Comment
rl/value_loss
rl/value_loss
group: seq_parallel_metrics_loss_42_no_seq_packing_2
group: seq_parallel_metrics_loss_42_2
group: 82d0dc2_loss_2
rl/policy_loss
rl/policy_loss
group: seq_parallel_metrics_loss_42_no_seq_packing_2
group: seq_parallel_metrics_loss_42_2
group: 82d0dc2_loss_2
rl/loss
rl/loss
group: seq_parallel_metrics_loss_42_no_seq_packing_2
group: seq_parallel_metrics_loss_42_2
group: 82d0dc2_loss_2
Run set
5
Name
5 visualized
State
Notes
User
Tags
Created
Runtime
Sweep
_cpu
_mixed_precision
actor.chunk_size
actor.discount_factor
actor.llm_max_rollouts
actor.log_each_n_secs
actor.max_reasoning_steps
actor.problem_queue_size
actor.result_queue_size
actor.rollout_policy
actor.rollout_workers
actor.sampling.method
actor.shared_memory_entry_size
actor.submit_delay
actor.system_prompt
actor.task_template
actor.threads_per_llm
actor.throughput_window_size
agent._target_
agent.llms.default.parameters.max_tokens
agent.llms.default.parameters.temperature
agent.max_iterations
agent.max_prompt_length
agent.name
agent.nodes
agent.store_llm_calls
agent.system_prompt
agent.templates.allowed_steps
agent.templates.allowed_tools
agent.templates.format
agent.templates.nodes
agent.templates.system_prompt
agent.templates.thought_format
agent_max_loops
also_save_steps
attempts
attn_implementation
auto_device_map
backend
config_name
cuda_empty_cache
dataset_loader
dataset_loader_params.dataset_path
dataset_loader_params.seeds
Killed
-
apiche
1h 15m
-
-
-
-
1
64
0
-
64
64
pipelinerl.domains.mcp.generate_mcp_rollout
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
tapeagents.agent.Agent
-
-
3
-
mcp_agent
["{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'act', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Then produce single function call for the next step. If the answer is ready, call MathAnswer. Put your final answer within \\\\boxed{}.', 'steps': ['pipelinerl.domains.mcp.steps.MathAnswer'], 'use_known_actions': True, 'use_function_calls': True}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.\\nBe specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.\\nConsider that next steps may depend on results of previous steps, so include conditional branching using \"if\" statements where needed.\\nStart with the title \"Plan\". Every step should have short name and description.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'reflect', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': \"1. Evaluate the action's success, explain its effect on current step, overall plan and task solution.\\n2. If the last action was not successful, describe errors and the possible reasons for failure.\\n3. Check if the current plan step is finished. \\n4. If the step is finished, update the following steps of the plan with new information and choose the next step.\\n${agent.templates.thought_format}\\n\", 'next_node': 'select'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'select', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Select the next step to do to move forward with the plan. Describe the expected effect of the proposed action.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'summarize', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Summarize last observation. If its an image, thoroughly describe it with all details.\\nDescribe the results of the last action and observed changes\\nDo not hallucinate or make up any information, only describe what you see in the observation.\\nDo not guess or assume action effects, describe only visible changes.\\n${agent.templates.thought_format}\\n'}"]
true
-
You have access to the following tools:
{tools_description}
You have access to the following tools:
{tools_description}
Output only a single JSON dict. Do not repeat the last thought again. If the last action does not change the observation, do not repeat it! DO NOT OUTPUT ANYTHING BESIDES THE JSON! DO NOT PLACE ANY COMMENTS INSIDE THE JSON. It will break the system that processes the output.
-
You are an expert AI Agent trained to assist users with complex information processing tasks.
Your role is to understand user queries and respond in a helpful and accurate manner.
Keep your replies concise and direct. Prioritize clarity and avoid over-elaboration.
Do not express emotions or opinions about user questions.
Important! Respond with the plain text, do not include any JSON or code.
Do not output anything besides what I asked in this message.
3
-
1
-
-
-
-
-
pipelinerl.domains.math.load_datasets
-
-
Crashed
-
apiche
4d 4h 38m 1s
-
-
-
-
1
64
0
-
64
64
pipelinerl.domains.mcp.generate_mcp_rollout
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
tapeagents.agent.Agent
-
-
3
-
mcp_agent
["{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'act', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Then produce single function call for the next step. YOU MUST CALL PYTHON. If the answer is ready, call MathAnswer. Put your final answer within \\\\boxed{}.', 'steps': ['pipelinerl.domains.mcp.steps.MathAnswer'], 'use_known_actions': True, 'use_function_calls': True}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.\\nBe specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.\\nConsider that next steps may depend on results of previous steps, so include conditional branching using \"if\" statements where needed.\\nStart with the title \"Plan\". Every step should have short name and description.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'reflect', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': \"1. Evaluate the action's success, explain its effect on current step, overall plan and task solution.\\n2. If the last action was not successful, describe errors and the possible reasons for failure.\\n3. Check if the current plan step is finished. \\n4. If the step is finished, update the following steps of the plan with new information and choose the next step.\\n${agent.templates.thought_format}\\n\", 'next_node': 'select'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'select', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Select the next step to do to move forward with the plan. Describe the expected effect of the proposed action.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'summarize', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Summarize last observation. If its an image, thoroughly describe it with all details.\\nDescribe the results of the last action and observed changes\\nDo not hallucinate or make up any information, only describe what you see in the observation.\\nDo not guess or assume action effects, describe only visible changes.\\n${agent.templates.thought_format}\\n'}"]
true
-
You have access to the following tools:
{tools_description}
You have access to the following tools:
{tools_description}
Output only a single JSON dict. Do not repeat the last thought again. If the last action does not change the observation, do not repeat it! DO NOT OUTPUT ANYTHING BESIDES THE JSON! DO NOT PLACE ANY COMMENTS INSIDE THE JSON. It will break the system that processes the output.
-
You are an expert AI Agent trained to assist users with complex information processing tasks.
Your role is to understand user queries and respond in a helpful and accurate manner.
Keep your replies concise and direct. Prioritize clarity and avoid over-elaboration.
Do not express emotions or opinions about user questions.
Important! Respond with the plain text, do not include any JSON or code.
Do not output anything besides what I asked in this message.
3
-
1
-
-
-
-
-
pipelinerl.domains.math.load_datasets
-
-
Crashed
-
apiche
4m 46s
-
-
-
-
1
1
0
-
64
64
pipelinerl.domains.tir_mcp.generate_mcp_rollout
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
tapeagents.agent.Agent
-
-
3
-
mcp_agent
["{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'act', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Then produce single function call for the next step. If the answer is ready, call MathAnswer. Put your final answer within \\\\boxed{}.', 'steps': ['pipelinerl.domains.tir_mcp.steps.MathAnswer'], 'use_known_actions': True, 'use_function_calls': True}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.\\nBe specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.\\nConsider that next steps may depend on results of previous steps, so include conditional branching using \"if\" statements where needed.\\nStart with the title \"Plan\". Every step should have short name and description.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'reflect', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': \"1. Evaluate the action's success, explain its effect on current step, overall plan and task solution.\\n2. If the last action was not successful, describe errors and the possible reasons for failure.\\n3. Check if the current plan step is finished. \\n4. If the step is finished, update the following steps of the plan with new information and choose the next step.\\n${agent.templates.thought_format}\\n\", 'next_node': 'select'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'select', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Select the next step to do to move forward with the plan. Describe the expected effect of the proposed action.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'summarize', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Summarize last observation. If its an image, thoroughly describe it with all details.\\nDescribe the results of the last action and observed changes\\nDo not hallucinate or make up any information, only describe what you see in the observation.\\nDo not guess or assume action effects, describe only visible changes.\\n${agent.templates.thought_format}\\n'}"]
true
-
You have access to the following tools:
{tools_description}
You have access to the following tools:
{tools_description}
Output only a single JSON dict. Do not repeat the last thought again. If the last action does not change the observation, do not repeat it! DO NOT OUTPUT ANYTHING BESIDES THE JSON! DO NOT PLACE ANY COMMENTS INSIDE THE JSON. It will break the system that processes the output.
-
You are an expert AI Agent trained to assist users with complex information processing tasks.
Your role is to understand user queries and respond in a helpful and accurate manner.
Keep your replies concise and direct. Prioritize clarity and avoid over-elaboration.
Do not express emotions or opinions about user questions.
Important! Respond with the plain text, do not include any JSON or code.
Do not output anything besides what I asked in this message.
3
-
1
-
-
-
-
-
pipelinerl.domains.math.load_datasets
-
-
Crashed
-
apiche
3h 1m 16s
-
-
-
-
1
1
0
-
64
64
pipelinerl.domains.tir_mcp.generate_math_rollout2
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
tapeagents.agent.Agent
-
-
3
-
mcp_agent
["{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'act', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Then produce single function call for the next step. If the answer is ready, call MathAnswer. Put your final answer within \\\\boxed{}.', 'steps': ['pipelinerl.domains.tir_mcp.steps.MathAnswer'], 'use_known_actions': True, 'use_function_calls': True}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.\\nBe specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.\\nConsider that next steps may depend on results of previous steps, so include conditional branching using \"if\" statements where needed.\\nStart with the title \"Plan\". Every step should have short name and description.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'reflect', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': \"1. Evaluate the action's success, explain its effect on current step, overall plan and task solution.\\n2. If the last action was not successful, describe errors and the possible reasons for failure.\\n3. Check if the current plan step is finished. \\n4. If the step is finished, update the following steps of the plan with new information and choose the next step.\\n${agent.templates.thought_format}\\n\", 'next_node': 'select'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'select', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Select the next step to do to move forward with the plan. Describe the expected effect of the proposed action.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'summarize', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Summarize last observation. If its an image, thoroughly describe it with all details.\\nDescribe the results of the last action and observed changes\\nDo not hallucinate or make up any information, only describe what you see in the observation.\\nDo not guess or assume action effects, describe only visible changes.\\n${agent.templates.thought_format}\\n'}"]
true
-
You have access to the following tools:
{tools_description}
You have access to the following tools:
{tools_description}
Output only a single JSON dict. Do not repeat the last thought again. If the last action does not change the observation, do not repeat it! DO NOT OUTPUT ANYTHING BESIDES THE JSON! DO NOT PLACE ANY COMMENTS INSIDE THE JSON. It will break the system that processes the output.
-
You are an expert AI Agent trained to assist users with complex information processing tasks.
Your role is to understand user queries and respond in a helpful and accurate manner.
Keep your replies concise and direct. Prioritize clarity and avoid over-elaboration.
Do not express emotions or opinions about user questions.
Important! Respond with the plain text, do not include any JSON or code.
Do not output anything besides what I asked in this message.
3
-
1
-
-
-
-
-
pipelinerl.domains.math.load_datasets
-
-
Crashed
-
apiche
5d 21h 2m 3s
-
False
no
-
1
43
0
-
64
64
pipelinerl.domains.tir_mcp.generate_math_rollout2
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
tapeagents.agent.Agent
8192
1
4.66667
-
mcp_agent
["{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'act', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Then produce single function call for the next step. If the answer is ready, call GaiaAnswer. Put your final answer within \\\\boxed{}.', 'steps': ['examples.gaia_agent.steps.GaiaAnswer'], 'use_known_actions': True, 'use_function_calls': True}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Use python to compute the correct answer\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.\\nBe specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.\\nConsider that next steps may depend on results of previous steps, so include conditional branching using \"if\" statements where needed.\\nStart with the title \"Plan\". Every step should have short name and description.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'reflect', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': \"1. Evaluate the action's success, explain its effect on current step, overall plan and task solution.\\n2. If the last action was not successful, describe errors and the possible reasons for failure.\\n3. Check if the current plan step is finished. \\n4. If the step is finished, update the following steps of the plan with new information and choose the next step.\\n${agent.templates.thought_format}\\n\", 'next_node': 'select'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'select', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Select the next step to do to move forward with the plan. Describe the expected effect of the proposed action.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'summarize', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 100, 'guidance': 'Summarize last observation. If its an image, thoroughly describe it with all details.\\nDescribe the results of the last action and observed changes\\nDo not hallucinate or make up any information, only describe what you see in the observation.\\nDo not guess or assume action effects, describe only visible changes.\\n${agent.templates.thought_format}\\n'}"]
true
-
You have access to the following tools:
{tools_description}
You have access to the following tools:
{tools_description}
Output only a single JSON dict. Do not repeat the last thought again. If the last action does not change the observation, do not repeat it! DO NOT OUTPUT ANYTHING BESIDES THE JSON! DO NOT PLACE ANY COMMENTS INSIDE THE JSON. It will break the system that processes the output.
["{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'act', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 1, 'guidance': 'Then produce single function call for the next step. If the answer is ready, call GaiaAnswer.', 'steps': ['examples.gaia_agent.steps.GaiaAnswer'], 'use_known_actions': True, 'use_function_calls': True}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'plan', 'system_prompt': '${agent.templates.system_prompt}', 'guidance': 'Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.\\nBe specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.\\nConsider that next steps may depend on results of previous steps, so include conditional branching using \"if\" statements where needed.\\nStart with the title \"Plan\". Every step should have short name and description.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'reflect', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 1, 'guidance': \"1. Evaluate the action's success, explain its effect on current step, overall plan and task solution.\\n2. If the last action was not successful, describe errors and the possible reasons for failure.\\n3. Check if the current plan step is finished. \\n4. If the step is finished, update the following steps of the plan with new information and choose the next step.\\n${agent.templates.thought_format}\\n\", 'next_node': 'select'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'select', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 1, 'guidance': 'Select the next step to do to move forward with the plan. Describe the expected effect of the proposed action.\\n${agent.templates.thought_format}\\n', 'steps_prompt': '${agent.templates.allowed_tools}'}","{'_target_': 'tapeagents.nodes.StandardNode', 'name': 'summarize', 'system_prompt': '${agent.templates.system_prompt}', 'trim_obs_except_last_n': 1, 'guidance': 'Summarize last observation. If its an image, thoroughly describe it with all details.\\nDescribe the results of the last action and observed changes\\nDo not hallucinate or make up any information, only describe what you see in the observation.\\nDo not guess or assume action effects, describe only visible changes.\\n${agent.templates.thought_format}\\n'}"]
["You are an expert AI Agent trained to assist users with complex information processing tasks.\nYour role is to understand user queries and respond in a helpful and accurate manner.\nKeep your replies concise and direct. Prioritize clarity and avoid over-elaboration.\nDo not express emotions or opinions about user questions. \n","You are an expert AI Agent trained to assist users with complex information processing tasks.\nYour role is to understand user queries and respond in a helpful and accurate manner.\nKeep your replies concise and direct. Prioritize clarity and avoid over-elaboration.\nDo not express emotions or opinions about user questions. You must use the python tool for computation.\n"]
Important! Respond with the plain text, do not include any JSON or code.
Do not output anything besides what I asked in this message.
2.33333
-
1
-
-
nccl
-
-
pipelinerl.domains.math.load_datasets
-
-
Crashed
-
apiche
2h 6m 26s
-
False
no
-
1
64
10
-
64
64
pipelinerl.swe.rollouts.generate_unified_swe_rollout
1
-
50000000
-
-
{task}
-
50
-
-
-
-
15000
-
-
-
-
-
-
-
-
-
-
-
-
1
-
-
nccl
-
-
pipelinerl.swe.load_datasets.load_local_swe_dataset
/mnt/llmd/data/swegym/ds
-
Crashed
-
apiche
11m 10s
-
-
-
-
1
1
10
-
64
64
pipelinerl.swe.rollouts.generate_unified_swe_rollout
1
-
50000000
-
-
{task}
-
50
-
-
-
-
15000
-
-
-
-
-
-
-
-
-
-
-
-
1
-
-
-
-
-
pipelinerl.swe.load_datasets.load_local_swe_dataset
/mnt/llmd/data/swegym/ds
-
Crashed
-
apiche
18h 47m 31s
-
-
-
-
1
1
0
8
64
64
pipelinerl.domains.tir.rollouts.generate_tir_rollout
1
-
10000000
-
-
-
50
pipelinerl.domains.tir.agent.TIRMathAgent
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
8
-
-
-
-
-
pipelinerl.domains.tir.datasets.load_datasets
-
-
Crashed
-
apiche
4d 4h 55m 35s
-
-
-
-
1
64
0
-
64
64
pipelinerl.domains.math.generate_math_rollout
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
1
-
-
-
-
-
pipelinerl.domains.math.load_datasets
-
-
Crashed
-
apiche
20h 34m 28s
-
-
-
-
1
64
0
-
64
64
pipelinerl.domains.math.generate_math_rollout
1
-
10000000
-
Please reason step by step, and put your final answer within \boxed{}.
{task}
-
50
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
1
-
-
-
-
-
pipelinerl.domains.math.load_datasets
-
-
1-10
of 122
Add a comment
Created with ❤️ on Weights & Biases.
https://wandb.ai/apiche/pipeline-rl/reports/test-data-loader-refactoring--VmlldzoxMzQ2MzE1MA