October 17th, according to "Volcano EngineTODAY, THE VOLCANIC ENGINE WAS LAUNCHED AND UPGRADED IN FOUR NEW UNITS AT THE FORCE LINK AI INNOVATION TOUR AT WUHAN STATIONBean bag large model:Bean curdLarge model 1.6 Upgrade, original support 4 thought lengthsbean bag model 1.6 lite, bean bag speech synthesis model 2.0, bean bag sound recapturing model 2.0.
1AI SUMMARIZES THE MAIN INFORMATION ON THE MODELS AS FOLLOWS:
1.6 Upgrade
Providing four thinking lengths for Minimal, Low, Medium, High, balancing the different needs of enterprises for model effects, time lags and costs in different contexts, and further enhancing the efficiency of thinking. This was also the first home-grown model to support “slotting-regulating thinking”。
Compared to the single thinking mode before the upgrade of the model, the upgraded bean bag 1.6 The total output of the model Tokens decreased by 77.51 TP3T, the reflection time decreased by 84.61 TP3T, and the model's effects remained unchanged。
Volcanic engines officially launched to further meet the diversification needs of enterprisesbean bag model 1.6 space(Doubao-Seed-1.6-lite), which is lighter, faster in reasoning and more price-for-money than the flagship bean bag 1.6. The effect, the modelbeyond the bean pack model 1.5 pro(Doubao-Seed-1.5-pro), 1.5 pro up 14% in the enterprise-level scene assessment, in the maximum-used 0-32k input slot, the combined-use cost is 1.5 pro down to 53.3%.
Bean bag speech synthesis, echo sound 2.0 model
Bean bag speech synthesis model 2.0 (Doubao-Seed-TTS 2.0), Bean bag sound recapturing model 2.0 (Doubao-Seed-ICL 2.0), availableStronger emotional performance, more precise command compliance, and accurate reading of complex formulas.
The model is based on the soybean language model, which develops a new structure for the synthesis of speech, allows both synthetic and regraving voices to be understood in depth semantic terms, and expands context-based reasoning to evolve from a simple text reading to “precision emotional expression after understanding”. Users can do it in their own languageSpeed, emotion, voice, tone, style changeThe precision of this is adjusted to significantly increase the controlability of the voice。
The two new voice models have been optimized and are now being read in complex formulas from primary to high schoolACCURACY OF 90%.