coinbasecryptogames| The price list of volcanic bean bun large model was announced: it supports the "prepayment/postpayment" dual model, claiming to be "the highest concurrency standard in China"
News from IT House on May 21CoinbasecryptogamesByte Jump launched the bean bag model (formerly known as "skylark") in the Spring Volcano engine FORCE Motivity Conference on May 15. The model is mainly oriented to industry scenarios and can provide a variety of image and text audio generation capabilities. The IT house is organized as follows:
Bean bag general model pro: byte beat self-developed LLM model professional version, supporting 128k long text, the whole series can be fine-tuned, with stronger comprehensive ability of understanding, generation and logic, adapting to rich scenarios such as question and answer, summary, creation, classification and so on.
Bean bag general model lite: byte beat self-developed LLM model lightweight version, which provides lower token cost and lower latency than the professional version, and provides flexible and economical model choices for enterprises.
Doubao role-playing model: personalized role-creation ability, stronger context awareness and plot promotion ability to meet flexible role-playing needs
Doubao speech synthesis model: provides natural and vivid speech synthesis ability, is good at expressing a variety of emotions, and deduces a variety of scenes
Bean bag sound reproduction model: sound 1 can be realized in 5 seconds.Coinbasecryptogames1 cloning, highly restore timbre similarity and naturalness, and support cross-language transfer of sound
Doubao speech recognition model: higher accuracy and sensitivity, lower speech recognition delay, support for correct multilingual recognition
Bean bag Wen Sheng picture model: more accurate text understanding ability, more accurate picture and text matching, more beautiful picture effect, good at the creation of Chinese cultural elements
Bean bag Function call model: provides more accurate function identification and parameter extraction capabilities, and is suitable for scenarios where complex tools are invoked.
Bean packet vectorization model: focus on the use of vector retrieval scenarios, provide core understanding capabilities for LLM knowledge base, and support multiple languages.
Today, the official website of Volcano engine updated the pricing details of the bean bag model, saying that "on the basis that the reasoning price of the model is much lower than the industry price, the TPM and RPM of the bean bag general model have reached the highest domestic standard" and "the price is 99% lower than that of the industry.Coinbasecryptogames.7x to 8x; in addition, related models can also use "prepaid" and "post-paid" models:
Take the bean bag general model pro-32k as an example: according to the unit price of the "prepaid" model, the monthly price of 10K TPM is 2000 yuan. 10K*60*24*30=43200K .
That is, the price of 432000K Tokens is 2000 yuan, and the average price is 0.0046 yuan / 1000 Tokens. According to the "post-payment" model, the inference input usually accounts for the vast majority of the computational cost of model reasoning, and the industry generally thinks that the reasoning input is five times the output.
According to the bean bag general model pro-32k reasoning input 0.0008 yuan / 1000 Tokens, reasoning output 0.002 yuan / 1000 Tokens calculation, the comprehensive price of model reasoning is 0.001 yuan / 1000 Tokens.
Officials say that in the country,CoinbasecryptogamesThe TPM limit of his model is mostly between 100K and 300K, while the RPM is between 60 and 120. the RPM limit of the lightweight model is relatively high, but only between 300K and 500k. According to the calculation of 10K RPM quota, enterprise customers can call the general bean bag model 167times per second on average, thus meeting the needs of large model applications in most business scenarios in production systems.
Officials also stressed that the relevant standards have reached the upper limit of the RPM provided by OpenAI for high-level customers (Tier4 and Tier5 customers). On the long text model with greater computational challenges, the 128k version of the bean bag general model pro and lite, with current limits of 1K RPM and 400K TPM, is also much higher than other 128k long text models in China, which can help enterprises to use large models at lower cost and accelerate the landing of large model applications.
Advertisement statement: the external jump links (including not limited to hyperlinks, QR codes, passwords, etc.) contained in the text are used to transmit more information and save selection time. The results are for reference only. All articles of IT House contain this statement.
Tags:
Prev: coinmasterfreespinapp2020| Analysis of Enterprise Growth and Investment Value in Stock Market
Next: JungleJackpotJubilation| What may be the cause of the abnormal sound in the BYD f3 rear suspension?