Karandeep singh comments

Results 35 comments of


                                            Karandeep singh

Getting TTS output voice close to the training data - Finetuning on different language

Any update on this?

how to change pitch and how to list all the available speaker ids

List speaker ids - `tts --model_path "{modelDir}{checkpointName}" --config_path "{modelDir}config.json" --list_language_idxs --list_speaker_idxs`

When is the Setu-transliterate pipeline scheduled for release?

Hi @StephennFernandes, I realized you are quite actively exploring a lot of indic models across modalities. Would love to connect with you on LinkedIn: https://linkedin.com/in/kdcyberdude

Dataset used to train HAT_GAN_Real_SRx4?

@chxy95 When can we expect the training settings to be released? Any rough estimate?

attention_mask

@netagl, Is your `audio_encoder_per_device_batch_size` 1?

OOM on training Mistral hypernet

Hi @bminixhofer, Do I need to update `max_position_embedding` while initializing `roberta-base` model to 128 in zett/model/__init__.py I tried without using pretrained hypernet model as well. It's still giving OOM. And...

Got an abnormally high loss when training Gemma-7B.

Despite adding **``** token and loading model and tokenizer using **unsloth**, I am still getting very high loss. Single sequence example - ```python model, tokenizer = FastLanguageModel.from_pretrained( model_name = "unsloth/gemma-7b-it-bnb-4bit",...

Got an abnormally high loss when training Gemma-7B.

Hi @danielhanchen, Isn't it better to use **[group_by_length](https://huggingface.co/docs/transformers/en/main_classes/trainer#transformers.Seq2SeqTrainingArguments.group_by_length)** to group short sequences if **packing** is disabled?

Got an abnormally high loss when training Gemma-7B.

Got it @danielhanchen, it does reduce the loss, but it's still `15.05`. way more than what it supposed to be.