Downloads · 30 days
11
8% of all-time downloads
LFM2.5-1.2B-JAPT-GGUF
Jabs2/LFM2.5-1.2B-JAPT-GGUF
LFM2.5-1.2B-JAPT-GGUF is a machine learning model from Jabs2. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
GGUF quants of a Japanese→Portuguese fine-tune of LiquidAI/LFM2.5-1.2B-Instruct, built for offline anime subtitle translation (GoAnime TV project).
All-time downloads
142
Public
Repo size
567 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.gguf567 MB · 100%
At a glance
- Access
- Public
- Created
- Sep 18, 2026
- Updated
- Sep 18, 2026
- SHA
- 7b076782
Model card
From the Hugging Face model README
LFM2.5-1.2B-JAPT-GGUF
GGUF quants of a Japanese→Portuguese fine-tune of LiquidAI/LFM2.5-1.2B-Instruct,
built for offline anime subtitle translation (GoAnime TV project).
Method
- Base:
LiquidAI/LFM2.5-1.2B-Instruct(license lfm1.0, see LICENSE file of the base model) - QLoRA SFT (r=32, 6000 steps): 147k JA-PT pairs (Tatoeba + OpenSubtitles v2024, cleaned)
- 15k EN-PT pairs (Tatoeba, keeps the EN→PT direction alive)
- Merge → GGUF (F16) → quantize with
llama-quantize - IQ3_M uses an importance matrix calibrated on JA-PT translation data
Quality gate (6 sentences, greedy, temp 0)
Same protocol as the project gate: PASS = meaning preserved + zero hallucination.
| Quant | Size | Score |
|---|---|---|
LFM2.5-1.2B-JAPT-Q4_K_M.gguf | 731 MB | 5/6 |
LFM2.5-1.2B-JAPT-IQ3_M.gguf | 567 MB | 5/6 |
Known gap: polite request with honorific (先輩…いただけますか) still fails in every small model tested — only the 1.8B teacher (Hy-MT2) gets it right.
Usage (llama.cpp)
llama-cli -m LFM2.5-1.2B-JAPT-IQ3_M.gguf \
-sys "You are a helpful translation assistant." \
-p "Translate the following text into Portuguese. Output only the translated result without any additional explanation:
おはよう、今日はいい天気だね." --temp 0 -n 128
- Created
- Sep 18, 2026
- Updated
- Sep 18, 2026