laion/Sera-4.6-Lite-T2-v4-1000

laion/Sera-4.6-Lite-T2-v4-1000

Row-subset of allenai/Sera-4.6-Lite-T2 (the dataset upstream SERA-8B was trained on), with OpenAI tool_calls pre-rendered into the content string as Hermes/Qwen3-style <tool_call>...</tool_call> wire tokens and tool responses wrapped as <tool_response>...</tool_response>.

This mirrors Ai2's sera/datagen/data/postprocess/utils.py::transform_traj_hermes (default tool_call_format: "hermes") which is the missing step between the public Sera-4.6-Lite-T2 dataset and axolotl training. Without this pre-render, axolotl's chat_template: chatml discards the structured tool_calls field and the SFT model never learns to emit tool calls.

Size: 1,000 rows (source: 36,083 rows).

Format: Raw JSONL. Per row, messages: list[{role, content, train}]. Roles are system | user | assistant. Tool observations are represented as role: user with <tool_response>...</tool_response> wrapping (per SERA convention). train: bool on each message is the per-message loss mask consumed by axolotl's message_field_training: train.

Sampling: deterministic random, seed=42, row-indexed into the full 36,083-row source. Row subsets are nested.

Usage (axolotl)

datasets:
  - path: laion/Sera-4.6-Lite-T2-v4-1000
    data_files:
      - sera-4.6-lite-t2_v4_1000.jsonl
    type: chat_template
    field_messages: messages
    ds_type: json
    message_field_training: train
chat_template: chatml