Riva 2.9.0 NMT does not produce translated output

Please provide the following information when requesting support.
Hardware - Nvidia A100
Hardware - AMD EPYC 7413
Operating System - Ubuntu 20.04.5
Riva Version - 2.9.0
TLT Version - N/a
How to reproduce the issue ? (This is for errors. Please share the command and the detailed log here)

python function, that does not produce error or output:

def send_nmt_request(self, text, target_lang):
	nmt_service = riva.client.NeuralMachineTranslationClient(self.auth)

	if target_lang == 'de-DE':
		model = 'en_de_24x6'
	elif target_lang == 'es-US':
		model = 'en_es_24x6'
	elif target_lang == 'zh-CN':
		model = 'en_zh_24x6'
	elif target_lang == 'ru-RU':
		model = 'en_ru_24x6'
	elif target_lang == 'fr-FR':
		model = 'en_fr_24x6'
	else:
		print(f'Model for language code ({target_lang}) not available')
		return None

	resp = nmt_service.translate(text, model=model, source_language="en-US", target_language=target_lang)
	print(f'Translation Response: {resp}')
	return resp

Attached are docker logs for riva. At the end of the file you can see that I tried twice with en → ru, and once with en → es. Both produced this error in docker logs and returned nothing to my python script.

docker_riva_logs.txt (52.5 KB)

Hi @armstrongsam25

Thanks for your interest in Riva

We will try to try this from our end, can you kindly share your config.sh used

Thanks

config.sh (12.8 KB)

@rvinobha sorry, didn’t reply to your comment or @ you. My bad.

Hi @armstrongsam25

Apologies for the delay

Found the Cause if the problem
the text (even if a single sentence must be provided as a list)

I also faced the same issue when I passes the text as string, it returned empty response and attribute error in riva-speech logs

Passing text as list will resolve the issue and provide transcript

resp = nmt_service.translate(text, model=model, source_language="en-US", target_language=target_lang)

i.e first param text needs to be of type list

Please try and let me know if it works

Thanks for your patience