Historical Qwen3.5-derived vision-language research weights for chart and market-context experiments. The published configuration uses Qwen3_5ForConditionalGeneration (qwen3_5). This family uses third-party foundation weights and is distinct from Qovaryx's random-initialized scratch bases.
Status — October 3, 2026: Qovaryx is on the back burner. Desktop downloads are paused; there is no current public installer, purchase, or trial. These repositories remain as research references. Published weights are separate from desktop product availability.
The repository contains a full sharded Safetensors checkpoint, configuration, tokenizer, and processor assets. The upstream family is Qwen3.5-9B. The earlier card used the noncanonical identifier Qwen3.5-9B-VL; the base-model reference has been corrected.
Franken-B is a historical merge in the VFAi-X lineage. The repository preserves its manifest and May 22 audit reports. Their names include historical words such as "production" and "ship"; those are records of earlier decisions, not current release status.
Historical matched-output comparisons and regression batteries evaluated particular prompts and internal criteria. They do not establish investment returns, future predictive accuracy, or live trading performance. This cleanup preserves those source artifacts and removes the old app-download promotions and broad performance framing from the card.
Use a Transformers release that supports the architecture in this checkpoint's configuration, together with the supplied processor and tokenizer. Follow the upstream Qwen documentation for compatible multimodal inputs, then verify compatibility with this particular checkpoint and revision. Loading or inference was not rerun during this documentation update.
The earlier vLLM instructions described a legacy serving path; they are not a description of Qovaryx's current integration. GPU-memory and quantization requirements depend on the runtime and workload and are not certified by this card.
Model output can be incorrect or unsupported by the input. Labels such as BUY, SELL, HOLD, or NO_TRADE are experimental output conventions, not authorization to trade or a validated investment recommendation.
Apache-2.0, as recorded in the existing model metadata and repository license. Retain applicable Qwen and other third-party notices. Attribution to the foundation model is required to describe this lineage accurately.
J.E. Herizon LLC, doing business as JE Horizon.
model-00002-of-00006.safetensors 4.0 GBmodel-00004-of-00006.safetensors 4.0 GBmodel-00003-of-00006.safetensors 4.0 GBmodel-00005-of-00006.safetensors 4.0 GBmodel-00001-of-00006.safetensors 2.0 GBmodel-00006-of-00006.safetensors 800 MBQwen/Qwen3.5-9B