jonah1024/Sped_ParakeetRomanian_110M_TDT-CTC_gguf overview
Acesta este un model convertit în format gguf al modelului fine tuned : gabrielpirlo/Sped ParakeetRomanian 110M TDT CTC , care este un model dedicat exclusiv l…
Runs locally from ~85.8 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| Sped_ParakeetRomanian_110M_TDT-CTC-F16.gguf | GGUF | F16 | 218.7 MB | Download |
| Sped_ParakeetRomanian_110M_TDT-CTC-F32.gguf | GGUF | F32 | 435.4 MB | Download |
| Sped_ParakeetRomanian_110M_TDT-CTC-Q4_K_M.gguf | GGUF | Q4_K_M | 85.8 MB | Download |
| Sped_ParakeetRomanian_110M_TDT-CTC-Q8_0.gguf | GGUF | Q8_0 | 129.1 MB | Download |
Model Details
| Model ID | jonah1024/Sped_ParakeetRomanian_110M_TDT-CTC_gguf |
|---|---|
| Author | jonah1024 |
| Pipeline | automatic-speech-recognition |
| License | apache-2.0 |
| Base model | gabrielpirlo/Sped_ParakeetRomanian_110M_TDT-CTC |
| Last modified | 2026-07-26T11:11:13.000Z |
Model README
---
license: apache-2.0
language:
- ro
base_model:
- gabrielpirlo/Sped_ParakeetRomanian_110M_TDT-CTC
pipeline_tag: automatic-speech-recognition
tags:
- gguf
---
Acesta este un model convertit în format gguf al modelului fine-tuned: gabrielpirlo/Sped_ParakeetRomanian_110M_TDT-CTC, care este un model dedicat exclusiv limbii române. Modelul a fost adaptat din nvidia/parakeet-tdt_ctc-110m prin transfer learning și antrenat pe 2.636 de ore de înregistrări vocale în limba română (sursa).
Conținut:
| File |Quantization| Size |
| ------------------------------------------------ |------------| -----: |
| Sped_ParakeetRomanian_110M_TDT-CTC-F32.gguf |F32 | 457 MB |
| Sped_ParakeetRomanian_110M_TDT-CTC-F16.gguf |F16 | 229 MB |
| Sped_ParakeetRomanian_110M_TDT-CTC-Q8_0.gguf |Q8_0 | 135 MB |
| Sped_ParakeetRomanian_110M_TDT-CTC-Q4_K_M.gguf |Q4_K_M | 90 MB |
| convert-parakeet.py |- | - |
Scop:
Model personalizat pentru handy-computer
Pentru a-l utiliza cu handy, copiați fișierul .gguf în directorul:
<App Data Directory>/models/
> Notă: App Data Directory poate fi găsit în secțiunea About a aplicației handy.
Cum s-a convertit:
Am actualizat scriptul convert-parakeet.py cu profilul:
# 110M Romanian hybrid TDT+CTC (SpeD).
"Sped_ParakeetRomanian_110M_TDT-CTC": {
"variant": "tdt_ctc-110m-ro",
"display_name": "SpeD Parakeet Romanian 110M TDT-CTC",
"version": "v1",
"size_label": "110M",
"head_kind": "tdt",
"expected_vocab_size": 1024,
"languages": ["ro"],
"lang_detect": False,
"prefer_direct_load": True,
"license": "cc-by-4.0",
"license_name": "Creative Commons Attribution 4.0",
"license_link": "https://creativecommons.org/licenses/by/4.0/",
}
Conversie:
cd transcribe.cpp \
uv run --project scripts/envs/parakeet \
python3 convert-parakeet.py gabrielpirlo/Sped_ParakeetRomanian_110M_TDT-CTC
Quantization:
~/transcribe.cpp/build/bin/transcribe-quantize \
models/Sped_ParakeetRomanian_110M_TDT-CTC/Sped_ParakeetRomanian_110M_TDT-CTC-F32.gguf \
models/Sped_ParakeetRomanian_110M_TDT-CTC/Sped_ParakeetRomanian_110M_TDT-CTC-F16.gguf --quant F16
Credits:
Lucrare: arXiv:2511.03361
Run jonah1024/Sped_ParakeetRomanian_110M_TDT-CTC_gguf with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models