MagicSource

python API support

4

Hi, any plan on support python API?

Streaming decode issue

3

Hello, for llama when decoding Chinese or Japanese characters, since one character mgith need 2 or more tokens to decode, so when streaming, the chunk returned one token decode result...

``` Sure, why not? What would you like to know? ``` why it can not stop at stop tokenizer, it will print out this token

Hello, I get a weired issue when serving using ctransformers. code: ```python model = AutoModelForCausalLM.from_pretrained( args.base_model, **config ) iterator: Generator = model.generate(gen_kwargs["inputs"]) for chat_chunk in iterator: new_text = model.detokenize(chat_chunk) print(new_text,...

I using chat prompt send to llm(query), but the generate result can not stop. Am using codellama, does there any chat example to reference?

How's the DownSampleBlock performance compare with CAbstractor?

3

How's the DownSampleBlock performance compare with CAbstractor?

Audio has info redundancy compare with picture.

No root transition

3

MHFormer seems didnt provide root transition, BTW, will it have a metahuman version?

onnxruntime option support?

2

eigen虽好但不好维护（应该也不是最快的方案），有没有兴趣将onnxruntime作为候选项支持？