diff --git a/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.html b/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.html index abd0612e7..e611b2bc3 100644 --- a/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.html +++ b/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.html @@ -21,7 +21,7 @@ "info": { "title": "[DRAFT] Llama Stack Specification", "version": "0.0.1", - "description": "This is the specification of the llama stack that provides\n a set of endpoints and their corresponding interfaces that are tailored to\n best leverage Llama Models. The specification is still in draft and subject to change.\n Generated at 2024-09-11 12:34:17.939867" + "description": "This is the specification of the llama stack that provides\n a set of endpoints and their corresponding interfaces that are tailored to\n best leverage Llama Models. The specification is still in draft and subject to change.\n Generated at 2024-09-11 14:17:44.361462" }, "servers": [ { @@ -51,7 +51,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/BatchChatCompletionRequestWrapper" + "$ref": "#/components/schemas/BatchChatCompletionRequest" } } }, @@ -81,7 +81,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/BatchCompletionRequestWrapper" + "$ref": "#/components/schemas/BatchCompletionRequest" } } }, @@ -277,7 +277,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/AgenticSystemTurnCreateRequestWrapper" + "$ref": "#/components/schemas/CreateAgenticSystemTurnRequest" } } }, @@ -300,7 +300,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/CreateDatasetRequestWrapper" + "$ref": "#/components/schemas/CreateDatasetRequest" } } }, @@ -330,7 +330,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/CreateExperimentRequestWrapper" + "$ref": "#/components/schemas/CreateExperimentRequest" } } }, @@ -390,7 +390,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/CreateRunRequestWrapper" + "$ref": "#/components/schemas/CreateRunRequest" } } }, @@ -572,7 +572,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/EvaluateQuestionAnsweringRequestWrapper" + "$ref": "#/components/schemas/EvaluateQuestionAnsweringRequest" } } }, @@ -602,7 +602,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/EvaluateSummarizationRequestWrapper" + "$ref": "#/components/schemas/EvaluateSummarizationRequest" } } }, @@ -632,7 +632,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/EvaluateTextGenerationRequestWrapper" + "$ref": "#/components/schemas/EvaluateTextGenerationRequest" } } }, @@ -1019,12 +1019,21 @@ "tags": [ "Telemetry" ], - "parameters": [], + "parameters": [ + { + "name": "query", + "in": "query", + "required": true, + "schema": { + "type": "string" + } + } + ], "requestBody": { "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/LogSearchRequestWrapper" + "$ref": "#/components/schemas/GetLogsRequest" } } }, @@ -1312,7 +1321,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/LogMessagesRequestWrapper" + "$ref": "#/components/schemas/LogMessagesRequest" } } }, @@ -1335,7 +1344,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/LogMetricsRequestWrapper" + "$ref": "#/components/schemas/LogMetricsRequest" } } }, @@ -1365,7 +1374,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/PostTrainingRLHFRequestWrapper" + "$ref": "#/components/schemas/PreferenceOptimizeRequest" } } }, @@ -1425,7 +1434,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/RewardScoringRequestWrapper" + "$ref": "#/components/schemas/RewardScoreRequest" } } }, @@ -1455,7 +1464,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/RunShieldRequestWrapper" + "$ref": "#/components/schemas/RunShieldsRequest" } } }, @@ -1485,7 +1494,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/PostTrainingSFTRequestWrapper" + "$ref": "#/components/schemas/SupervisedFineTuneRequest" } } }, @@ -1515,7 +1524,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/SyntheticDataGenerationRequestWrapper" + "$ref": "#/components/schemas/SyntheticDataGenerateRequest" } } }, @@ -1568,7 +1577,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/UpdateExperimentRequestWrapper" + "$ref": "#/components/schemas/UpdateExperimentRequest" } } }, @@ -1598,7 +1607,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/UpdateRunRequestWrapper" + "$ref": "#/components/schemas/UpdateRunRequest" } } }, @@ -1628,7 +1637,7 @@ "content": { "application/json": { "schema": { - "$ref": "#/components/schemas/UploadArtifactRequestWrapper" + "$ref": "#/components/schemas/UploadArtifactRequest" } } }, @@ -1640,65 +1649,6 @@ "jsonSchemaDialect": "https://json-schema.org/draft/2020-12/schema", "components": { "schemas": { - "BatchChatCompletionRequest": { - "type": "object", - "properties": { - "model": { - "type": "string" - }, - "messages_batch": { - "type": "array", - "items": { - "type": "array", - "items": { - "oneOf": [ - { - "$ref": "#/components/schemas/UserMessage" - }, - { - "$ref": "#/components/schemas/SystemMessage" - }, - { - "$ref": "#/components/schemas/ToolResponseMessage" - }, - { - "$ref": "#/components/schemas/CompletionMessage" - } - ] - } - } - }, - "sampling_params": { - "$ref": "#/components/schemas/SamplingParams" - }, - "tools": { - "type": "array", - "items": { - "$ref": "#/components/schemas/ToolDefinition" - } - }, - "tool_choice": { - "$ref": "#/components/schemas/ToolChoice" - }, - "tool_prompt_format": { - "$ref": "#/components/schemas/ToolPromptFormat" - }, - "logprobs": { - "type": "object", - "properties": { - "top_k": { - "type": "integer" - } - }, - "additionalProperties": false - } - }, - "additionalProperties": false, - "required": [ - "model", - "messages_batch" - ] - }, "BuiltinTool": { "type": "string", "enum": [ @@ -2050,16 +2000,63 @@ "content" ] }, - "BatchChatCompletionRequestWrapper": { + "BatchChatCompletionRequest": { "type": "object", "properties": { - "request": { - "$ref": "#/components/schemas/BatchChatCompletionRequest" + "model": { + "type": "string" + }, + "messages_batch": { + "type": "array", + "items": { + "type": "array", + "items": { + "oneOf": [ + { + "$ref": "#/components/schemas/UserMessage" + }, + { + "$ref": "#/components/schemas/SystemMessage" + }, + { + "$ref": "#/components/schemas/ToolResponseMessage" + }, + { + "$ref": "#/components/schemas/CompletionMessage" + } + ] + } + } + }, + "sampling_params": { + "$ref": "#/components/schemas/SamplingParams" + }, + "tools": { + "type": "array", + "items": { + "$ref": "#/components/schemas/ToolDefinition" + } + }, + "tool_choice": { + "$ref": "#/components/schemas/ToolChoice" + }, + "tool_prompt_format": { + "$ref": "#/components/schemas/ToolPromptFormat" + }, + "logprobs": { + "type": "object", + "properties": { + "top_k": { + "type": "integer" + } + }, + "additionalProperties": false } }, "additionalProperties": false, "required": [ - "request" + "model", + "messages_batch" ] }, "BatchChatCompletionResponse": { @@ -2118,18 +2115,6 @@ "content_batch" ] }, - "BatchCompletionRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/BatchCompletionRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "BatchCompletionResponse": { "type": "object", "properties": { @@ -3007,234 +2992,38 @@ "session_id" ] }, - "AgenticSystemTurnCreateRequest": { + "Attachment": { "type": "object", "properties": { - "sampling_params": { - "$ref": "#/components/schemas/SamplingParams" - }, - "input_shields": { - "type": "array", - "items": { - "$ref": "#/components/schemas/ShieldDefinition" - } - }, - "output_shields": { - "type": "array", - "items": { - "$ref": "#/components/schemas/ShieldDefinition" - } - }, - "tools": { - "type": "array", - "items": { - "oneOf": [ - { - "$ref": "#/components/schemas/SearchToolDefinition" - }, - { - "$ref": "#/components/schemas/WolframAlphaToolDefinition" - }, - { - "$ref": "#/components/schemas/PhotogenToolDefinition" - }, - { - "$ref": "#/components/schemas/CodeInterpreterToolDefinition" - }, - { - "$ref": "#/components/schemas/FunctionCallToolDefinition" - }, - { - "type": "object", - "properties": { - "input_shields": { - "type": "array", - "items": { - "$ref": "#/components/schemas/ShieldDefinition" - } - }, - "output_shields": { - "type": "array", - "items": { - "$ref": "#/components/schemas/ShieldDefinition" - } - }, - "type": { - "type": "string", - "const": "memory" - }, - "memory_bank_configs": { - "type": "array", - "items": { - "oneOf": [ - { - "type": "object", - "properties": { - "bank_id": { - "type": "string" - }, - "type": { - "type": "string", - "const": "vector" - } - }, - "additionalProperties": false, - "required": [ - "bank_id", - "type" - ] - }, - { - "type": "object", - "properties": { - "bank_id": { - "type": "string" - }, - "type": { - "type": "string", - "const": "keyvalue" - }, - "keys": { - "type": "array", - "items": { - "type": "string" - } - } - }, - "additionalProperties": false, - "required": [ - "bank_id", - "type", - "keys" - ] - }, - { - "type": "object", - "properties": { - "bank_id": { - "type": "string" - }, - "type": { - "type": "string", - "const": "keyword" - } - }, - "additionalProperties": false, - "required": [ - "bank_id", - "type" - ] - }, - { - "type": "object", - "properties": { - "bank_id": { - "type": "string" - }, - "type": { - "type": "string", - "const": "graph" - }, - "entities": { - "type": "array", - "items": { - "type": "string" - } - } - }, - "additionalProperties": false, - "required": [ - "bank_id", - "type", - "entities" - ] - } - ] - } - }, - "query_generator_config": { - "oneOf": [ - { - "type": "object", - "properties": { - "type": { - "type": "string", - "const": "default" - }, - "sep": { - "type": "string" - } - }, - "additionalProperties": false, - "required": [ - "type", - "sep" - ] - }, - { - "type": "object", - "properties": { - "type": { - "type": "string", - "const": "llm" - }, - "model": { - "type": "string" - }, - "template": { - "type": "string" - } - }, - "additionalProperties": false, - "required": [ - "type", - "model", - "template" - ] - }, - { - "type": "object", - "properties": { - "type": { - "type": "string", - "const": "custom" - } - }, - "additionalProperties": false, - "required": [ - "type" - ] - } - ] - }, - "max_tokens_in_context": { - "type": "integer" - }, - "max_chunks": { - "type": "integer" - } - }, - "additionalProperties": false, - "required": [ - "type", - "memory_bank_configs", - "query_generator_config", - "max_tokens_in_context", - "max_chunks" - ] + "content": { + "oneOf": [ + { + "type": "string" + }, + { + "type": "array", + "items": { + "type": "string" } - ] - } + }, + { + "$ref": "#/components/schemas/URL" + } + ] }, - "tool_choice": { - "$ref": "#/components/schemas/ToolChoice" - }, - "tool_prompt_format": { - "$ref": "#/components/schemas/ToolPromptFormat" - }, - "instructions": { + "mime_type": { "type": "string" - }, + } + }, + "additionalProperties": false, + "required": [ + "content", + "mime_type" + ] + }, + "CreateAgenticSystemTurnRequest": { + "type": "object", + "properties": { "agent_id": { "type": "string" }, @@ -3271,47 +3060,6 @@ "messages" ] }, - "Attachment": { - "type": "object", - "properties": { - "content": { - "oneOf": [ - { - "type": "string" - }, - { - "type": "array", - "items": { - "type": "string" - } - }, - { - "$ref": "#/components/schemas/URL" - } - ] - }, - "mime_type": { - "type": "string" - } - }, - "additionalProperties": false, - "required": [ - "content", - "mime_type" - ] - }, - "AgenticSystemTurnCreateRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/AgenticSystemTurnCreateRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "AgenticSystemTurnResponseEvent": { "type": "object", "properties": { @@ -3811,23 +3559,6 @@ ], "title": "A single turn in an interaction with an Agentic System." }, - "CreateDatasetRequest": { - "type": "object", - "properties": { - "uuid": { - "type": "string" - }, - "dataset": { - "$ref": "#/components/schemas/TrainEvalDataset" - } - }, - "additionalProperties": false, - "required": [ - "uuid", - "dataset" - ], - "title": "Request to create a dataset." - }, "TrainEvalDataset": { "type": "object", "properties": { @@ -3883,16 +3614,20 @@ "json" ] }, - "CreateDatasetRequestWrapper": { + "CreateDatasetRequest": { "type": "object", "properties": { - "request": { - "$ref": "#/components/schemas/CreateDatasetRequest" + "uuid": { + "type": "string" + }, + "dataset": { + "$ref": "#/components/schemas/TrainEvalDataset" } }, "additionalProperties": false, "required": [ - "request" + "uuid", + "dataset" ] }, "CreateExperimentRequest": { @@ -3932,18 +3667,6 @@ "name" ] }, - "CreateExperimentRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/CreateExperimentRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "Experiment": { "type": "object", "properties": { @@ -4216,18 +3939,6 @@ "experiment_id" ] }, - "CreateRunRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/CreateRunRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "Run": { "type": "object", "properties": { @@ -4401,24 +4112,9 @@ "embeddings" ] }, - "Checkpoint": { - "description": "Checkpoint created during training runs" - }, "EvaluateQuestionAnsweringRequest": { "type": "object", "properties": { - "job_uuid": { - "type": "string" - }, - "dataset": { - "$ref": "#/components/schemas/TrainEvalDataset" - }, - "checkpoint": { - "$ref": "#/components/schemas/Checkpoint" - }, - "sampling_params": { - "$ref": "#/components/schemas/SamplingParams" - }, "metrics": { "type": "array", "items": { @@ -4432,24 +4128,7 @@ }, "additionalProperties": false, "required": [ - "job_uuid", - "dataset", - "checkpoint", - "sampling_params", "metrics" - ], - "title": "Request to evaluate question answering." - }, - "EvaluateQuestionAnsweringRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/EvaluateQuestionAnsweringRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" ] }, "EvaluationJob": { @@ -4467,18 +4146,6 @@ "EvaluateSummarizationRequest": { "type": "object", "properties": { - "job_uuid": { - "type": "string" - }, - "dataset": { - "$ref": "#/components/schemas/TrainEvalDataset" - }, - "checkpoint": { - "$ref": "#/components/schemas/Checkpoint" - }, - "sampling_params": { - "$ref": "#/components/schemas/SamplingParams" - }, "metrics": { "type": "array", "items": { @@ -4492,41 +4159,12 @@ }, "additionalProperties": false, "required": [ - "job_uuid", - "dataset", - "checkpoint", - "sampling_params", "metrics" - ], - "title": "Request to evaluate summarization." - }, - "EvaluateSummarizationRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/EvaluateSummarizationRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" ] }, "EvaluateTextGenerationRequest": { "type": "object", "properties": { - "job_uuid": { - "type": "string" - }, - "dataset": { - "$ref": "#/components/schemas/TrainEvalDataset" - }, - "checkpoint": { - "$ref": "#/components/schemas/Checkpoint" - }, - "sampling_params": { - "$ref": "#/components/schemas/SamplingParams" - }, "metrics": { "type": "array", "items": { @@ -4541,24 +4179,7 @@ }, "additionalProperties": false, "required": [ - "job_uuid", - "dataset", - "checkpoint", - "sampling_params", "metrics" - ], - "title": "Request to evaluate text generation." - }, - "EvaluateTextGenerationRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/EvaluateTextGenerationRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" ] }, "GetAgenticSystemSessionRequest": { @@ -4809,12 +4430,9 @@ "job_uuid" ] }, - "LogSearchRequest": { + "GetLogsRequest": { "type": "object", "properties": { - "query": { - "type": "string" - }, "filters": { "type": "object", "additionalProperties": { @@ -4841,22 +4459,7 @@ } } }, - "additionalProperties": false, - "required": [ - "query" - ] - }, - "LogSearchRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/LogSearchRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] + "additionalProperties": false }, "Log": { "type": "object", @@ -4943,6 +4546,9 @@ "run_id" ] }, + "Checkpoint": { + "description": "Checkpoint created during training runs" + }, "PostTrainingJobArtifactsResponse": { "type": "object", "properties": { @@ -5117,18 +4723,6 @@ "logs" ] }, - "LogMessagesRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/LogMessagesRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "LogMetricsRequest": { "type": "object", "properties": { @@ -5148,18 +4742,6 @@ "metrics" ] }, - "LogMetricsRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/LogMetricsRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "DPOAlignmentConfig": { "type": "object", "properties": { @@ -5213,7 +4795,49 @@ "weight_decay" ] }, - "PostTrainingRLHFRequest": { + "RLHFAlgorithm": { + "type": "string", + "enum": [ + "dpo" + ] + }, + "TrainingConfig": { + "type": "object", + "properties": { + "n_epochs": { + "type": "integer" + }, + "batch_size": { + "type": "integer" + }, + "shuffle": { + "type": "boolean" + }, + "n_iters": { + "type": "integer" + }, + "enable_activation_checkpointing": { + "type": "boolean" + }, + "memory_efficient_fsdp_wrap": { + "type": "boolean" + }, + "fsdp_cpu_offload": { + "type": "boolean" + } + }, + "additionalProperties": false, + "required": [ + "n_epochs", + "batch_size", + "shuffle", + "n_iters", + "enable_activation_checkpointing", + "memory_efficient_fsdp_wrap", + "fsdp_cpu_offload" + ] + }, + "PreferenceOptimizeRequest": { "type": "object", "properties": { "job_uuid": { @@ -5303,61 +4927,6 @@ "training_config", "hyperparam_search_config", "logger_config" - ], - "title": "Request to finetune a model." - }, - "RLHFAlgorithm": { - "type": "string", - "enum": [ - "dpo" - ] - }, - "TrainingConfig": { - "type": "object", - "properties": { - "n_epochs": { - "type": "integer" - }, - "batch_size": { - "type": "integer" - }, - "shuffle": { - "type": "boolean" - }, - "n_iters": { - "type": "integer" - }, - "enable_activation_checkpointing": { - "type": "boolean" - }, - "memory_efficient_fsdp_wrap": { - "type": "boolean" - }, - "fsdp_cpu_offload": { - "type": "boolean" - } - }, - "additionalProperties": false, - "required": [ - "n_epochs", - "batch_size", - "shuffle", - "n_iters", - "enable_activation_checkpointing", - "memory_efficient_fsdp_wrap", - "fsdp_cpu_offload" - ] - }, - "PostTrainingRLHFRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/PostTrainingRLHFRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" ] }, "QueryDocumentsRequest": { @@ -5508,7 +5077,7 @@ "sampled_generations" ] }, - "RewardScoringRequest": { + "RewardScoreRequest": { "type": "object", "properties": { "dialog_generations": { @@ -5525,19 +5094,6 @@ "required": [ "dialog_generations", "model" - ], - "title": "Request to score a reward function. A list of prompts and a list of responses per prompt." - }, - "RewardScoringRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/RewardScoringRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" ] }, "RewardScoringResponse": { @@ -5620,7 +5176,7 @@ "score" ] }, - "RunShieldRequest": { + "RunShieldsRequest": { "type": "object", "properties": { "messages": { @@ -5655,18 +5211,6 @@ "shields" ] }, - "RunShieldRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/RunShieldRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "RunShieldResponse": { "type": "object", "properties": { @@ -5753,7 +5297,38 @@ "alpha" ] }, - "PostTrainingSFTRequest": { + "QLoraFinetuningConfig": { + "type": "object", + "properties": { + "lora_attn_modules": { + "type": "array", + "items": { + "type": "string" + } + }, + "apply_lora_to_mlp": { + "type": "boolean" + }, + "apply_lora_to_output": { + "type": "boolean" + }, + "rank": { + "type": "integer" + }, + "alpha": { + "type": "integer" + } + }, + "additionalProperties": false, + "required": [ + "lora_attn_modules", + "apply_lora_to_mlp", + "apply_lora_to_output", + "rank", + "alpha" + ] + }, + "SupervisedFineTuneRequest": { "type": "object", "properties": { "job_uuid": { @@ -5853,53 +5428,9 @@ "training_config", "hyperparam_search_config", "logger_config" - ], - "title": "Request to finetune a model." - }, - "QLoraFinetuningConfig": { - "type": "object", - "properties": { - "lora_attn_modules": { - "type": "array", - "items": { - "type": "string" - } - }, - "apply_lora_to_mlp": { - "type": "boolean" - }, - "apply_lora_to_output": { - "type": "boolean" - }, - "rank": { - "type": "integer" - }, - "alpha": { - "type": "integer" - } - }, - "additionalProperties": false, - "required": [ - "lora_attn_modules", - "apply_lora_to_mlp", - "apply_lora_to_output", - "rank", - "alpha" ] }, - "PostTrainingSFTRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/PostTrainingSFTRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, - "SyntheticDataGenerationRequest": { + "SyntheticDataGenerateRequest": { "type": "object", "properties": { "dialogs": { @@ -5941,19 +5472,6 @@ "required": [ "dialogs", "filtering_function" - ], - "title": "Request to generate synthetic data. A small batch of prompts and a filtering function" - }, - "SyntheticDataGenerationRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/SyntheticDataGenerationRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" ] }, "SyntheticDataGenerationResponse": { @@ -6056,18 +5574,6 @@ "experiment_id" ] }, - "UpdateExperimentRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/UpdateExperimentRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "UpdateRunRequest": { "type": "object", "properties": { @@ -6112,18 +5618,6 @@ "run_id" ] }, - "UpdateRunRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/UpdateRunRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] - }, "UploadArtifactRequest": { "type": "object", "properties": { @@ -6173,18 +5667,6 @@ "artifact_type", "content" ] - }, - "UploadArtifactRequestWrapper": { - "type": "object", - "properties": { - "request": { - "$ref": "#/components/schemas/UploadArtifactRequest" - } - }, - "additionalProperties": false, - "required": [ - "request" - ] } }, "responses": {} @@ -6195,17 +5677,17 @@ } ], "tags": [ + { + "name": "Telemetry" + }, + { + "name": "Evaluations" + }, { "name": "PostTraining" }, { - "name": "RewardScoring" - }, - { - "name": "Safety" - }, - { - "name": "Memory" + "name": "Datasets" }, { "name": "SyntheticDataGeneration" @@ -6213,24 +5695,20 @@ { "name": "BatchInference" }, - { - "name": "AgenticSystem" - }, - { - "name": "Telemetry" - }, - { - "name": "Evaluations" - }, { "name": "Inference" }, { - "name": "Datasets" + "name": "AgenticSystem" }, { - "name": "BatchChatCompletionRequest", - "description": "" + "name": "Safety" + }, + { + "name": "Memory" + }, + { + "name": "RewardScoring" }, { "name": "BuiltinTool", @@ -6285,8 +5763,8 @@ "description": "" }, { - "name": "BatchChatCompletionRequestWrapper", - "description": "" + "name": "BatchChatCompletionRequest", + "description": "" }, { "name": "BatchChatCompletionResponse", @@ -6296,10 +5774,6 @@ "name": "BatchCompletionRequest", "description": "" }, - { - "name": "BatchCompletionRequestWrapper", - "description": "" - }, { "name": "BatchCompletionResponse", "description": "" @@ -6412,17 +5886,13 @@ "name": "AgenticSystemSessionCreateResponse", "description": "" }, - { - "name": "AgenticSystemTurnCreateRequest", - "description": "" - }, { "name": "Attachment", "description": "" }, { - "name": "AgenticSystemTurnCreateRequestWrapper", - "description": "" + "name": "CreateAgenticSystemTurnRequest", + "description": "" }, { "name": "AgenticSystemTurnResponseEvent", @@ -6480,10 +5950,6 @@ "name": "Turn", "description": "A single turn in an interaction with an Agentic System.\n\n" }, - { - "name": "CreateDatasetRequest", - "description": "Request to create a dataset.\n\n" - }, { "name": "TrainEvalDataset", "description": "Dataset to be used for training or evaluating language models.\n\n" @@ -6493,17 +5959,13 @@ "description": "" }, { - "name": "CreateDatasetRequestWrapper", - "description": "" + "name": "CreateDatasetRequest", + "description": "" }, { "name": "CreateExperimentRequest", "description": "" }, - { - "name": "CreateExperimentRequestWrapper", - "description": "" - }, { "name": "Experiment", "description": "" @@ -6524,10 +5986,6 @@ "name": "CreateRunRequest", "description": "" }, - { - "name": "CreateRunRequestWrapper", - "description": "" - }, { "name": "Run", "description": "" @@ -6560,17 +6018,9 @@ "name": "EmbeddingsResponse", "description": "" }, - { - "name": "Checkpoint", - "description": "Checkpoint created during training runs\n\n" - }, { "name": "EvaluateQuestionAnsweringRequest", - "description": "Request to evaluate question answering.\n\n" - }, - { - "name": "EvaluateQuestionAnsweringRequestWrapper", - "description": "" + "description": "" }, { "name": "EvaluationJob", @@ -6578,19 +6028,11 @@ }, { "name": "EvaluateSummarizationRequest", - "description": "Request to evaluate summarization.\n\n" - }, - { - "name": "EvaluateSummarizationRequestWrapper", - "description": "" + "description": "" }, { "name": "EvaluateTextGenerationRequest", - "description": "Request to evaluate text generation.\n\n" - }, - { - "name": "EvaluateTextGenerationRequestWrapper", - "description": "" + "description": "" }, { "name": "GetAgenticSystemSessionRequest", @@ -6633,12 +6075,8 @@ "description": "" }, { - "name": "LogSearchRequest", - "description": "" - }, - { - "name": "LogSearchRequestWrapper", - "description": "" + "name": "GetLogsRequest", + "description": "" }, { "name": "Log", @@ -6648,6 +6086,10 @@ "name": "Metric", "description": "" }, + { + "name": "Checkpoint", + "description": "Checkpoint created during training runs\n\n" + }, { "name": "PostTrainingJobArtifactsResponse", "description": "Artifacts of a finetuning job.\n\n" @@ -6680,18 +6122,10 @@ "name": "LogMessagesRequest", "description": "" }, - { - "name": "LogMessagesRequestWrapper", - "description": "" - }, { "name": "LogMetricsRequest", "description": "" }, - { - "name": "LogMetricsRequestWrapper", - "description": "" - }, { "name": "DPOAlignmentConfig", "description": "" @@ -6700,10 +6134,6 @@ "name": "OptimizerConfig", "description": "" }, - { - "name": "PostTrainingRLHFRequest", - "description": "Request to finetune a model.\n\n" - }, { "name": "RLHFAlgorithm", "description": "" @@ -6713,8 +6143,8 @@ "description": "" }, { - "name": "PostTrainingRLHFRequestWrapper", - "description": "" + "name": "PreferenceOptimizeRequest", + "description": "" }, { "name": "QueryDocumentsRequest", @@ -6729,12 +6159,8 @@ "description": "" }, { - "name": "RewardScoringRequest", - "description": "Request to score a reward function. A list of prompts and a list of responses per prompt.\n\n" - }, - { - "name": "RewardScoringRequestWrapper", - "description": "" + "name": "RewardScoreRequest", + "description": "" }, { "name": "RewardScoringResponse", @@ -6749,12 +6175,8 @@ "description": "" }, { - "name": "RunShieldRequest", - "description": "" - }, - { - "name": "RunShieldRequestWrapper", - "description": "" + "name": "RunShieldsRequest", + "description": "" }, { "name": "RunShieldResponse", @@ -6772,25 +6194,17 @@ "name": "LoraFinetuningConfig", "description": "" }, - { - "name": "PostTrainingSFTRequest", - "description": "Request to finetune a model.\n\n" - }, { "name": "QLoraFinetuningConfig", "description": "" }, { - "name": "PostTrainingSFTRequestWrapper", - "description": "" + "name": "SupervisedFineTuneRequest", + "description": "" }, { - "name": "SyntheticDataGenerationRequest", - "description": "Request to generate synthetic data. A small batch of prompts and a filtering function\n\n" - }, - { - "name": "SyntheticDataGenerationRequestWrapper", - "description": "" + "name": "SyntheticDataGenerateRequest", + "description": "" }, { "name": "SyntheticDataGenerationResponse", @@ -6804,25 +6218,13 @@ "name": "UpdateExperimentRequest", "description": "" }, - { - "name": "UpdateExperimentRequestWrapper", - "description": "" - }, { "name": "UpdateRunRequest", "description": "" }, - { - "name": "UpdateRunRequestWrapper", - "description": "" - }, { "name": "UploadArtifactRequest", "description": "" - }, - { - "name": "UploadArtifactRequestWrapper", - "description": "" } ], "x-tagGroups": [ @@ -6849,8 +6251,6 @@ "AgenticSystemCreateResponse", "AgenticSystemSessionCreateResponse", "AgenticSystemStepResponse", - "AgenticSystemTurnCreateRequest", - "AgenticSystemTurnCreateRequestWrapper", "AgenticSystemTurnResponseEvent", "AgenticSystemTurnResponseStepCompletePayload", "AgenticSystemTurnResponseStepProgressPayload", @@ -6862,10 +6262,8 @@ "ArtifactType", "Attachment", "BatchChatCompletionRequest", - "BatchChatCompletionRequestWrapper", "BatchChatCompletionResponse", "BatchCompletionRequest", - "BatchCompletionRequestWrapper", "BatchCompletionResponse", "BuiltinShield", "BuiltinTool", @@ -6882,13 +6280,11 @@ "CompletionResponseStreamChunk", "CreateAgenticSystemRequest", "CreateAgenticSystemSessionRequest", + "CreateAgenticSystemTurnRequest", "CreateDatasetRequest", - "CreateDatasetRequestWrapper", "CreateExperimentRequest", - "CreateExperimentRequestWrapper", "CreateMemoryBankRequest", "CreateRunRequest", - "CreateRunRequestWrapper", "DPOAlignmentConfig", "DeleteAgenticSystemRequest", "DeleteAgenticSystemSessionRequest", @@ -6900,11 +6296,8 @@ "EmbeddingsRequest", "EmbeddingsResponse", "EvaluateQuestionAnsweringRequest", - "EvaluateQuestionAnsweringRequestWrapper", "EvaluateSummarizationRequest", - "EvaluateSummarizationRequestWrapper", "EvaluateTextGenerationRequest", - "EvaluateTextGenerationRequestWrapper", "EvaluationJob", "EvaluationJobArtifactsResponse", "EvaluationJobLogStream", @@ -6915,16 +6308,13 @@ "FunctionCallToolDefinition", "GetAgenticSystemSessionRequest", "GetDocumentsRequest", + "GetLogsRequest", "InferenceStep", "InsertDocumentsRequest", "ListArtifactsRequest", "Log", "LogMessagesRequest", - "LogMessagesRequestWrapper", "LogMetricsRequest", - "LogMetricsRequestWrapper", - "LogSearchRequest", - "LogSearchRequestWrapper", "LoraFinetuningConfig", "MemoryBank", "MemoryBankDocument", @@ -6938,23 +6328,18 @@ "PostTrainingJobLogStream", "PostTrainingJobStatus", "PostTrainingJobStatusResponse", - "PostTrainingRLHFRequest", - "PostTrainingRLHFRequestWrapper", - "PostTrainingSFTRequest", - "PostTrainingSFTRequestWrapper", + "PreferenceOptimizeRequest", "QLoraFinetuningConfig", "QueryDocumentsRequest", "QueryDocumentsResponse", "RLHFAlgorithm", "RestAPIExecutionConfig", "RestAPIMethod", - "RewardScoringRequest", - "RewardScoringRequestWrapper", + "RewardScoreRequest", "RewardScoringResponse", "Run", - "RunShieldRequest", - "RunShieldRequestWrapper", "RunShieldResponse", + "RunShieldsRequest", "SamplingParams", "SamplingStrategy", "ScoredDialogGenerations", @@ -6965,8 +6350,8 @@ "ShieldDefinition", "ShieldResponse", "StopReason", - "SyntheticDataGenerationRequest", - "SyntheticDataGenerationRequestWrapper", + "SupervisedFineTuneRequest", + "SyntheticDataGenerateRequest", "SyntheticDataGenerationResponse", "SystemMessage", "TokenLogProbs", @@ -6987,11 +6372,8 @@ "URL", "UpdateDocumentsRequest", "UpdateExperimentRequest", - "UpdateExperimentRequestWrapper", "UpdateRunRequest", - "UpdateRunRequestWrapper", "UploadArtifactRequest", - "UploadArtifactRequestWrapper", "UserMessage", "WolframAlphaToolDefinition" ] diff --git a/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.yaml b/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.yaml index 877639f51..b029f9adf 100644 --- a/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.yaml +++ b/rfcs/RFC-0001-llama-stack-assets/llama-stack-spec.yaml @@ -180,180 +180,6 @@ components: required: - step type: object - AgenticSystemTurnCreateRequest: - additionalProperties: false - properties: - agent_id: - type: string - attachments: - items: - $ref: '#/components/schemas/Attachment' - type: array - input_shields: - items: - $ref: '#/components/schemas/ShieldDefinition' - type: array - instructions: - type: string - messages: - items: - oneOf: - - $ref: '#/components/schemas/UserMessage' - - $ref: '#/components/schemas/ToolResponseMessage' - type: array - output_shields: - items: - $ref: '#/components/schemas/ShieldDefinition' - type: array - sampling_params: - $ref: '#/components/schemas/SamplingParams' - session_id: - type: string - stream: - type: boolean - tool_choice: - $ref: '#/components/schemas/ToolChoice' - tool_prompt_format: - $ref: '#/components/schemas/ToolPromptFormat' - tools: - items: - oneOf: - - $ref: '#/components/schemas/SearchToolDefinition' - - $ref: '#/components/schemas/WolframAlphaToolDefinition' - - $ref: '#/components/schemas/PhotogenToolDefinition' - - $ref: '#/components/schemas/CodeInterpreterToolDefinition' - - $ref: '#/components/schemas/FunctionCallToolDefinition' - - additionalProperties: false - properties: - input_shields: - items: - $ref: '#/components/schemas/ShieldDefinition' - type: array - max_chunks: - type: integer - max_tokens_in_context: - type: integer - memory_bank_configs: - items: - oneOf: - - additionalProperties: false - properties: - bank_id: - type: string - type: - const: vector - type: string - required: - - bank_id - - type - type: object - - additionalProperties: false - properties: - bank_id: - type: string - keys: - items: - type: string - type: array - type: - const: keyvalue - type: string - required: - - bank_id - - type - - keys - type: object - - additionalProperties: false - properties: - bank_id: - type: string - type: - const: keyword - type: string - required: - - bank_id - - type - type: object - - additionalProperties: false - properties: - bank_id: - type: string - entities: - items: - type: string - type: array - type: - const: graph - type: string - required: - - bank_id - - type - - entities - type: object - type: array - output_shields: - items: - $ref: '#/components/schemas/ShieldDefinition' - type: array - query_generator_config: - oneOf: - - additionalProperties: false - properties: - sep: - type: string - type: - const: default - type: string - required: - - type - - sep - type: object - - additionalProperties: false - properties: - model: - type: string - template: - type: string - type: - const: llm - type: string - required: - - type - - model - - template - type: object - - additionalProperties: false - properties: - type: - const: custom - type: string - required: - - type - type: object - type: - const: memory - type: string - required: - - type - - memory_bank_configs - - query_generator_config - - max_tokens_in_context - - max_chunks - type: object - type: array - required: - - agent_id - - session_id - - messages - type: object - AgenticSystemTurnCreateRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/AgenticSystemTurnCreateRequest' - required: - - request - type: object AgenticSystemTurnResponseEvent: additionalProperties: false properties: @@ -574,14 +400,6 @@ components: - model - messages_batch type: object - BatchChatCompletionRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/BatchChatCompletionRequest' - required: - - request - type: object BatchChatCompletionResponse: additionalProperties: false properties: @@ -617,14 +435,6 @@ components: - model - content_batch type: object - BatchCompletionRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/BatchCompletionRequest' - required: - - request - type: object BatchCompletionResponse: additionalProperties: false properties: @@ -843,6 +653,30 @@ components: - agent_id - session_name type: object + CreateAgenticSystemTurnRequest: + additionalProperties: false + properties: + agent_id: + type: string + attachments: + items: + $ref: '#/components/schemas/Attachment' + type: array + messages: + items: + oneOf: + - $ref: '#/components/schemas/UserMessage' + - $ref: '#/components/schemas/ToolResponseMessage' + type: array + session_id: + type: string + stream: + type: boolean + required: + - agent_id + - session_id + - messages + type: object CreateDatasetRequest: additionalProperties: false properties: @@ -853,15 +687,6 @@ components: required: - uuid - dataset - title: Request to create a dataset. - type: object - CreateDatasetRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/CreateDatasetRequest' - required: - - request type: object CreateExperimentRequest: additionalProperties: false @@ -881,14 +706,6 @@ components: required: - name type: object - CreateExperimentRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/CreateExperimentRequest' - required: - - request - type: object CreateMemoryBankRequest: additionalProperties: false properties: @@ -960,14 +777,6 @@ components: required: - experiment_id type: object - CreateRunRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/CreateRunRequest' - required: - - request - type: object DPOAlignmentConfig: additionalProperties: false properties: @@ -1110,12 +919,6 @@ components: EvaluateQuestionAnsweringRequest: additionalProperties: false properties: - checkpoint: - $ref: '#/components/schemas/Checkpoint' - dataset: - $ref: '#/components/schemas/TrainEvalDataset' - job_uuid: - type: string metrics: items: enum: @@ -1123,33 +926,12 @@ components: - f1 type: string type: array - sampling_params: - $ref: '#/components/schemas/SamplingParams' required: - - job_uuid - - dataset - - checkpoint - - sampling_params - metrics - title: Request to evaluate question answering. - type: object - EvaluateQuestionAnsweringRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/EvaluateQuestionAnsweringRequest' - required: - - request type: object EvaluateSummarizationRequest: additionalProperties: false properties: - checkpoint: - $ref: '#/components/schemas/Checkpoint' - dataset: - $ref: '#/components/schemas/TrainEvalDataset' - job_uuid: - type: string metrics: items: enum: @@ -1157,33 +939,12 @@ components: - bleu type: string type: array - sampling_params: - $ref: '#/components/schemas/SamplingParams' required: - - job_uuid - - dataset - - checkpoint - - sampling_params - metrics - title: Request to evaluate summarization. - type: object - EvaluateSummarizationRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/EvaluateSummarizationRequest' - required: - - request type: object EvaluateTextGenerationRequest: additionalProperties: false properties: - checkpoint: - $ref: '#/components/schemas/Checkpoint' - dataset: - $ref: '#/components/schemas/TrainEvalDataset' - job_uuid: - type: string metrics: items: enum: @@ -1192,23 +953,8 @@ components: - bleu type: string type: array - sampling_params: - $ref: '#/components/schemas/SamplingParams' required: - - job_uuid - - dataset - - checkpoint - - sampling_params - metrics - title: Request to evaluate text generation. - type: object - EvaluateTextGenerationRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/EvaluateTextGenerationRequest' - required: - - request type: object EvaluationJob: additionalProperties: false @@ -1338,6 +1084,20 @@ components: required: - document_ids type: object + GetLogsRequest: + additionalProperties: false + properties: + filters: + additionalProperties: + oneOf: + - type: 'null' + - type: boolean + - type: number + - type: string + - type: array + - type: object + type: object + type: object InferenceStep: additionalProperties: false properties: @@ -1423,14 +1183,6 @@ components: required: - logs type: object - LogMessagesRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/LogMessagesRequest' - required: - - request - type: object LogMetricsRequest: additionalProperties: false properties: @@ -1444,40 +1196,6 @@ components: - run_id - metrics type: object - LogMetricsRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/LogMetricsRequest' - required: - - request - type: object - LogSearchRequest: - additionalProperties: false - properties: - filters: - additionalProperties: - oneOf: - - type: 'null' - - type: boolean - - type: number - - type: string - - type: array - - type: object - type: object - query: - type: string - required: - - query - type: object - LogSearchRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/LogSearchRequest' - required: - - request - type: object LoraFinetuningConfig: additionalProperties: false properties: @@ -1766,7 +1484,7 @@ components: - checkpoints title: Status of a finetuning job. type: object - PostTrainingRLHFRequest: + PreferenceOptimizeRequest: additionalProperties: false properties: algorithm: @@ -1816,78 +1534,6 @@ components: - training_config - hyperparam_search_config - logger_config - title: Request to finetune a model. - type: object - PostTrainingRLHFRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/PostTrainingRLHFRequest' - required: - - request - type: object - PostTrainingSFTRequest: - additionalProperties: false - properties: - algorithm: - $ref: '#/components/schemas/FinetuningAlgorithm' - algorithm_config: - oneOf: - - $ref: '#/components/schemas/LoraFinetuningConfig' - - $ref: '#/components/schemas/QLoraFinetuningConfig' - - $ref: '#/components/schemas/DoraFinetuningConfig' - dataset: - $ref: '#/components/schemas/TrainEvalDataset' - hyperparam_search_config: - additionalProperties: - oneOf: - - type: 'null' - - type: boolean - - type: number - - type: string - - type: array - - type: object - type: object - job_uuid: - type: string - logger_config: - additionalProperties: - oneOf: - - type: 'null' - - type: boolean - - type: number - - type: string - - type: array - - type: object - type: object - model: - type: string - optimizer_config: - $ref: '#/components/schemas/OptimizerConfig' - training_config: - $ref: '#/components/schemas/TrainingConfig' - validation_dataset: - $ref: '#/components/schemas/TrainEvalDataset' - required: - - job_uuid - - model - - dataset - - validation_dataset - - algorithm - - algorithm_config - - optimizer_config - - training_config - - hyperparam_search_config - - logger_config - title: Request to finetune a model. - type: object - PostTrainingSFTRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/PostTrainingSFTRequest' - required: - - request type: object QLoraFinetuningConfig: additionalProperties: false @@ -2019,7 +1665,7 @@ components: - PUT - DELETE type: string - RewardScoringRequest: + RewardScoreRequest: additionalProperties: false properties: dialog_generations: @@ -2031,16 +1677,6 @@ components: required: - dialog_generations - model - title: Request to score a reward function. A list of prompts and a list of responses - per prompt. - type: object - RewardScoringRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/RewardScoringRequest' - required: - - request type: object RewardScoringResponse: additionalProperties: false @@ -2086,7 +1722,17 @@ components: - started_at - metadata type: object - RunShieldRequest: + RunShieldResponse: + additionalProperties: false + properties: + responses: + items: + $ref: '#/components/schemas/ShieldResponse' + type: array + required: + - responses + type: object + RunShieldsRequest: additionalProperties: false properties: messages: @@ -2105,24 +1751,6 @@ components: - messages - shields type: object - RunShieldRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/RunShieldRequest' - required: - - request - type: object - RunShieldResponse: - additionalProperties: false - properties: - responses: - items: - $ref: '#/components/schemas/ShieldResponse' - type: array - required: - - responses - type: object SamplingParams: additionalProperties: false properties: @@ -2297,7 +1925,61 @@ components: - end_of_message - out_of_tokens type: string - SyntheticDataGenerationRequest: + SupervisedFineTuneRequest: + additionalProperties: false + properties: + algorithm: + $ref: '#/components/schemas/FinetuningAlgorithm' + algorithm_config: + oneOf: + - $ref: '#/components/schemas/LoraFinetuningConfig' + - $ref: '#/components/schemas/QLoraFinetuningConfig' + - $ref: '#/components/schemas/DoraFinetuningConfig' + dataset: + $ref: '#/components/schemas/TrainEvalDataset' + hyperparam_search_config: + additionalProperties: + oneOf: + - type: 'null' + - type: boolean + - type: number + - type: string + - type: array + - type: object + type: object + job_uuid: + type: string + logger_config: + additionalProperties: + oneOf: + - type: 'null' + - type: boolean + - type: number + - type: string + - type: array + - type: object + type: object + model: + type: string + optimizer_config: + $ref: '#/components/schemas/OptimizerConfig' + training_config: + $ref: '#/components/schemas/TrainingConfig' + validation_dataset: + $ref: '#/components/schemas/TrainEvalDataset' + required: + - job_uuid + - model + - dataset + - validation_dataset + - algorithm + - algorithm_config + - optimizer_config + - training_config + - hyperparam_search_config + - logger_config + type: object + SyntheticDataGenerateRequest: additionalProperties: false properties: dialogs: @@ -2323,16 +2005,6 @@ components: required: - dialogs - filtering_function - title: Request to generate synthetic data. A small batch of prompts and a filtering - function - type: object - SyntheticDataGenerationRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/SyntheticDataGenerationRequest' - required: - - request type: object SyntheticDataGenerationResponse: additionalProperties: false @@ -2702,14 +2374,6 @@ components: required: - experiment_id type: object - UpdateExperimentRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/UpdateExperimentRequest' - required: - - request - type: object UpdateRunRequest: additionalProperties: false properties: @@ -2733,14 +2397,6 @@ components: required: - run_id type: object - UpdateRunRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/UpdateRunRequest' - required: - - request - type: object UploadArtifactRequest: additionalProperties: false properties: @@ -2769,14 +2425,6 @@ components: - artifact_type - content type: object - UploadArtifactRequestWrapper: - additionalProperties: false - properties: - request: - $ref: '#/components/schemas/UploadArtifactRequest' - required: - - request - type: object UserMessage: additionalProperties: false properties: @@ -2822,7 +2470,7 @@ info: description: "This is the specification of the llama stack that provides\n \ \ a set of endpoints and their corresponding interfaces that are tailored\ \ to\n best leverage Llama Models. The specification is still in\ - \ draft and subject to change.\n Generated at 2024-09-11 12:34:17.939867" + \ draft and subject to change.\n Generated at 2024-09-11 14:17:44.361462" title: '[DRAFT] Llama Stack Specification' version: 0.0.1 jsonSchemaDialect: https://json-schema.org/draft/2020-12/schema @@ -2954,7 +2602,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/AgenticSystemTurnCreateRequestWrapper' + $ref: '#/components/schemas/CreateAgenticSystemTurnRequest' required: true responses: '200': @@ -3011,7 +2659,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/BatchChatCompletionRequestWrapper' + $ref: '#/components/schemas/BatchChatCompletionRequest' required: true responses: '200': @@ -3029,7 +2677,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/BatchCompletionRequestWrapper' + $ref: '#/components/schemas/BatchCompletionRequest' required: true responses: '200': @@ -3047,7 +2695,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/CreateDatasetRequestWrapper' + $ref: '#/components/schemas/CreateDatasetRequest' required: true responses: '200': @@ -3169,7 +2817,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/EvaluateQuestionAnsweringRequestWrapper' + $ref: '#/components/schemas/EvaluateQuestionAnsweringRequest' required: true responses: '200': @@ -3187,7 +2835,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/EvaluateSummarizationRequestWrapper' + $ref: '#/components/schemas/EvaluateSummarizationRequest' required: true responses: '200': @@ -3205,7 +2853,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/EvaluateTextGenerationRequestWrapper' + $ref: '#/components/schemas/EvaluateTextGenerationRequest' required: true responses: '200': @@ -3241,7 +2889,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/UploadArtifactRequestWrapper' + $ref: '#/components/schemas/UploadArtifactRequest' required: true responses: '200': @@ -3259,7 +2907,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/CreateExperimentRequestWrapper' + $ref: '#/components/schemas/CreateExperimentRequest' required: true responses: '200': @@ -3277,7 +2925,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/CreateRunRequestWrapper' + $ref: '#/components/schemas/CreateRunRequest' required: true responses: '200': @@ -3324,7 +2972,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/UpdateExperimentRequestWrapper' + $ref: '#/components/schemas/UpdateExperimentRequest' required: true responses: '200': @@ -3391,12 +3039,17 @@ paths: - Inference /logging/get_logs: post: - parameters: [] + parameters: + - in: query + name: query + required: true + schema: + type: string requestBody: content: application/json: schema: - $ref: '#/components/schemas/LogSearchRequestWrapper' + $ref: '#/components/schemas/GetLogsRequest' required: true responses: '200': @@ -3414,7 +3067,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/LogMessagesRequestWrapper' + $ref: '#/components/schemas/LogMessagesRequest' required: true responses: '200': @@ -3655,7 +3308,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/PostTrainingRLHFRequestWrapper' + $ref: '#/components/schemas/PreferenceOptimizeRequest' required: true responses: '200': @@ -3673,7 +3326,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/PostTrainingSFTRequestWrapper' + $ref: '#/components/schemas/SupervisedFineTuneRequest' required: true responses: '200': @@ -3691,7 +3344,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/RewardScoringRequestWrapper' + $ref: '#/components/schemas/RewardScoreRequest' required: true responses: '200': @@ -3709,7 +3362,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/LogMetricsRequestWrapper' + $ref: '#/components/schemas/LogMetricsRequest' required: true responses: '200': @@ -3740,7 +3393,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/UpdateRunRequestWrapper' + $ref: '#/components/schemas/UpdateRunRequest' required: true responses: '200': @@ -3758,7 +3411,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/RunShieldRequestWrapper' + $ref: '#/components/schemas/RunShieldsRequest' required: true responses: '200': @@ -3776,7 +3429,7 @@ paths: content: application/json: schema: - $ref: '#/components/schemas/SyntheticDataGenerationRequestWrapper' + $ref: '#/components/schemas/SyntheticDataGenerateRequest' required: true responses: '200': @@ -3792,20 +3445,17 @@ security: servers: - url: http://any-hosted-llama-stack.com tags: -- name: PostTraining -- name: RewardScoring -- name: Safety -- name: Memory -- name: SyntheticDataGeneration -- name: BatchInference -- name: AgenticSystem - name: Telemetry - name: Evaluations -- name: Inference +- name: PostTraining - name: Datasets -- description: - name: BatchChatCompletionRequest +- name: SyntheticDataGeneration +- name: BatchInference +- name: Inference +- name: AgenticSystem +- name: Safety +- name: Memory +- name: RewardScoring - description: name: BuiltinTool - description: name: UserMessage -- description: - name: BatchChatCompletionRequestWrapper + name: BatchChatCompletionRequest - description: name: BatchChatCompletionResponse - description: name: BatchCompletionRequest -- description: - name: BatchCompletionRequestWrapper - description: name: BatchCompletionResponse @@ -3944,14 +3591,11 @@ tags: - description: name: AgenticSystemSessionCreateResponse -- description: - name: AgenticSystemTurnCreateRequest - description: name: Attachment -- description: - name: AgenticSystemTurnCreateRequestWrapper + name: CreateAgenticSystemTurnRequest - description: 'Streamed agent execution response. @@ -3995,11 +3639,6 @@ tags: ' name: Turn -- description: 'Request to create a dataset. - - - ' - name: CreateDatasetRequest - description: 'Dataset to be used for training or evaluating language models. @@ -4008,15 +3647,12 @@ tags: - description: name: TrainEvalDatasetColumnType -- description: - name: CreateDatasetRequestWrapper + name: CreateDatasetRequest - description: name: CreateExperimentRequest -- description: - name: CreateExperimentRequestWrapper - description: name: Experiment - description: name: CreateRunRequest -- description: - name: CreateRunRequestWrapper - description: name: Run - description: name: EmbeddingsResponse -- description: 'Checkpoint created during training runs - - - ' - name: Checkpoint -- description: 'Request to evaluate question answering. - - - ' - name: EvaluateQuestionAnsweringRequest -- description: - name: EvaluateQuestionAnsweringRequestWrapper + name: EvaluateQuestionAnsweringRequest - description: name: EvaluationJob -- description: 'Request to evaluate summarization. - - - ' +- description: name: EvaluateSummarizationRequest -- description: - name: EvaluateSummarizationRequestWrapper -- description: 'Request to evaluate text generation. - - - ' name: EvaluateTextGenerationRequest -- description: - name: EvaluateTextGenerationRequestWrapper - description: name: GetAgenticSystemSessionRequest @@ -4123,16 +3733,17 @@ tags: - description: name: EvaluationJobStatusResponse -- description: - name: LogSearchRequest -- description: - name: LogSearchRequestWrapper +- description: + name: GetLogsRequest - description: name: Log - description: name: Metric +- description: 'Checkpoint created during training runs + + + ' + name: Checkpoint - description: 'Artifacts of a finetuning job. @@ -4165,33 +3776,22 @@ tags: - description: name: LogMessagesRequest -- description: - name: LogMessagesRequestWrapper - description: name: LogMetricsRequest -- description: - name: LogMetricsRequestWrapper - description: name: DPOAlignmentConfig - description: name: OptimizerConfig -- description: 'Request to finetune a model. - - - ' - name: PostTrainingRLHFRequest - description: name: RLHFAlgorithm - description: name: TrainingConfig -- description: - name: PostTrainingRLHFRequestWrapper + name: PreferenceOptimizeRequest - description: name: QueryDocumentsRequest @@ -4201,15 +3801,9 @@ tags: - description: name: DialogGenerations -- description: 'Request to score a reward function. A list of prompts and a list of - responses per prompt. - - - ' - name: RewardScoringRequest -- description: - name: RewardScoringRequestWrapper + name: RewardScoreRequest - description: 'Response from the reward scoring. Batch of (prompt, response, score) tuples that pass the threshold. @@ -4221,12 +3815,9 @@ tags: name: ScoredDialogGenerations - description: name: ScoredMessage -- description: - name: RunShieldRequest -- description: - name: RunShieldRequestWrapper + name: RunShieldsRequest - description: name: RunShieldResponse @@ -4239,27 +3830,15 @@ tags: - description: name: LoraFinetuningConfig -- description: 'Request to finetune a model. - - - ' - name: PostTrainingSFTRequest - description: name: QLoraFinetuningConfig -- description: - name: PostTrainingSFTRequestWrapper -- description: 'Request to generate synthetic data. A small batch of prompts and a - filtering function - - - ' - name: SyntheticDataGenerationRequest -- description: - name: SyntheticDataGenerationRequestWrapper + name: SyntheticDataGenerateRequest - description: 'Response from the synthetic data generation. Batch of (prompt, response, score) tuples that pass the threshold. @@ -4273,21 +3852,12 @@ tags: - description: name: UpdateExperimentRequest -- description: - name: UpdateExperimentRequestWrapper - description: name: UpdateRunRequest -- description: - name: UpdateRunRequestWrapper - description: name: UploadArtifactRequest -- description: - name: UploadArtifactRequestWrapper x-tagGroups: - name: Operations tags: @@ -4308,8 +3878,6 @@ x-tagGroups: - AgenticSystemCreateResponse - AgenticSystemSessionCreateResponse - AgenticSystemStepResponse - - AgenticSystemTurnCreateRequest - - AgenticSystemTurnCreateRequestWrapper - AgenticSystemTurnResponseEvent - AgenticSystemTurnResponseStepCompletePayload - AgenticSystemTurnResponseStepProgressPayload @@ -4321,10 +3889,8 @@ x-tagGroups: - ArtifactType - Attachment - BatchChatCompletionRequest - - BatchChatCompletionRequestWrapper - BatchChatCompletionResponse - BatchCompletionRequest - - BatchCompletionRequestWrapper - BatchCompletionResponse - BuiltinShield - BuiltinTool @@ -4341,13 +3907,11 @@ x-tagGroups: - CompletionResponseStreamChunk - CreateAgenticSystemRequest - CreateAgenticSystemSessionRequest + - CreateAgenticSystemTurnRequest - CreateDatasetRequest - - CreateDatasetRequestWrapper - CreateExperimentRequest - - CreateExperimentRequestWrapper - CreateMemoryBankRequest - CreateRunRequest - - CreateRunRequestWrapper - DPOAlignmentConfig - DeleteAgenticSystemRequest - DeleteAgenticSystemSessionRequest @@ -4359,11 +3923,8 @@ x-tagGroups: - EmbeddingsRequest - EmbeddingsResponse - EvaluateQuestionAnsweringRequest - - EvaluateQuestionAnsweringRequestWrapper - EvaluateSummarizationRequest - - EvaluateSummarizationRequestWrapper - EvaluateTextGenerationRequest - - EvaluateTextGenerationRequestWrapper - EvaluationJob - EvaluationJobArtifactsResponse - EvaluationJobLogStream @@ -4374,16 +3935,13 @@ x-tagGroups: - FunctionCallToolDefinition - GetAgenticSystemSessionRequest - GetDocumentsRequest + - GetLogsRequest - InferenceStep - InsertDocumentsRequest - ListArtifactsRequest - Log - LogMessagesRequest - - LogMessagesRequestWrapper - LogMetricsRequest - - LogMetricsRequestWrapper - - LogSearchRequest - - LogSearchRequestWrapper - LoraFinetuningConfig - MemoryBank - MemoryBankDocument @@ -4397,23 +3955,18 @@ x-tagGroups: - PostTrainingJobLogStream - PostTrainingJobStatus - PostTrainingJobStatusResponse - - PostTrainingRLHFRequest - - PostTrainingRLHFRequestWrapper - - PostTrainingSFTRequest - - PostTrainingSFTRequestWrapper + - PreferenceOptimizeRequest - QLoraFinetuningConfig - QueryDocumentsRequest - QueryDocumentsResponse - RLHFAlgorithm - RestAPIExecutionConfig - RestAPIMethod - - RewardScoringRequest - - RewardScoringRequestWrapper + - RewardScoreRequest - RewardScoringResponse - Run - - RunShieldRequest - - RunShieldRequestWrapper - RunShieldResponse + - RunShieldsRequest - SamplingParams - SamplingStrategy - ScoredDialogGenerations @@ -4424,8 +3977,8 @@ x-tagGroups: - ShieldDefinition - ShieldResponse - StopReason - - SyntheticDataGenerationRequest - - SyntheticDataGenerationRequestWrapper + - SupervisedFineTuneRequest + - SyntheticDataGenerateRequest - SyntheticDataGenerationResponse - SystemMessage - TokenLogProbs @@ -4446,10 +3999,7 @@ x-tagGroups: - URL - UpdateDocumentsRequest - UpdateExperimentRequest - - UpdateExperimentRequestWrapper - UpdateRunRequest - - UpdateRunRequestWrapper - UploadArtifactRequest - - UploadArtifactRequestWrapper - UserMessage - WolframAlphaToolDefinition