Streaming breaks with finetuned model

[mbittristan]

Self Checks

• This template is only for bug reports. For questions, please visit Discussions.
• I have thoroughly reviewed the project documentation (installation, training, inference) but couldn't find information to solve my problem. English 中文 日本語 Portuguese (Brazil)
• I have searched for existing issues, including closed ones. Search issues
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:）
• Please do not modify this template and fill in all required fields.

Cloud or Self Hosted

Self Hosted (Source)

Environment Details

OS: Ubuntu 24.04
Python: 3.10.16

FFMPEG: 4.4.2

Steps to Reproduce

1. Run server

python -m tools.api_server --listen 0.0.0.0:50005 --llama-checkpoint-path "checkpoints/fish-speech-1.5" --decoder-checkpoint-path "checkpoints/fish-speech-1.5/firefly-gan-vq-fsq-8x1024-21hz-generator.pth" --decoder-config-name firefly_gan_vq --compile

2. Make a streaming api request using postman or node

{ "reference_id": "tom",
"normalize": 1,
"max_new_tokens": 1024,
"chunk_length": 200,
"top_p": 0.8,
"repetition_penalty": 1.5,
"temperature": 0.7,
"streaming": 1,
"use_memory_cache": "off",
"rate": 8000,
"channels": 1,
"seed": 6,
"text": "Hi there, my name is Tom. This is what I sound like." }

✔️ Expected Behavior

A wav stream that can be played.

❌ Actual Behavior

A non-working wav stream is returned with a malformed header.
Postman is unable to play the stream, and ffmpeg is unable to convert the stream to mulaw at 8000hz.

However, non-streamed output still generates fine.

[mbittristan]

It should also be noted that this is using 1.5.0

[Stardust-minus]

We didn't support the 8khz output, you should use ffmpeg and convert the output stream to 44100Hz files.