KasuleTrevor

cdli-qwen3-asr-lg-typical-1p7b-base-finetune

Deploy Dedicated

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Model Details

Model Provider

KasuleTrevor

Model Tree

Base

Qwen/Qwen3-ASR-1.7B

Fine-tuned

this model

Input Modalities

Audio

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Training Setup

Base model: Qwen/Qwen3-ASR-1.7B
Datasets: KasuleTrevor/lg_100hrs + dmusingu/yogera-dataset (Luganda split)
Training language tag: Luganda
Forced inference language: disabled
Epochs: 10
Batch size: 4
Gradient accumulation: 2
Learning rate: 2e-05
Scheduler: cosine
Warmup ratio: 0.03
Save steps: 500
Selected checkpoint: checkpoint-61500
Selection reason: fallback best eval_loss (0.181851)

Final Test Metrics

Corpus WER (normalized): 0.226814
Corpus CER (normalized): 0.066090
Average utterance WER (normalized): 0.230013
Average utterance CER (normalized): 0.067483

Checkpoint Selection Evidence

Table with columns: checkpoint, step, eval_loss
checkpoint	step	eval_loss
checkpoint-61500	61500	0.1818510740995407
checkpoint-62000	62000	0.181895226240158
checkpoint-62500	62500	0.18197962641716
checkpoint-63000	63000	0.1819778531789779

Source Dataset Breakdown

Table with columns: source_dataset, n_samples, mean_wer, mean_cer, median_wer, median_cer
source_dataset	n_samples	mean_wer	mean_cer	median_wer	median_cer
lg100	2828	0.23	0.0675	0.1667	0.0274

Artifacts

Result folder: results/checkpoint-61500/
Includes checkpoint validation summaries, final test predictions, scored outputs, and grouped analyses.