Downloads · 30 days
0
KasuleTrevor/cdli-slam-asr-luganda-atypical-llm-lora-e2s107
cdli-slam-asr-luganda-atypical-llm-lora-e2s107 is a automatic speech recognition model from KasuleTrevor. Use it when you need speech turned into text. It is set up for slam-llm. The card lists the license as other.
LLM-LoRA plus projector atypical-speech adaptation checkpoint for SLAM-ASR on the CDLI Luganda atypical speech dataset. The Whisper encoder remains frozen; the linear projector and decoder LoRA adapters are updated fr…
Downloads · 30 days
0
Access
Public
Updated Aug 21, 2026
Repo size
115 MB
Likes
0
Public
Click a slice to open those files.
.pt115 MB · 100%
From the Hugging Face model README
LLM-LoRA plus projector atypical-speech adaptation checkpoint for SLAM-ASR on the CDLI Luganda atypical speech dataset. The Whisper encoder remains frozen; the linear projector and decoder LoRA adapters are updated from the ASR-adapted starting checkpoint.
This Hub repository stores a partial SLAM-ASR checkpoint for use with the
SLAM-LLM codebase. It is not a standalone transformers checkpoint.
llm_lora_projectorSunbird/asr-whisper-large-v3-saltSunbird/Sunflower-14Bmodel.ptcdli/ugandan_luganda_nonstandard_speech_v1.0testTest decode used MAX_NEW_TOKENS=200, NUM_BEAMS=4, REPETITION_PENALTY=2.0, NO_REPEAT_NGRAM_SIZE=2, USE_LLM_PEFT=true, LLM_TARGET_MODULES=[q_proj,v_proj].
Notebook-style subgroup breakdown on the test split: Mild 48.94% WER, Moderate 52.78%, Severe 62.60%. By disorder: Dysarthria 50.15%, Articulation Disorders 53.70%, Stuttering 54.28%, Voice disorder 67.68%. This checkpoint shows stable decoding with hyp/ref ratio 92.88%.
Load through SLAM-LLM; this repository stores a partial SLAM-ASR checkpoint, not a standalone Transformers model.
Typical decode flow in this project uses:
examples/asr_luganda/scripts/decode_luganda_sunflower.shUSE_ENCODER_PEFT=true for encoder-LoRA checkpointsThis addendum keeps the archived test predictions fixed and swaps in
the cleaned Luganda references from experiments/luganda/test_cleaned.tsv
where a match exists. It is a reference-quality sensitivity analysis,
not a new decode run.
2026-08-211027/1028 test items matchedUG022_LUG048keep-original for the single unmatched item54.42%19.10%47.46%16.32%-6.96 points-2.78 points52.67%20.50%51.18%20.24%The metric change reflects cleaner references, not a model update. Use the archived metrics for strict apples-to-apples comparison with the original SLAM-ASR benchmark, and use this section to quantify the impact of transcript cleanup on the same predictions.
<!-- cleaned-reference-side-analysis:end -->