Inkling Reasoning Parser (reasoning/inkling_parser.py)

Parses Inkling‑style reasoning streams, switching channels via opener tokens to separate thinking and visible content for the VMLX engine. Use when handling Inkling‑formatted model outputs.

# Source: https://github.com/jjang-ai/vmlx | License: Apache-2.0 | Copyright You must retain, in the Source form of any Derivative Works
# Provided via justclaw.it.com; keep this notice with the code.
# SPDX-License-Identifier: Apache-2.0
"""Inkling reasoning parser — CHANNEL-SWITCHED, not tag-delimited.

Every other reasoning family in this engine wraps thought in a matched pair
(``<think>…</think>``, ``<mm:think>…</mm:think>``). Inkling does not, and
building it on the think-tag base would be wrong in a way that only shows up on
a real bundle. Its contract, read from
``Inkling-Small-JANG-AWQ-90GB/jang_config.json`` (``chat.reasoning``):

    thinking_channel_token      <|content_thinking|>
    visible_channel_token       <|content_text|>
    message_end_token           <|end_message|>
  

... (truncated -- full source via MCP)