Evasion-detection task
According to the model card, the model classifies earnings-call Q&A pairs as direct, intermediate, or fully_evasive following the Rasiah framework.
Open Source Model Profile · FutureMa
Eva-4B
Eva-4B is a 4.02B-parameter Qwen3 model from FutureMa for detecting evasive answers in earnings-call Q&A. Its model card documents EvasionBench training and a Qwen3-4B-Instruct base.
- Publisher
- FutureMa
- Task
- text-generation
- Model type
- qwen3
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Eva-4B is published by FutureMa as a text-generation model for financial evasion detection. The captured configuration identifies Qwen3ForCausalLM with model type qwen3, and Safetensors metadata reports 4,022,468,096 parameters. According to the model card, it is a full-parameter fine-tune of Qwen3-4B-Instruct-2507 for 3-way classification of earnings-call Q&A, with Apache-2.0 recorded as the license.
Recorded capabilities
Documented Qwen3 base and full fine-tune
Hub tags and the model card identify Qwen/Qwen3-4B-Instruct-2507 as the base, with full-parameter fine-tuning reported.
EvasionBench data and reported scores
The model card describes 30,000 balanced training samples with multi-model annotation, and reports 81.3% accuracy on the human test set.
Training setup and stated limits
According to the model card, training used 2 epochs, 2048 sequence length, and 2x B200 GPUs, with English-only and domain-specific limits noted.
Use cases in the source record
- Research and tooling around corporate disclosure quality by classifying earnings-call Q&A as direct, intermediate, or fully_evasive.
- Prompted JSON classification experiments using the model card's documented template and Transformers generation example.
Limitations and unknowns
- According to the model card, evaluation is English-only, domain-specific to earnings-call Q&A, and the intermediate class has lower agreement.
- The model card notes judge position-bias risk, possible self-preference, higher annotation cost, and temporal drift from 2005-2023 data.
- According to the model card, Eva-4B is a research artifact and not financial advice, with human review recommended for high-stakes use.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: FutureMa/Eva-4B
Captured: Unknown. Processed: 2026-09-07T19:35:39.574890+00:00.
Eva-4B: Financial Evasion Detection Model Eva-4B is a 4B-parameter model for detecting evasive answers in earnings call Q&A . Model Summary Model name: Eva-4B Task: 3-way classification of Q&A pairs into: direct intermediate fully_evasive Base model: Qwen/Qwen3-4B-Instruct-2507 Training method: full-parameter fine-tuning Training data: EvasionBench training set (30,000 samples; 10,000 per class) Intended Use Eva-4B is intended for research and tooling around corporate disclosure quality and evasiveness in earnings call Q&A. Task Definition Given an earnings call Question (analyst) and Answer (management), the model predicts one of:…
F001F002F003F004F005F006F007F009F010F011F012F013F015F016F017F018F019F020F021