Datasetpublic

LiquidAI/antidoom-mix-v1.0-LFM2.5-1.2B-Base

Antidoom Mix v1.0 – FTPO Pairs for LFM2.5-1.2B-Base [!Note] 📓 Tutorial: This is the dataset used in Antidoom Training with TRL Pre-extracted FTPO (Final Token Preference Optimization) training pairs for eliminating doom loops in LiquidAI/LFM2.5-1.2B-Base. Derived from LiquidAI/antidoom-mix-v1.0 by generating completions with LFM2.5-1.2B-Base. Schema Each row contains: Field Type Description full_prompt string The formatted prompt sent to the model… See the full description on the dataset page: https://huggingface.co/datasets/LiquidAI/antidoom-mix-v1.0-LFM2.5-1.2B-Base.

sourceHugging Faceupdated 20d agoView on Hugging Face

2likes94downloads

Dataset Card

Antidoom Mix v1.0 – FTPO Pairs for LFM2.5-1.2B-Base

[!Note] 📓 Tutorial: This is the dataset used in Antidoom Training with TRL

Pre-extracted FTPO (Final Token Preference Optimization) training pairs for eliminating doom loops in `LiquidAI/LFM2.5-1.2B-Base`.

Derived from `LiquidAI/antidoom-mix-v1.0` by generating completions with LFM2.5-1.2B-Base.

Schema

Each row contains:

Field	Type	Description
`full_prompt`	`string`	The formatted prompt sent to the model
`prompt_ids`	`list[int]`	Token IDs up to (but not including) the rejected token
`chosen_ids`	`list[int]`	~20 alternative token IDs ranked by logit score
`rejected_token_id`	`int`	The token that starts the doom loop

Generation Setup

Completions were generated using llama.cpp and the default configurations from the Antidoom repo:

bash

llama-server -m LFM2.5-1.2B-Base-BF16.gguf --port 8080 -ngl 99 -np 4 --temp 0.01 --top-k 50 --top-p 1.0 --min-p 0.01

Parameter	Value
`max_tokens`	1000
`temperature`	0.01
`top_k`	50
`top_p`	1.0
`min_p`	0.01

Citation

@article{liquidAI2026Antidoom,
    author = {Liquid AI},
    title = {Reducing Doom Loops with Final Token Preference Optimization},
    journal = {Liquid AI Blog},
    year = {2026},
    note = {www.liquid.ai/blog/antidoom}
}