Manasseh @devmanasseh

Introducing Context .NET 🚀 a small .NET CLI tool I built that gives code agents better context . it clones the source snapshot for the exact nuget package version you're using, locally under .context/ link below 👇

1 0 5 2K 1

View Details

Manasseh @devmanasseh

2 months ago

@jullerino thanks!

0 0 0 29 0

View Details

Manasseh @devmanasseh

2 months ago

hey @jullerino would be nice to have a copy button to get the whole agents response as markdown. been loving t3code so far amazing tool!

2 0 4 400 0

View Details

Manasseh @devmanasseh

2 months ago

@_chenglou wow

0 0 1 17 0

View Details

Manasseh @devmanasseh

3 months ago

There is widespread speaker ID contamination across the dataset, different speakers and genders collapsed under the same IDs throughout, this was inherited from the original google release. I'm running a full programmatic relabelling and will publish this aswell when ready.

0 0 1 26 0

View Details

Manasseh @devmanasseh

3 months ago

huggingface.co/datasets/manas…

1 0 1 22 0

View Details

Manasseh @devmanasseh

3 months ago

I've been curating a Shona 🇿🇼 speech dataset adapted from Google's recent WaxalNLP corpus. 86+ hours of cleaned speech with: >VAD-based silence trimming >Inter-utterance gap normalisation >SNR-based quality scoring >Normalised transcriptions for Shona orthography link below.

1 0 3 50 0

View Details

Manasseh @devmanasseh

3 months ago

@alfxad hey, thanks for the dm, trying to reply but the chat is readonly, i think it's a privacy settings because I'm not on premium

0 0 0 3 0

View Details

Manasseh @devmanasseh

3 months ago

@alfxad hey, i just saw your work on huggingface with the swahili csm finetuned model, i was wondering if i can pick your brain a little i'm doing a similiar project but for shona, i'td be a huge help

0 0 1 16 0

View Details

Manasseh @devmanasseh

3 months ago

@bernhardsson any chance i can get academic credits as a student outside the US? I'm burning through my remaining balance, still tuning the learning rate and fixing some overfitting issues. I'd love to get this over the line and open-source the Shona weights!

Manasseh @devmanasseh

3 months ago

I've been using @modal to finetune the sesame csm 1b speech model to support Shona for my final year capstone project, and its really good. I can spin up an a100 from my terminal and watch the training progress without leaving my IDE. i love the code first approach.