share more about the model, what it excels in, why it's valuable for developers.
+
+## Key model capabilities
+List of 3-4 valuable core capabilities by model, include a value proposition statement.
+
+# Pricing
+Pricing is based on a number of factors. See pricing details here. (Please link the last sentence to ACOM pricing page)
+
+# Use cases
+See Responsible AI for additional consideration for responsible use.
+
+## Key use cases
+Key use cases are the main practical applications of a model. Industry and task specific. Short paragraph describing the intended uses of the model. Highlight specific scenarios where the model excels.
+
+## Out of scope use cases
+Short paragraph describing model limitations constraints and identifying any restricted or prohibited uses.
+
+# Distribution channels
+A list of the methods of distribution (e.g. enterprise or subscription-based access through existing software suites or enterprise-specific solutions; public or subscription-based access through an API; public or proprietary access through integrated development environments, device-specific applications or firmware, open-source repositories) through which the model can be made available to downstream providers both within and outside the EU.
+
+# Azure Direct Models
+For Microsoft PMs: Include 1-2 sentences as your core value prop of Azure Direct Models.
+
+# More information
+Whatever you would like to go here!

azureml/registry/data/evaluation.md.template

@@ -1,2 +1,10 @@
-<!-- REMOVE THESE HEADERS AFTER UPDATING -->
-<!-- `evaluation.md` is highly recommended, but not required. It captures information about the performance of the model. We highly recommend including this section as this information is often used to decide what model to use. -->
+# Benchmarks
+
+## Quality and performance evaluations
+Summarize the model’s performance across public and internal benchmarks and describe how it compares to other third-party models. Include the types of benchmarks used, grouped by capability areas (reasoning, language understanding, math, multilingual, etc.) and limitations (knowledge capacity, etc.). Include specific datasets and tasks that demonstrate the model’s performance across diverse domains, highlight any custom or adversarial evaluations, and note any planned mitigations
+
+## Benchmarking methodology
+Share additional details about your team’s benchmarking methodology, including how prompts are standardized across models for fair comparison. Note any exceptions and clarify what is and isn’t allowed when adapting prompts. If needed, organize the appendix (A, B, C, etc.) based on specific benchmarks such as robustness, short & long context, multilingual, etc.
+
+## Public data summary
+Link to the relevant public data summary or summaries for this model

azureml/registry/data/model.schema.json

@@ -251,7 +251,7 @@
         "inputModalities": {
           "description": "Input modalities supported (e.g., text, image) as a comma-separated string",
           "type": "string",
-          "pattern": "^(audio|csv|embeddings|image|json|pdf|text|video)(?:\\s*,\\s*(audio|csv|embeddings|image|json|pdf|text|video))*$"
+          "pattern": "^(audio|csv|embeddings|image|json|pdf|text|video|code)(?:\\s*,\\s*(audio|csv|embeddings|image|json|pdf|text|video|code))*$"
         },
         "inference_supported_envs": {
           "description": "Supported inference environments",
@@ -361,7 +361,7 @@
         "modelCapabilities": {
           "description": "Model capabilities (e.g., agents, assistants) as a comma-separated string",
           "type": "string",
-          "pattern": "^(agents|assistants|routing|reasoning|streaming|tool-calling)(?:\\s*,\\s*(agents|assistants|routing|reasoning|streaming|tool-calling))*$"
+          "pattern": "^(agents|agentsV2|assistants|routing|reasoning|streaming|tool-calling|function-calling|image-input)(?:\\s*,\\s*(agents|agentsV2|assistants|routing|reasoning|streaming|tool-calling|function-calling|image-input))*$"
         },
         "modelHash": {
           "description": "Hash of the model",
@@ -395,7 +395,7 @@
         "task": {
           "description": "Tasks supported by the model as a comma-separated string",
           "type": "string",
-          "pattern": "^(audio-analysis|audio-classification|audio-generation|automatic-speech-recognition|chat-completion|completions|content-filters|content-safety|conversational-ai|custom-extraction|data-generation|document-analysis|document-ingestion|document-translation|embeddings|face-detection|fill-mask|forecasting|image-analysis|image-classification|image-feature-extraction|image-text-to-text|image-to-image|image-to-text|intelligent-content-processing|intelligent-document-processing|text-pii-extraction|conversation-pii-extraction|document-pii-extraction|detect-language|optical-character-recognition|protein-sequence-generation|protein-structure-prediction|responses|responsible-ai|retrosynthesis-prediction|summarization|text-analysis|text-analytics|text-classification|text-generation|text-to-image|text-to-speech|time-series-forecasting|translation|speech-to-text|speech-translation|video-analysis|video-generation|video-text-to-text|visual-question-answering|zero-shot-classification|zero-shot-image-classification|materials-design|atomistic-modelling|image-to-3D|text-to-3D|3D-generation|task-completion-verification|action-affordance|next-plausible-action-prediction|Structure-Prediction|Genomics|biomolecular-complex-structure-prediction|structure-prediction|protein-folding)(?:\\s*,\\s*(audio-analysis|audio-classification|audio-generation|automatic-speech-recognition|chat-completion|completions|content-filters|content-safety|conversational-ai|custom-extraction|data-generation|document-analysis|document-ingestion|document-translation|embeddings|face-detection|fill-mask|forecasting|image-analysis|image-classification|image-feature-extraction|image-text-to-text|image-to-image|image-to-text|intelligent-content-processing|intelligent-document-processing|optical-character-recognition|protein-sequence-generation|protein-structure-prediction|responses|responsible-ai|summarization|text-analysis|text-analytics|text-classification|text-generation|text-pii-extraction|conversation-pii-extraction|document-pii-extraction|detect-language|text-to-image|text-to-speech|time-series-forecasting|translation|speech-to-text|speech-translation|video-analysis|video-generation|video-text-to-text|visual-question-answering|zero-shot-classification|zero-shot-image-classification|materials-design|atomistic-modelling|image-to-3D|text-to-3D|3D-generation|task-completion-verification|action-affordance|next-plausible-action-prediction|Structure-Prediction|Genomics|biomolecular-complex-structure-prediction|structure-prediction|protein-folding))*$"
+          "pattern": "^(audio-analysis|audio-classification|audio-generation|automatic-speech-recognition|chat-completion|completions|content-filters|content-safety|conversational-ai|custom-extraction|data-generation|document-analysis|document-ingestion|document-translation|embeddings|face-detection|fill-mask|forecasting|image-analysis|image-classification|image-feature-extraction|image-text-to-text|image-to-image|image-to-text|intelligent-content-processing|intelligent-document-processing|text-pii-extraction|conversation-pii-extraction|document-pii-extraction|detect-language|optical-character-recognition|protein-sequence-generation|protein-structure-prediction|responses|responsible-ai|retrosynthesis-prediction|summarization|text-analysis|text-analytics|text-classification|text-generation|text-to-image|text-to-speech|time-series-forecasting|translation|speech-to-text|speech-translation|video-analysis|video-generation|video-text-to-text|visual-question-answering|zero-shot-classification|zero-shot-image-classification|materials-design|atomistic-modelling|image-to-3D|text-to-3D|3D-generation|task-completion-verification|action-affordance|next-plausible-action-prediction|Structure-Prediction|Genomics|biomolecular-complex-structure-prediction|structure-prediction|protein-folding|web-agent-tasks|gui-grounding|messages)(?:\\s*,\\s*(audio-analysis|audio-classification|audio-generation|automatic-speech-recognition|chat-completion|completions|content-filters|content-safety|conversational-ai|custom-extraction|data-generation|document-analysis|document-ingestion|document-translation|embeddings|face-detection|fill-mask|forecasting|image-analysis|image-classification|image-feature-extraction|image-text-to-text|image-to-image|image-to-text|intelligent-content-processing|intelligent-document-processing|optical-character-recognition|protein-sequence-generation|protein-structure-prediction|responses|responsible-ai|summarization|text-analysis|text-analytics|text-classification|text-generation|text-pii-extraction|conversation-pii-extraction|document-pii-extraction|detect-language|text-to-image|text-to-speech|time-series-forecasting|translation|speech-to-text|speech-translation|video-analysis|video-generation|video-text-to-text|visual-question-answering|zero-shot-classification|zero-shot-image-classification|materials-design|atomistic-modelling|image-to-3D|text-to-3D|3D-generation|task-completion-verification|action-affordance|next-plausible-action-prediction|Structure-Prediction|Genomics|biomolecular-complex-structure-prediction|structure-prediction|protein-folding|web-agent-tasks|gui-grounding|messages))*$"
         },
         "textContextWindow": {
           "description": "Context window size",
@@ -593,11 +593,14 @@
             "type": "string",
             "enum": [
               "agents",
+              "agentsV2",
               "assistants",
               "routing",
               "reasoning",
               "streaming",
-              "tool-calling"
+              "tool-calling",
+              "function-calling",
+              "image-input"
             ]
           },
           "description": "Model capabilities (e.g., agents, assistants)"

azureml/registry/data/notes.md.template

@@ -1,2 +1,22 @@
-<!-- REMOVE THESE HEADERS AFTER UPDATING -->
-<!-- `notes.md` is highly recommended, but not required. It captures information about how your model is created. We highly recommend including this section to provide transparency for the customers. -->
+# Responsible AI considerations
+
+## Safety techniques
+Policies, techniques and safety frameworks applied to enable model hosting on Foundry (i.e. Provenance). 
+
+Describe the safety alignment strategy used during post-training. Include the types of datasets leveraged, specify the techniques applied along with the safety objectives they target.
+
+## Safety evaluations
+Describe the evaluation methods used to assess model safety prior to release. Include both quantitative and qualitive approaches and specify the risk categories evaluated [i.e.  disallowed content (sexual, violent, hateful, or self-harm content), copyright content/IP, and jailbreaks] and any collaboration with internal or external safety teams. Refer to other relevant documentation for more details.
+
+## Known limitations
+Outline known limitations and potential risks associated with the model, including fairness, representation, offensive content, reliability, and misuse. Clearly state areas where the model may underperform (non-English languages, sensitive domains, etc.). Provide specific guidance for developers on applying responsible AI practices, legal compliance, and appropriate safeguards for high-risk or consequential use cases.
+
+# Acceptable use
+
+## Acceptable use policy
+Link to any relevant acceptable use policies. Otherwise state N/A.
+
+# Terms of Service
+
+## Terms of Service Link
+Type or category of license (e.g. free/open source or proprietary). If there is no license, describe how access to the model is provided. State the license.

azureml/registry/mgmt/asset_management.py

@@ -18,6 +18,7 @@ from azure.identity import DefaultAzureCredential
 from azureml.registry.data.validate_model_schema import validate_model_schema
 from azureml.registry.data.validate_model_variant_schema import validate_model_variant_schema
 from azureml.registry._rest_client.registry_management_client import RegistryManagementClient
+from azureml.registry.mgmt.util import resolve_from_file_for_asset
 
 # Windows compatibility patch - must be applied before importing azureml.assets
 from subprocess import run
@@ -98,6 +99,9 @@ def put_system_metadata(ml_client: MLClient, asset: AssetConfig, registry: str,
         registry (str): Name of registry.
         system_metadata (dict): System metadata payload.
     """
+    # First, transform system metadata such that any files are read and their content used
+    system_metadata = {k: resolve_from_file_for_asset(asset, v) for k, v in system_metadata.items()}
+
     # Use RegistryManagementClient for discovery
     registry_mgmt_client = RegistryManagementClient(registry_name=registry)
     discovery = registry_mgmt_client.discovery()
@@ -120,7 +124,7 @@ def put_system_metadata(ml_client: MLClient, asset: AssetConfig, registry: str,
     }
 
     print("Attempting to PUT system metadata")
-    print(f"Payload: {system_metadata}")
+    print(f"System Metadata Payload: {system_metadata}")
 
     response = requests.put(url, headers=headers, json=system_metadata)
 

azureml/registry/mgmt/util.py

@@ -0,0 +1,75 @@
+"""File resolution utilities for asset management."""
+
+import os
+from pathlib import Path, PurePath
+from typing import Tuple, Union, Any
+
+import azureml.assets as assets  # noqa: E402
+
+
+def is_file_relative_to_asset_path(asset: assets.AssetConfig, value: Any) -> bool:
+    """Check if the value is a file with respect to the asset path.
+
+    Args:
+        asset (AssetConfig): the asset to try and resolve the value for
+        value: value to check
+
+    Returns:
+        bool: True if value represents a file relative to asset path, False otherwise
+    """
+    if not isinstance(value, str) and not isinstance(value, PurePath):
+        return False
+
+    path_value = value if isinstance(value, Path) else Path(value)
+
+    if not path_value.is_relative_to(asset.file_path):
+        path_value = asset._append_to_file_path(path_value)
+
+    return os.path.isfile(path_value)
+
+
+def resolve_from_file_for_asset(asset: assets.AssetConfig, value: Any) -> Any:
+    """Resolve the value from a file for an asset if it is a file, otherwise returns the value.
+
+    Args:
+        asset (AssetConfig): the asset to try and resolve the value for
+        value: value to try and resolve
+
+    Returns:
+        Any: File content if value is a file path relative to asset, otherwise the original value
+    """
+    if not is_file_relative_to_asset_path(asset, value):
+        return value
+
+    path_value = value if isinstance(value, Path) else Path(value)
+
+    if not path_value.is_relative_to(asset.file_path):
+        path_value = asset._append_to_file_path(path_value)
+
+    is_resolved_from_file, resolved_value = _resolve_from_file(path_value)
+
+    if is_resolved_from_file:
+        return resolved_value
+    else:
+        return value
+
+
+def _resolve_from_file(value: Union[str, Path]) -> Tuple[bool, Union[str, None]]:
+    """Resolve file content (internal helper).
+
+    Args:
+        value: File path to resolve
+
+    Returns:
+        Tuple[bool, Union[str, None]]: (success, content) where success indicates
+                                       if file was read and content is the file content
+    """
+    if os.path.isfile(value):
+        try:
+            with open(value, 'r') as f:
+                content = f.read()
+                return (True, content)
+        except Exception as e:
+            raise Exception(f"Failed to read file {value}: {e}")
+    else:
+        return (False, None)

{azureml_registry_tools-0.1.0a27.dist-info → azureml_registry_tools-0.1.0a29.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: azureml-registry-tools
-Version: 0.1.0a27
+Version: 0.1.0a29
 Summary: AzureML Registry tools and CLI
 Author: Microsoft Corp
 License: https://aka.ms/azureml-sdk-license

{azureml_registry_tools-0.1.0a27.dist-info → azureml_registry_tools-0.1.0a29.dist-info}/RECORD

@@ -10,29 +10,30 @@ azureml/registry/_rest_client/registry_management_client.py,sha256=NsqWRaDdOlPyI
 azureml/registry/_rest_client/registry_model_client.py,sha256=LqJGTuYQtBnHSeSbOl5KVbiO6vGBkKQ7HD5jc4fvV3k,14466
 azureml/registry/data/__init__.py,sha256=cW3X6HATz6XF-K_7uKdybTbJb9EZSecBN4J27NGdZmU,231
 azureml/registry/data/asset.yaml.template,sha256=WTgfuvKEBp-EVFSQ0JpU0h4z_ULJdULO9kHmB_9Fr1o,96
-azureml/registry/data/description.md.template,sha256=wQLk54U8hoXU1y9235R4irc6FGYPXGO-x9EHUorP15Q,84
-azureml/registry/data/evaluation.md.template,sha256=JaDecIfLV9vZDUrZzVRPzVHHnKD-BQGBgQ-cw1dHavU,277
+azureml/registry/data/description.md.template,sha256=DiVAQEXGXoKmhV4LPqE3NupxVtcsuDZ1pn2UA5Fzd6U,2821
+azureml/registry/data/evaluation.md.template,sha256=FC9U8EI_1Dg9Vz18ftAFDDmTqvpwELDkIDlYqB8C9Dk,1031
 azureml/registry/data/model-variant.schema.json,sha256=AT4Dy6cCtp_SFUfSqYIqcER8AldpYm0QIEy1abY3QWE,1699
-azureml/registry/data/model.schema.json,sha256=LAclWqHYMY0f62Hp33tGETjv2mjSffAwVHNIuXngp6o,46832
+azureml/registry/data/model.schema.json,sha256=GbpT9kqZ-7t_2YxoU4QX1ojI_yo9XJfR2bAhrG0LgZc,47085
 azureml/registry/data/model.yaml.template,sha256=h5uqAN22FLaWrbPxIb8yVKH9cGDBrIwooXYYfsKhxDw,245
-azureml/registry/data/notes.md.template,sha256=zSRyOR__9NGL2j0tugY7HgFkwkAdcE2pJyyyGsz1SAk,248
+azureml/registry/data/notes.md.template,sha256=rgGGHQaxfVg6COIzZU8EVBa48sRPnNezVkCMGNyRRNo,1528
 azureml/registry/data/validate_model_schema.py,sha256=OQp2E01kdxSphvUQYQvelSiD24-qUG6nTFuzW60wX2c,8322
 azureml/registry/data/validate_model_variant_schema.py,sha256=JPVNtRBn6qciMu4PaRXOvS86OGGW0cocL2Rri4xYKo8,3629
 azureml/registry/mgmt/__init__.py,sha256=LMhqcEC8ItmmpKZljElGXH-6olHlT3SLl0dJU01OvuM,226
-azureml/registry/mgmt/asset_management.py,sha256=kt607xr81-sTa7qq5ndpVfq4IaCutNsrqZWmmBoZz2g,14189
+azureml/registry/mgmt/asset_management.py,sha256=NeYjjtOFlXJT1c0s9pW5mBjJgyxQ7zqZbB-YzmNOg9s,14465
 azureml/registry/mgmt/create_asset_template.py,sha256=ejwLuIsmzJOoUePoxbM-eGMg2E3QHfdX-nPMBzYUVMQ,3525
 azureml/registry/mgmt/create_manifest.py,sha256=N9wRmjAKO09A3utN_lCUsM_Ufpj7PL0SJz-XHPHWuyM,9528
 azureml/registry/mgmt/create_model_spec.py,sha256=1PdAcUf-LomvljoT8wKQihXMTLd7DoTgN0qDX4Lol1A,10473
 azureml/registry/mgmt/model_management.py,sha256=STTr_uvdPKV2NaJ5UvS5aMi3yejVF6Hkj9DjofJLQik,7453
 azureml/registry/mgmt/syndication_manifest.py,sha256=8Sfd49QuCA5en5_mIOLE21kZVpnReUXowx_g0TVRgWg,9025
+azureml/registry/mgmt/util.py,sha256=BeVUsiMbZdO6rc9tHlsFg0AkVX62lzKn4Ko_y329r-I,2385
 azureml/registry/tools/__init__.py,sha256=IAuWWpGfZm__pAkBIxmpJz84QskpkxBr0yDk1TUSnkE,223
 azureml/registry/tools/config.py,sha256=tjPaoBsWtPXBL8Ww1hcJtsr2SuIjPKt79dR8iovcebg,3639
 azureml/registry/tools/create_or_update_assets.py,sha256=7LcuBzwU-HNE9ADG2igFXI696aKR028yTYxMuDtjVmA,16095
 azureml/registry/tools/registry_utils.py,sha256=zgYlCiOONtQJ4yZ9wg8tKVoE8dh6rrjB8hYBGhpV9-0,1403
 azureml/registry/tools/repo2registry_config.py,sha256=eXp_tU8Jyi30g8xGf7wbpLgKEPpieohBANKxMSLzq7s,4873
-azureml_registry_tools-0.1.0a27.dist-info/licenses/LICENSE.txt,sha256=n20rxwp7_NGrrShv9Qvcs90sjI1l3Pkt3m-5OPCWzgs,845
-azureml_registry_tools-0.1.0a27.dist-info/METADATA,sha256=vLbpMUGdaBjXBOIPby_z9ZtQ3xgdbizApSaYpcVccDQ,522
-azureml_registry_tools-0.1.0a27.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-azureml_registry_tools-0.1.0a27.dist-info/entry_points.txt,sha256=iRUkAeQidMnO6RQzpLqMUBTcyYtNzAfSin9WnSdVGLw,147
-azureml_registry_tools-0.1.0a27.dist-info/top_level.txt,sha256=ZOeEa0TAXo6i5wOjwBoqfIGEuxOcKuscGgNSpizqREY,8
-azureml_registry_tools-0.1.0a27.dist-info/RECORD,,
+azureml_registry_tools-0.1.0a29.dist-info/licenses/LICENSE.txt,sha256=n20rxwp7_NGrrShv9Qvcs90sjI1l3Pkt3m-5OPCWzgs,845
+azureml_registry_tools-0.1.0a29.dist-info/METADATA,sha256=l6pLVmMr-nw8Cd7XPT2XSlcxd2r6prXEkAK8ch6OGqE,522
+azureml_registry_tools-0.1.0a29.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+azureml_registry_tools-0.1.0a29.dist-info/entry_points.txt,sha256=iRUkAeQidMnO6RQzpLqMUBTcyYtNzAfSin9WnSdVGLw,147
+azureml_registry_tools-0.1.0a29.dist-info/top_level.txt,sha256=ZOeEa0TAXo6i5wOjwBoqfIGEuxOcKuscGgNSpizqREY,8
+azureml_registry_tools-0.1.0a29.dist-info/RECORD,,