Skip to content

Datum

Source code in docs_src/capo_sagemakertrainingsessionruntime/types/datum.py 
14
15
16
17
18
19
20
class Datum(TypedDict, closed=True):
    model_input: "capo_sagemakertrainingsessionruntime.types.model_input.ModelInput"
    """<p>The tokenized input to feed to the model.</p>"""
    loss_fn_inputs: NotRequired[
        "capo_sagemakertrainingsessionruntime.types.tensor_data_map.TensorDataMap"
    ]
    """<p>The tensor inputs for the loss function, keyed by name. Unsupported keys are rejected. Supported keys: <code>target_tokens</code>, <code>weights</code>, <code>advantages</code>, <code>logprobs</code>, <code>ref_log_prob</code>, <code>temperature</code>. Required keys depend on the loss function type: SFT (<code>cross_entropy</code>) requires <code>target_tokens</code> and <code>weights</code>; RL (<code>ppo</code>, <code>cispo</code>, <code>importance_sampling</code>) requires <code>target_tokens</code>, <code>weights</code>, <code>advantages</code>, and <code>logprobs</code>.</p>"""

loss_fn_inputs instance-attribute 

loss_fn_inputs: NotRequired[TensorDataMap]

The tensor inputs for the loss function, keyed by name. Unsupported keys are rejected. Supported keys: target_tokens, weights, advantages, logprobs, ref_log_prob, temperature. Required keys depend on the loss function type: SFT (cross_entropy) requires target_tokens and weights; RL (ppo, cispo, importance_sampling) requires target_tokens, weights, advantages, and logprobs.

model_input instance-attribute 

model_input: ModelInput

The tokenized input to feed to the model.